chore: ✨ stash audio experiments

closes #120

.dockerignore

@@ -0,0 +1,34 @@
+# Include any files or directories that you don't want to be copied to your
+# container here (e.g., local build artifacts, temporary files, etc.).
+#
+# For more help, visit the .dockerignore file reference guide at
+# https://docs.docker.com/engine/reference/builder/#dockerignore-file
+
+**/.DS_Store
+**/__pycache__
+**/.venv
+**/.classpath
+**/.dockerignore
+**/.env
+**/.git
+**/.gitignore
+**/.project
+**/.settings
+**/.toolstarget
+**/.vs
+**/.vscode
+**/*.*proj.user
+**/*.dbmdl
+**/*.jfm
+**/bin
+**/charts
+**/docker-compose*
+**/compose*
+**/Dockerfile*
+**/node_modules
+**/npm-debug.log
+**/obj
+**/secrets.dev.yaml
+**/values.dev.yaml
+LICENSE
+README.md

.github/CODEOWNERS

@@ -0,0 +1,5 @@
+* @melMass
+extern/GFPGAN/* @TencentARC
+extern/SadTalker/* @OpenTalker
+nodes/deep_bump.py @HugoTini
+web/imageFeed.js @pythongosssss @melMass

.github/FUNDING.yml

@@ -0,0 +1,14 @@
+# These are supported funding model platforms
+
+github: [melMass]
+custom: ["https://www.buymeacoffee.com/melmass"]
+patreon: # Replace with a single Patreon username
+open_collective: # Replace with a single Open Collective username
+ko_fi: # Replace with a single Ko-fi username
+tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
+community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
+liberapay: # Replace with a single Liberapay username
+issuehunt: # Replace with a single IssueHunt username
+otechie: # Replace with a single Otechie username
+lfx_crowdfunding: # Replace with a single LFX Crowdfunding project-name e.g., cloud-foundry
+

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -0,0 +1,83 @@
+name: 🐞 Bug Report
+title: "[bug] "
+description: Report a bug
+labels: ["type: 🐛 bug", "status: 🧹 needs triage"]
+assignees:
+  - melMass
+  
+body:
+  - type: markdown
+    attributes:
+      value: |
+        ## Before submiting an issue
+        - Make sure to read the README & INSTALL instructions.  
+        - Please search for [existing issues](https://github.com/melMass/comfy_mtb/issues?q=is%3Aissue) around your problem before filing a report.
+
+        ### Try using the debug mode to get more info
+
+        If you use the env variable `MTB_DEBUG=true`, debug message from the extension will appear in the terminal.
+
+  - type: textarea
+    id: description
+    attributes:
+      label: Describe the bug
+      description: A clear description of what the bug is. Include screenshots if applicable.
+      placeholder: Bug description
+    validations:
+      required: true
+
+  - type: textarea
+    id: reproduction
+    attributes:
+      label: Reproduction
+      description: Steps to reproduce the behavior.
+      placeholder: |
+        1. Add node xxx ...
+        2. Connect to xxx ...
+        3. See error
+
+  - type: textarea
+    id: expected-behavior
+    attributes:
+      label: Expected behavior
+      description: A clear description of what you expected to happen.
+
+  - type: dropdown
+    id: os
+    attributes:
+      label: Operating System
+      description: What OS are you using?
+      options:
+        - Windows (Default)
+        - Linux
+        - Mac
+      default: 0
+    validations:
+      required: true
+      
+  - type: dropdown
+    id: comfy_mode
+    attributes:
+      label: Comfy Mode
+      description: What flavor of Comfy do you use?
+      options:
+        - Comfy Portable (embed) (Default)
+        - In a custom virtual env (venv, virtualenv, conda...)
+        - Google Colab
+        - Other (online services, containers etc..)
+      default: 0
+    validations:
+      required: true
+
+  - type: textarea
+    id: logs
+    attributes:
+      label: Console output
+      description: Paste the console output without backticks
+      render: sh
+
+  - type: textarea
+    id: context
+    attributes:
+      label: Additional context
+      description: Add any other context about the problem here.

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1 @@
+blank_issues_enabled: false

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -0,0 +1,35 @@
+name: 💡 Feature Request
+title: "[feat] "
+description: Suggest an idea
+labels: ["type: 🤚 feature request"]
+
+body:
+  - type: textarea
+    id: problem
+    attributes:
+      label: Describe the problem
+      description: A clear description of the problem this feature would solve
+      placeholder: "I'm always frustrated when..."
+    validations:
+      required: true
+
+  - type: textarea
+    id: solution
+    attributes:
+      label: "Describe the solution you'd like"
+      description: A clear description of what change you would like
+      placeholder: "I would like to..."
+    validations:
+      required: true
+
+  - type: textarea
+    id: alternatives
+    attributes:
+      label: Alternatives considered
+      description: "Any alternative solutions you've considered"
+
+  - type: textarea
+    id: context
+    attributes:
+      label: Additional context
+      description: Add any other context about the problem here.

.github/workflows/package_wheels.yml

@@ -27,17 +27,15 @@ jobs:
     steps:
       - name: ♻️ Checking out the repository
         uses: actions/checkout@v3
-      - name: "🐍 Setting up Python"
+      - name: '🐍 Setting up Python'
         uses: actions/setup-python@v4
         with:
-          python-version: "3.10.9"
+          python-version: '3.10.9'
 
       - name: 📦 Building and Bundling wheels
         shell: bash
         run: |
-          python -m pip wheel --no-cache-dir -r requirements-wheels.txt -w ./wheels > build.log
-
-          cat build.log
+          python -m pip wheel --no-cache-dir -r reqs.txt -w ./wheels 2>&1 | tee build.log
 
           # find source wheels
           packages=$(cat build.log | awk -F 'Building wheels for collected packages: ' '{print $2}')
@@ -45,6 +43,13 @@ jobs:
 
           IFS=', ' read -r -a package_array <<< "$packages"
 
+          # Save reversed package_array to wheel_order.txt
+          reversed_array=()
+          for ((idx=${#package_array[@]}-1; idx>=0; idx--)); do
+            reversed_array+=("${package_array[idx]}")
+          done
+          printf '%s\n' "${reversed_array[@]}" > ./wheels/wheel_order.txt
+
           printf "Autodetect this source package: \e[32m%s\e[0m\n" "${package_array[@]}"
 
           # Iterate through the wheel files and remove those that are not source built
@@ -71,4 +76,4 @@ jobs:
         uses: actions/cache/save@v3
         with:
           path: ${{ env.archive_name }}.zip
-          key: ${{ env.archive_name }}
+          key: ${{ env.archive_name }}-${{ hashFiles('reqs.txt') }}

.github/workflows/release.yml

@@ -6,7 +6,7 @@ on:
       name:
         description: Release tag / name ?
         required: true
-        default: "latest"
+        default: 'latest'
         type: string
       environment:
         description: Environment to run tests against
@@ -27,8 +27,36 @@ jobs:
       - name: ♻️ Checking out the repository
         uses: actions/checkout@v3
         with:
-          submodules: "recursive"
+          submodules: 'recursive'
           path: ${{ env.repo_name }}
+
+      # - name: 📝 Prepare file with paths to remove
+      #   run: |
+      #     find ${{ env.repo_name }} -type f -size +10M > .release_ignore
+      #     find ${{ env.repo_name }} -type d -empty >> .release_ignore
+      #   shell: bash
+
+      - name: 🗑️ Remove files and directories listed in .release_ignore
+        shell: bash
+        run: |
+          release_ignore="${{ env.repo_name }}/.release_ignore"
+          if [ -f "$release_ignore" ]; then
+            while IFS= read -r entry || [ -n "$entry" ]; do
+              target="${{ env.repo_name }}/$entry"
+              if [ -e "$target" ]; then
+                if [ -f "$target" ]; then
+                  rm "$target"
+                elif [ -d "$target" ]; then
+                  rm -r "$target"
+                fi
+              else
+                echo "Warning: $entry does not exist in the repository. Skipping removal."
+              fi
+            done < "$release_ignore"
+          else
+            echo "No .release_ignore file found. Skipping removal of files and directories."
+          fi
+
       - name: 📦 Building custom comfy nodes
         shell: bash
         run: |
@@ -70,10 +98,18 @@ jobs:
         id: cache
         with:
           path: ${{ env.archive_name }}.zip
-          key: ${{ env.archive_name }}
+          key: ${{ env.archive_name }}-${{ hashFiles('reqs.txt') }}
+      - name: 📦 Unzip wheels
+        shell: bash
+        run: |
+          mkdir -p wheels
+          unzip -j ${{ env.archive_name }}.zip "**/*.whl" -d wheels
+          unzip -j ${{ env.archive_name }}.zip "**/*.txt" -d wheels
+        if: success()
       - name: ✅ Add wheels to release
         uses: softprops/action-gh-release@v1
         with:
           tag_name: ${{ inputs.name }}
           files: |
-            ${{ env.archive_name }}.zip
+            wheels/*.whl
+            wheels/wheel_order.txt

.github/workflows/test_embedded.yml

@@ -0,0 +1,71 @@
+name: 🧪 Test Comfy Portable
+
+on: workflow_dispatch
+jobs:
+  install-comfy:
+    runs-on: windows-latest
+    env:
+      repo_name: ${{ github.event.repository.name }}
+    steps:
+      - name: ⚡️ Restore Cache if Available
+        id: cache-comfy
+        uses: actions/cache/restore@v3
+        with:
+          path: ComfyUI_windows_portable
+          key: ${{ runner.os }}-comfy-env
+
+      - name: 🚡 Download and Extract Comfy
+        id: download-extract-comfy
+        if: steps.cache-comfy.outputs.cache-hit != 'true'
+        shell: bash
+        run: |
+          mkdir comfy_temp
+          curl -L -o comfy_temp/comfyui.7z https://github.com/comfyanonymous/ComfyUI/releases/download/latest/ComfyUI_windows_portable_nvidia_cu118_or_cpu.7z
+
+          7z x comfy_temp/comfyui.7z -o./comfy_temp
+
+
+          # mv comfy_temp/ComfyUI_windows_portable/python_embeded .
+          # mv comfy_temp/ComfyUI_windows_portable/ComfyUI .
+          # mv comfy_temp/ComfyUI_windows_portable/update .
+          ls
+          mv comfy_temp/ComfyUI_windows_portable .
+
+      - name: 💾 Store cache
+        uses: actions/cache/save@v3
+        if: steps.cache-comfy.outputs.cache-hit != 'true'
+        with:
+          path: ComfyUI_windows_portable
+          key: ${{ runner.os }}-comfy-env
+      - name: ⏬ Install other extensions
+        shell: bash
+        run: |
+          export COMFY_PYTHON="${GITHUB_WORKSPACE}/ComfyUI_windows_portable/python_embeded/python.exe"
+          cd "${GITHUB_WORKSPACE}/ComfyUI_windows_portable/ComfyUI/custom_nodes"
+
+          git clone https://github.com/Fannovel16/comfy_controlnet_preprocessors
+          cd comfy_controlnet_preprocessors
+          $COMFY_PYTHON -m pip install -r requirements.txt
+          
+      - name: ♻️ Checking out comfy_mtb to custom_nodes
+        uses: actions/checkout@v3
+        with:
+          submodules: 'recursive'
+          path: ComfyUI_windows_portable/ComfyUI/custom_nodes/${{ env.repo_name }}
+
+      - name: 📦 Install mtb nodes
+        shell: bash
+        run: |
+          # run install
+          export COMFY_PYTHON="${GITHUB_WORKSPACE}/ComfyUI_windows_portable/python_embeded/python.exe"
+          cd "${GITHUB_WORKSPACE}/ComfyUI_windows_portable/ComfyUI/custom_nodes"
+          $COMFY_PYTHON ${{ env.repo_name }}/install.py -w
+
+      - name: ⏬ Import mtb_nodes
+        shell: bash
+        run: |
+          export COMFY_PYTHON="${GITHUB_WORKSPACE}/ComfyUI_windows_portable/python_embeded/python.exe"
+          cd "${GITHUB_WORKSPACE}/ComfyUI_windows_portable/ComfyUI"
+          $COMFY_PYTHON -s main.py --quick-test-for-ci --cpu
+
+          $COMFY_PYTHON -m pip freeze

.gitignore

@@ -1,3 +1,5 @@
 __pycache__
 *.py[cod]
-*.onnx
+*.onnx
+wheels/
+node_modules/

.gitmodules

@@ -1,3 +1,9 @@
-[submodule "extern/SadTalker"]
-	path = extern/SadTalker
-	url = https://github.com/OpenTalker/SadTalker.git
+[submodule "extern/google-FILM"]
+	path = extern/frame_interpolation
+	url = https://github.com/google-research/frame-interpolation
+[submodule "extern/GFPGAN"]
+	path = extern/GFPGAN
+	url = https://github.com/TencentARC/GFPGAN.git
+[submodule "extern/frame_interpolation"]
+	path = extern/frame_interpolation
+	url = https://github.com/google-research/frame-interpolation

.prettierrc

@@ -0,0 +1,6 @@
+{
+  "semi": false,
+  "singleQuote": true,
+  "tabWidth": 2,
+  "useTabs": false
+}

.release_ignore

@@ -0,0 +1,4 @@
+extern/frame_interpolation/moment.gif
+extern/frame_interpolation/photos
+extern/GFPGAN/inputs
+.git

INSTALL-CN.md

@@ -0,0 +1,93 @@
+# 安装
+- [安装](#安装)
+    - [自动安装（推荐）](#自动安装推荐)
+    - [ComfyUI 管理器](#comfyui-管理器)
+    - [虚拟环境](#虚拟环境)
+  - [模型下载](#模型下载)
+      - [网络扩展](#网络扩展)
+    - [旧的安装方法 (MANUAL)](#旧的安装方法-manual)
+    - [依赖关系](#依赖关系)
+### 自动安装（推荐）
+
+### ComfyUI 管理器
+
+从 0.1.0 版开始，该扩展将使用 [ComfyUI-Manager](https://github.com/ltdrdata/ComfyUI-Manager) 进行安装，这对处理各种环境下的各种安装问题大有帮助。
+
+### 虚拟环境
+还有一种试验性的单行安装方法，即在 ComfyUI 根目录下使用以下命令进行安装。它将下载代码、安装依赖项并运行安装脚本：
+
+
+```bash
+curl -sSL "https://raw.githubusercontent.com/username/repo/main/install.py" | python3 -
+```
+
+## 模型下载
+某些节点需要下载额外的模型，您可以使用与上述相同的 python 环境以交互方式完成下载：
+
+```bash
+python scripts/download_models.py
+```
+
+然后根据提示或直接按回车键下载每个模型。
+
+> **Note**
+> 您可以使用以下方法下载所有型号，无需提示：
+  ```bash
+  python scripts/download_models.py -y
+  ```
+
+#### 网络扩展
+
+首次运行时，脚本会尝试将 [网络扩展](https://github.com/melMass/comfy_mtb/tree/main/web)链接到你的 "web/extensions "文件夹，[请参阅](https://github.com/melMass/comfy_mtb/blob/d982b69a58c05ccead9c49370764beaa4549992a/__init__.py#L45-L61)。
+
+<img alt="color widget preview" src="https://github.com/melMass/comfy_mtb/assets/7041726/cff7e66a-4cc4-4866-b35b-10af0bb2d110" width=450>
+
+### 旧的安装方法 (MANUAL)
+### 依赖关系
+<details><summary><h4>Custom Virtualenv（我主要用这个）</h4></summary
+    
+1. 确保您处于用于 ComfyUI 的 Python 环境中。
+2. 运行以下命令安装所需的依赖项：
+  ```bash
+  pip install -r comfy_mtb/reqs.txt
+  ```
+
+</details>
+
+<details><summary><h4>Comfy 便携式/单机版（来自 ComfyUI 版本）</h4></summary>
+
+如果您使用 ComfyUI 单机版中的 `python-embeded `，那么当二进制文件没有轮子时，您就无法使用 pip 安装二进制文件的依赖项，在这种情况下，请查看最近的 [发布](https://github.com/melMass/comfy_mtb/releases)，那里有一个预编译轮子的 linux 和 windows 捆绑包（只有那些需要从源代码编译的轮子），请查看 [此问题 (#1)](https://github.com/melMass/comfy_mtb/issues/1) 以获取更多信息。
+![image](https://github.com/melMass/comfy_mtb/assets/7041726/2934fa14-3725-427c-8b9e-2b4f60ba1b7b)
+
+
+</details>
+
+<details><summary><h4>Google Colab</h4></summary>
+
+在 **Run ComfyUI with localtunnel (Recommended Way)** 标题之后（代码单元格之前）添加一个新的代码单元格
+
+![preview of where to add it on colab](https://github.com/melMass/comfy_mtb/assets/7041726/35df2ef1-14f9-44cd-aa65-353829188cd7)
+
+
+```python
+# download the nodes
+!git clone --recursive https://github.com/melMass/comfy_mtb.git custom_nodes/comfy_mtb
+
+# download all models
+!python custom_nodes/comfy_mtb/scripts/download_models.py -y
+
+# install the dependencies
+!pip install -r custom_nodes/comfy_mtb/reqs.txt -f https://download.openmmlab.com/mmcv/dist/cu118/torch2.0/index.html
+```
+
+如果运行后 colab 抱怨需要重新启动运行时，请重新启动，然后不要重新运行之前的单元格，只运行运行本地隧道的单元格。(可能需要先添加一个包含 `%cd ComfyUI` 的单元格）
+
+
+> **Note**:
+> If you don't need all models, remove the `-y` as collab actually supports user input: ![image](https://github.com/melMass/comfy_mtb/assets/7041726/40fc3602-f1d4-432a-98fd-ce2240f5ad06)
+
+> **Preview**
+> ![image](https://github.com/melMass/comfy_mtb/assets/7041726/b5b2b2d9-f1e8-4c43-b1db-7dfc5e07be86)
+
+</details>
+

INSTALL-JP.md

@@ -0,0 +1,93 @@
+# インストール
+
+- [インストール](#インストール)
+  - [自動インストール (推奨)](#自動インストール-推奨)
+    - [ComfyUI マネージャ](#comfyui-マネージャ)
+    - [仮想環境](#仮想環境)
+  - [モデルのダウンロード](#モデルのダウンロード)
+    - [ウェブ拡張機能](#ウェブ拡張機能)
+  - [旧インストール方法 (MANUAL)](#旧インストール方法-manual)
+    - [依存関係](#依存関係)
+
+
+## 自動インストール (推奨)
+
+### ComfyUI マネージャ
+
+バージョン0.1.0では、この拡張機能は[ComfyUI-Manager](https://github.com/ltdrdata/ComfyUI-Manager)と一緒にインストールすることを想定しています。これは、様々な環境で直面する様々なインストール問題を処理するのに非常に役立ちます。
+
+### 仮想環境
+また、ComfyUIのルートから以下のコマンドを使用する実験的なワンライナー・インストールもあります。これはコードをダウンロードし、依存関係をインストールし、インストールスクリプトを実行します：
+
+```bash
+curl -sSL "https://raw.githubusercontent.com/username/repo/main/install.py" | python3 -
+```
+
+## モデルのダウンロード
+ノードによっては、追加モデルのダウンロードが必要な場合があるので、上記と同じ python 環境を使って対話的に行うことができる：
+```bash
+python scripts/download_models.py
+```
+
+プロンプトに従うか、Enterを押すだけで全てのモデルをダウンロードできます。
+
+
+> **Note**
+> プロンプトを出さずに全てのモデルをダウンロードするには、以下のようにします:
+  ```bash
+  python scripts/download_models.py -y
+  ```
+
+### ウェブ拡張機能
+
+初回実行時にスクリプトは[web extensions](https://github.com/melMass/comfy_mtb/tree/main/web)をあなたの快適な `web/extensions` フォルダに[シンボリックリンク](https://github.com/melMass/comfy_mtb/blob/d982b69a58c05ccead9c49370764beaa4549992a/__init__.py#L45-L61)しようとします。万が一失敗した場合は、mtbフォルダを手動で`ComfyUI/web/extensions`にコピーしてください：
+
+<img alt="color widget preview" src="https://github.com/melMass/comfy_mtb/assets/7041726/cff7e66a-4cc4-4866-b35b-10af0bb2d110" width=450>
+
+## 旧インストール方法 (MANUAL)
+### 依存関係
+
+<details><summary><h4>カスタム Virtualenv (私は主にこれを使っています)</h4></summary>
+    
+1. ComfyUIで使用しているPython環境であることを確認してください。
+2. 以下のコマンドを実行して、必要な依存関係をインストールします：
+  ```bash
+  pip install -r comfy_mtb/reqs.txt
+  ```
+
+</details> 
+
+<details><summary><h4>Comfy-portable / standalone (ComfyUI リリースより)</h4></summary>。
+    
+もしあなたがComfyUIスタンドアロンから`python-embeded`を使用している場合、バイナリがホイールを持っていない場合、依存関係をpipでインストールすることができません。この場合、最後の[リリース](https://github.com/melMass/comfy_mtb/releases)をチェックしてください。(ソースからのビルドが必要なもののみ)あらかじめビルドされたホイールがあるlinuxとwindows用のバンドルがあります。詳細は[この問題(#1)](https://github.com/melMass/comfy_mtb/issues/1)をチェックしてください。
+
+![image](https://github.com/melMass/comfy_mtb/assets/7041726/2934fa14-3725-427c-8b9e-2b4f60ba1b7b)
+
+</details>
+
+<details><summary><h4>Google Colab</h4></summary>
+
+ComfyUI with localtunnel (Recommended Way)**ヘッダーのすぐ後（コードセルの前）に、新しいコードセルを追加してください。
+![colabに追加する場所のプレビュー](https://github.com/melMass/comfy_mtb/assets/7041726/35df2ef1-14f9-44cd-aa65-353829188cd7)
+
+```python
+# download the nodes
+!git clone --recursive https://github.com/melMass/comfy_mtb.git custom_nodes/comfy_mtb
+
+# download all models
+!python custom_nodes/comfy_mtb/scripts/download_models.py -y
+
+# install the dependencies
+!pip install -r custom_nodes/comfy_mtb/reqs.txt -f https://download.openmmlab.com/mmcv/dist/cu118/torch2.0/index.html
+```
+これを実行した後、colabがランタイムを再起動する必要があると文句を言ったら、それを実行し、それ以前のセルは再実行せず、localtunnelを実行するセルだけを再実行してください。(最初に`%cd ComfyUI`のセルを追加する必要があるかもしれません...)
+
+
+> **Note**:
+> すべてのモデルが必要でない場合は、`-y`を削除してください : ![image](https://github.com/melMass/comfy_mtb/assets/7041726/40fc3602-f1d4-432a-98fd-ce2240f5ad06)
+
+> **プレビュー**
+> ![image](https://github.com/melMass/comfy_mtb/assets/7041726/b5b2b2d9-f1e8-4c43-b1db-7dfc5e07be86)
+
+</details>
+

INSTALL.md

@@ -0,0 +1,85 @@
+# Installation
+- [Installation](#installation)
+  - [Automatic Install (Recommended)](#automatic-install-recommended)
+    - [ComfyUI Manager](#comfyui-manager)
+    - [Virtual Env](#virtual-env)
+  - [Models Download](#models-download)
+  - [Old installation method (MANUAL)](#old-installation-method-manual)
+    - [Dependencies](#dependencies)
+
+## Automatic Install (Recommended)
+
+### ComfyUI Manager
+
+As of version 0.1.0, this extension is meant to be installed with the [ComfyUI-Manager](https://github.com/ltdrdata/ComfyUI-Manager), which helps a lot with handling the various install issues faced by various environments.
+
+### Virtual Env
+There is also an experimental one liner install using the following command from ComfyUI's root. It will download the code, install the dependencies and run the install script:
+
+```bash
+curl -sSL "https://raw.githubusercontent.com/username/repo/main/install.py" | python3 -
+```
+
+## Models Download
+Some nodes require extra models to be downloaded, you can interactively do it using the same python environment as above:
+```bash
+python scripts/download_models.py
+```
+
+then follow the prompt or just press enter to download every models.
+
+> **Note**
+> You can use the following to download all models without prompt:
+  ```bash
+  python scripts/download_models.py -y
+  ```
+
+
+## Old installation method (MANUAL)
+### Dependencies
+<details><summary><h4>Custom Virtualenv (I use this mainly)</h4></summary>
+    
+1. Make sure you are in the Python environment you use for ComfyUI.
+2. Install the required dependencies by running the following command:
+  ```bash
+  pip install -r comfy_mtb/reqs.txt
+  ```
+
+</details>
+
+<details><summary><h4>Comfy-portable / standalone (from ComfyUI releases)</h4></summary>
+    
+If you use the `python-embeded` from ComfyUI standalone then you are not able to pip install dependencies with binaries when they don't have wheels, in this case check the last [release](https://github.com/melMass/comfy_mtb/releases) there is a bundle for linux and windows with prebuilt wheels (only the ones that require building from source), check [this issue (#1)](https://github.com/melMass/comfy_mtb/issues/1) for more info.
+![image](https://github.com/melMass/comfy_mtb/assets/7041726/2934fa14-3725-427c-8b9e-2b4f60ba1b7b)
+
+
+
+</details>
+
+<details><summary><h4>Google Colab</h4></summary>
+
+Add a new code cell just after the **Run ComfyUI with localtunnel (Recommended Way)** header (before the code cell)
+![preview of where to add it on colab](https://github.com/melMass/comfy_mtb/assets/7041726/35df2ef1-14f9-44cd-aa65-353829188cd7)
+
+
+```python
+# download the nodes
+!git clone --recursive https://github.com/melMass/comfy_mtb.git custom_nodes/comfy_mtb
+
+# download all models
+!python custom_nodes/comfy_mtb/scripts/download_models.py -y
+
+# install the dependencies
+!pip install -r custom_nodes/comfy_mtb/reqs.txt -f https://download.openmmlab.com/mmcv/dist/cu118/torch2.0/index.html
+```
+If after running this, colab complains about needing to restart runtime, do it, and then do not rerun earlier cells, just the one to run the localtunnel. (you might have to add a cell with `%cd ComfyUI` first...)
+
+
+> **Note**:
+> If you don't need all models, remove the `-y` as collab actually supports user input: ![image](https://github.com/melMass/comfy_mtb/assets/7041726/40fc3602-f1d4-432a-98fd-ce2240f5ad06)
+
+> **Preview**
+> ![image](https://github.com/melMass/comfy_mtb/assets/7041726/b5b2b2d9-f1e8-4c43-b1db-7dfc5e07be86)
+
+</details>
+

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2023 Mel Massadian
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README-CN.md

@@ -0,0 +1,99 @@
+# MTB Nodes
+
+<a href="https://www.buymeacoffee.com/melmass" target="_blank"><img src="https://www.buymeacoffee.com/assets/img/custom_images/orange_img.png" alt="Buy Me A Coffee" style="height: 32px !important;width: 140px !important;box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;-webkit-box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;" ></a>
+
+[** 安装指南**](./INSTALL-CN.md) | [** 示例**](https://github.com/melMass/comfy_mtb/wiki/Examples)
+
+欢迎使用 MTB Nodes 项目！这个代码库是开放的，您可以自由地探索和利用。它的主要目的是构建用于 [MLOPs](https://github.com/Bismuth-Consultancy-BV/MLOPs) 中的概念验证（POCs）。该项目中的许多节点都是受到现有社区贡献或内置功能的启发而创建的。
+
+在继续之前，请注意与此项目中使用的某些库相关的许可证。例如，`deepbump` 库采用 [GPLv3](https://github.com/HugoTini/DeepBump/blob/master/LICENSE) 许可证。
+
+- [节点列表](#节点列表)
+  - [bbox](#bbox)
+  - [colors](#colors)
+  - [人脸检测/交换](#人脸检测交换)
+  - [图像插值（动画）](#图像插值动画)
+  - [图像操作](#图像操作)
+  - [潜在变量工具](#潜在变量工具)
+  - [其他工具](#其他工具)
+  - [纹理](#纹理)
+- [Comfy 资源](#comfy-资源)
+
+
+
+
+# 节点列表
+
+## bbox
+- `Bounding Box`: BBox 构造函数（自定义类型）
+- `BBox From Mask`: 从遮罩中提取边界框
+- `Crop`: 根据边界框裁剪图像
+- `Uncrop`: 根据边界框还原图像
+
+## colors
+- `Colored Image`: 给定尺寸的纯色图像
+- `RGB to HSV`: -
+- `HSV to RGB`: -
+- `Color Correct`: 基本颜色校正工具  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/7c20ac83-31ff-40ea-a1a0-06c2acefb2ef" width=345/>
+
+## 人脸检测/交换
+- `Face Swap`: 使用 deepinsight/insightface 模型进行人脸交换（该节点在早期版本中称为 `Roop`，功能相同，`Roop` 只是使用这些模型的应用程序）
+  > **注意**
+  > 人脸索引允许您选择要替换的人脸，如下所示：
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/2e9d6066-c466-4a01-bd6c-315f7f1e8b42" width=320/>
+- `Load Face Swap Model`: 加载 insightface 模型用于人脸交换
+- `Restore Face`: 使用 [GFPGan](https://github.com/TencentARC/GFPGAN) 还原人脸，与 `Face Swap` 配合使用效果很好，并支持 `bg_upscaler` 的 Comfy 原生放大器
+  
+## 图像插值（动画）
+- `Load Film Model`: 加载 [FILM](https://github.com/google-research/frame-interpolation) 模型
+- `Film Interpolation`: 使用 [FILM](https://github.com/google-research/frame-interpolation) 处理输入帧  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/3afd1647-6634-4b92-a34b-51432e6a9834" width=400/>
+- `Export to Prores (experimental)`: 将输入帧导出为 ProRes 4444 mov 文件。这使用 ffmpeg stdin 发送原始的 NumPy 数组，与 `Film Interpolation` 一起使用，目前很简单，但可以进一步扩展。
+
+## 图像操作
+- `Blur`: 使用高斯滤波器对图像进行模糊处理。
+- `Deglaze Image`: 从 [FN16](https://github.com/Fannovel16/FN16-ComfyUI-nodes/blob/main/DeglazeImage.py) 中提取
+- `Denoise`: 对输入图像进行降噪处理
+- `Image Compare`: 比较两个图像并返回差异图像
+- `Image Premultiply`: 使用掩码对图像进行预乘处理
+- `Image Remove Background Rembg`: 使用 [RemBG](https://github.com/danielgatis/rembg) 进行背景去除  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/e69253b4-c03c-45e9-92b5-aa46fb887be8" width=320/>
+- `Image Resize Factor`: 大部分提取自 [WAS Node Suite](https://github.com/WASasquatch/was-node-suite-comfyui)，经过一些编辑（特别是支持多个图像）和较少的功能。
+- `Mask To Image`: 将遮罩（Alpha）转换为带有颜色和背景的 RGB 图像
+- `Save Image Grid`: 将输入批次中的所有图像保存为图像网格。
+
+## 潜在变量工具
+- `Latent Lerp`: 两个潜在变量之间的线性插值（混合）
+
+
+## 其他工具
+- `Concat Images`: 接受两个图像流，并将它们合并为其他 Comfy 管道支持的图像批次。
+- `Image Resize Factor`: **已弃用**，因为我后来发现了内
+
+置的图像调整大小功能。
+- `Text To Image`: 使用字体将文本转换为图像的工具
+- `Styles Loader`: 加载 csv 文件并从行中填充下拉列表（类似于 A111）  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/02fe3211-18ee-4e54-a029-931388f5fde8" width=320/>
+- `Smart Step`: 一个非常基本的节点，用于获取在 KSampler 高级中使用的步骤百分比
+- `Qr Code`: 基本的 QR Code 生成器
+- `Save Tensors`: 调试节点，将来可能会被删除
+- `Int to Number`: 用于 WASSuite 数字节点的补充
+- `Smart Step`: 使用百分比来控制 `KAdvancedSampler` 的步骤（开始/停止）
+
+## 纹理
+
+- `DeepBump`: 从单张图片生成法线图和高度图
+
+# Comfy 资源
+
+**指南**:
+- [官方示例（英文）](https://comfyanonymous.github.io/ComfyUI_examples/)
+- @BlenderNeko 的[ComfyUI 社区手册（英文）](https://blenderneko.github.io/ComfyUI-docs/)
+
+- @tjhayasaka 的[Tomoaki 个人 Wiki（日文）](https://comfyui.creamlab.net/guides/)
+
+**扩展和自定义节点**:
+- @WASasquatch 的[Comfy 列表插件（英文）](https://github.com/WASasquatch/comfyui-plugins)
+
+- [CivitAI 上的 ComfyUI 标签（英文）](https://civitai.com/tag/comfyui)

README-JP.md

@@ -0,0 +1,96 @@
+# MTB Nodes
+
+<a href="https://www.buymeacoffee.com/melmass" target="_blank"><img src="https://www.buymeacoffee.com/assets/img/custom_images/orange_img.png" alt="Buy Me A Coffee" style="height: 32px !important;width: 140px !important;box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;-webkit-box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;" ></a>
+
+[**インストールガイド**](./INSTALL-JP.md) | [**サンプル**](https://github.com/melMass/comfy_mtb/wiki/Examples)
+
+MTB Nodesプロジェクトへようこそ！このコードベースは、自由に探索し、利用することができます。主な目的は、[MLOPs](https://github.com/Bismuth-Consultancy-BV/MLOPs)の実装のための概念実証（POC）を構築することです。このプロジェクトの多くのノードは、既存のコミュニティの貢献や組み込みの機能に触発されています。
+
+続行する前に、このプロジェクトで使用されている特定のライブラリに関連するライセンスに注意してください。たとえば、「deepbump」ライブラリは、[GPLv3](https://github.com/HugoTini/DeepBump/blob/master/LICENSE)の下でライセンスされています。
+
+- [ノードリスト](#ノードリスト)
+  - [bbox](#bbox)
+  - [colors](#colors)
+  - [顔検出 / スワッピング](#顔検出--スワッピング)
+  - [画像補間（アニメーション）](#画像補間アニメーション)
+  - [画像操作](#画像操作)
+  - [潜在的なユーティリティ](#潜在的なユーティリティ)
+  - [その他のユーティリティ](#その他のユーティリティ)
+  - [テクスチャ](#テクスチャ)
+- [Comfyリソース](#comfyリソース)
+
+
+# ノードリスト
+
+## bbox
+- `Bounding Box`: BBoxコンストラクタ（カスタムタイプ）
+- `BBox From Mask`: マスクからバウンディングボックスを抽出
+- `Crop`: BBoxから画像を切り抜く
+- `Uncrop`: BBoxから画像を元に戻す
+
+## colors
+- `Colored Image`: 指定されたサイズの一定の色の画像
+- `RGB to HSV`: -
+- `HSV to RGB`: -
+- `Color Correct`: 基本的なカラーコレクションツール  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/7c20ac83-31ff-40ea-a1a0-06c2acefb2ef" width=345/>
+
+## 顔検出 / スワッピング
+- `Face Swap`: deepinsight/insightfaceモデルを使用した顔の入れ替え（このノードは初期バージョンでは「Roop」と呼ばれていましたが、同じ機能を提供します。Roopは単にこれらのモデルを使用するアプリです）
+  > **注意**
+  > 顔のインデックスを使用して置き換える顔を選択できます。以下を参照してください：
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/2e9d6066-c466-4a01-bd6c-315f7f1e8b42" width=320/>
+- `Load Face Swap Model`: 顔の交換のためのinsightfaceモデルを読み込む
+- `Restore Face`: [GFPGan](https://github.com/TencentARC/GFPGAN)を使用して顔を復元し、`Face Swap`と組み合わせて使用すると非常に効果的であり、`bg_upscaler`のComfyネイティブアップスケーラーもサポートしています。
+
+## 画像補間（アニメーション）
+- `Load Film Model`: [FILM](https://github.com/google-research/frame-interpolation)モデルを読み込む
+- `Film Interpolation`: [FILM](https://github.com/google-research/frame-interpolation)を使用して入力フレームを処理する  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/3afd1647-6634-4b92-a34b-51432e6a9834" width=400/>
+- `Export to Prores (experimental)`: 入力フレームをProRes 4444 movファイルにエクスポートします。これは現在は単純なものですが、`Film Interpolation`と組み合わせて使用するためのffmpegのstdinを使用して生のNumPy配列を送信するもので、拡張することもできます。
+
+## 画像操作
+- `Blur`: ガウスフィルタを使用して画像をぼかす
+- `Deglaze Image`: [FN16](https://github.com/Fannovel16/FN16-ComfyUI-nodes/blob/main/DeglazeImage.py)から取得
+- `Denoise`: 入力画像のノイズを除去する
+- `Image Compare`: 2つの画像を比較し、差分画像を返す
+- `Image Premultiply`: 画像をマスクで乗算
+- `Image Remove Background Rembg`: [RemBG](https://github.com/danielgatis/rembg)を使用した背景除去  
+  <img src="https://github.com/melMass/comfy_mtb/assets/704172
+
+6/e69253b4-c03c-45e9-92b5-aa46fb887be8" width=320/>
+- `Image Resize Factor`: [WAS Node Suite](https://github.com/WASasquatch/was-node-suite-comfyui)から抽出され、いくつかの編集（特に複数の画像のサポート）と機能の削減が行われました。
+- `Mask To Image`: マスク（アルファ）をカラーと背景を持つRGBイメージに変換します。
+- `Save Image Grid`: 入力バッチのすべての画像を画像グリッドとして保存します。
+
+## 潜在的なユーティリティ
+- `Latent Lerp`: 2つの潜在的なベクトルの間の線形補間（ブレンド）
+
+## その他のユーティリティ
+- `Concat Images`: 2つの画像ストリームを取り、他のComfyパイプラインでサポートされている画像のバッチとしてマージします。
+- `Image Resize Factor`: **非推奨**。組み込みの画像リサイズ機能を発見したため、削除される予定です。
+- `Text To Image`: フォントを使用してテキストを画像に変換するためのユーティリティ
+- `Styles Loader`: csvファイルをロードし、行からドロップダウンを作成します（A111のようなもの）  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/02fe3211-18ee-4e54-a029-931388f5fde8" width=320/>
+- `Smart Step`: KSamplerの高度な使用に使用するステップパーセントを取得する非常に基本的なノード
+- `Qr Code`: 基本的なQRコード生成器
+- `Save Tensors`: 将来的に削除される可能性のあるデバッグノード
+- `Int to Number`: WASSuiteの数値ノードの補完
+- `Smart Step`: `KAdvancedSampler`のステップ（開始/停止）を制御するための非常に基本的なツールで、パーセンテージを使用します。
+
+## テクスチャ
+
+- `DeepBump`: 1枚の画像から法線マップと高さマップを生成します。
+
+# Comfyリソース
+
+**ガイド**:
+- [公式の例（英語）](https://comfyanonymous.github.io/ComfyUI_examples/)
+- @BlenderNekoによる[ComfyUIコミュニティマニュアル（英語）](https://blenderneko.github.io/ComfyUI-docs/)
+
+- @tjhayasakaによる[Tomoakiの個人Wiki（日本語）](https://comfyui.creamlab.net/guides/)
+
+**拡張機能とカスタムノード**:
+- @WASasquatchによる[Comfyリスト用のプラグイン（英語）](https://github.com/WASasquatch/comfyui-plugins)
+
+- [CivitAIのComfyUIタグ（英語）](https://civitai.com/tag/comfyui)

README.md

@@ -1,51 +1,166 @@
-## MTB Nodes
+# MTB Nodes
+[![embedded test](https://github.com/melMass/comfy_mtb/actions/workflows/test_embedded.yml/badge.svg)](https://github.com/melMass/comfy_mtb/actions/workflows/test_embedded.yml)
 
-Feel free to do whatever you want with this codebase, I'm mainly using Comfy to build POCs to implement in [MLOPs](https://github.com/Bismuth-Consultancy-BV/MLOPs). And a lot of nodes are inspired by existing ones from the community or builtin 
-Just beware of the licenses of some libraries (deepbump for instance is [GPLv3](https://github.com/HugoTini/DeepBump/blob/master/LICENSE))
+![home](https://repository-images.githubusercontent.com/649047066/a3eef9a7-20dd-4ef9-b839-884502d4e873)
 
-## Install
+<!-- omit in toc -->
 
-From within the python environment you already use for ComfyUI install the requirements.
-```bash
-pip install -r comfy_mtb/requirements.txt
-```
+**Translated Readme (using DeepTranslate, PRs are welcome)**:  
+![image](https://github.com/melMass/comfy_mtb/assets/7041726/f8429c14-3521-4e28-82a3-863d781976c0)
+[日本語による説明](./README-JP.md)  
+![image](https://github.com/melMass/comfy_mtb/assets/7041726/d5cc1fdd-2820-4a5c-b2d7-482f1c222063)
+[中文说明](./README-CN.md)
 
-## Screenshots
+<a href="https://www.buymeacoffee.com/melmass" target="_blank"><img src="https://www.buymeacoffee.com/assets/img/custom_images/orange_img.png" alt="Buy Me A Coffee" style="height: 32px !important;width: 140px !important;box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;-webkit-box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;" ></a>
 
-- **FaceSwap [roop]** (using [roop](https://github.com/s0md3v/roop/))  
-  The face index allow you to choose which face to replace as you can see here:
-  ![ComfyUI_909](https://github.com/melMass/comfy_mtb/assets/7041726/2e9d6066-c466-4a01-bd6c-315f7f1e8b42)
+[**Install Guide**](./INSTALL.md) | [**Examples**](https://github.com/melMass/comfy_mtb/wiki/Examples)
 
-- **Style Loader**: A111 like csv styles in Comfy  
-  ![image](https://github.com/melMass/comfy_mtb/assets/7041726/02fe3211-18ee-4e54-a029-931388f5fde8)
+Welcome to the MTB Nodes project! This codebase is open for you to explore and utilize as you wish. Its primary purpose is to build proof-of-concepts (POCs) for implementation in [MLOPs](https://github.com/Bismuth-Consultancy-BV/MLOPs). Many nodes in this project are inspired by existing community contributions or built-in functionalities.
 
-- **Color Correction**: basic color correction node  
-  ![image](https://github.com/melMass/comfy_mtb/assets/7041726/7c20ac83-31ff-40ea-a1a0-06c2acefb2ef)
+Before proceeding, please be aware of the licenses associated with certain libraries used in this project. For example, the `deepbump` library is licensed under [GPLv3](https://github.com/HugoTini/DeepBump/blob/master/LICENSE).
 
-- **Image Remove Background [RemBG]**: (using [rembg](https://github.com/danielgatis/rembg))
-  ![image](https://github.com/melMass/comfy_mtb/assets/7041726/e69253b4-c03c-45e9-92b5-aa46fb887be8)
+- [Web Extensions](#web-extensions)
+- [Node List](#node-list)
+  - [Animation](#animation)
+  - [bbox](#bbox)
+  - [colors](#colors)
+  - [image ops](#image-ops)
+  - [latent utils](#latent-utils)
+  - [textures](#textures)
+  - [misc utils](#misc-utils)
+  - [Optional nodes](#optional-nodes)
+    - [face detection / swapping](#face-detection--swapping)
+    - [image interpolation (animation)](#image-interpolation-animation)
+- [Comfy Resources](#comfy-resources)
+
+# Web Extensions
+mtb add a few widgets like `COLOR`
+
+<img alt="color widget preview" src="https://github.com/melMass/comfy_mtb/assets/7041726/cff7e66a-4cc4-4866-b35b-10af0bb2d110" width=450>
+
+A few nodes have the concept of "dynamic" inputs:  
+<img alt="dynamic inputs" width=450  src="https://github.com/melMass/comfy_mtb/assets/7041726/10b3976e-b212-4968-91eb-f34c02bb80c3" />
 
 
+# Node List
 
-### Node List
+## Animation
+- `Animation Builder`: Convenient way to manage basic animation maths at the core of many of my workflows (both worflows for the following GIFs are in the [examples](https://github.com/melMass/comfy_mtb/wiki/Examples))  
 
-- `Latent Lerp`: Linear Interpolate between two latents,
-- `Int to Number`: Supplement for WASSuite number nodes,
+  **[Example lerping two conditions (blue car -> yellow car)](https://github.com/melMass/comfy_mtb/blob/main/examples/03-animation_builder-condition-lerp.json)**
+  
+  <img width=300 src="https://user-images.githubusercontent.com/7041726/260258970-d6d66d96-fb34-40d0-9038-cbabf0714c5d.gif"/>  
+  
+
+  **[Example using image transforms a feedback for a fake deforum effect](https://github.com/melMass/comfy_mtb/blob/main/examples/04-animation_builder-deforum.json)**
+  
+  <img width=300 src="https://user-images.githubusercontent.com/7041726/260261504-303a1037-60d3-4b31-a589-b15d549752f6.gif"/>
+  
+- `Batch Float`: Generates a batch of float values with interpolation.
+- `Batch Shape`: Generates a batch of 2D shapes with optional shading (experimental).
+- `Batch Transform`: Transform a batch of images using a batch of keyframes.  
+  <img width=400 src="https://github.com/melMass/comfy_mtb/assets/7041726/3f217de1-79aa-49b0-a66a-35cf29dd8f01"/>
+- `Export With Ffmpeg`: Export with FFmpeg, it used to be export to Proress and is still tailored for YUV
+- `Fit Number` : Fit the input float using a source and target range, you can also control the interpolation curve from a list of presets (default to linear)
+  
+## bbox
 - `Bounding Box`: BBox constructor (custom type),
-- `Crop`: Crop image from BBox,
-- `Uncrop`: Uncrop image from BBox,
-- `ImageBlur`: Blur the input image,
-- `Denoise`: Denoise the input image,
-- `ImageCompare`: Compare image,
+- `BBox From Mask`: From a mask extract the bounding box
+- `Crop`: Crop image from BBox
+- `Uncrop`: Uncrop image from BBox
+
+## colors
+- `Colored Image`: Constant color image of given size
 - `RGB to HSV`: -,
 - `HSV to RGB`: -,
-- `Color Correct`: Basic color correction tools,
-- `Modulo`: Modulo (useful for loops),
+- `Color Correct`: Basic color correction tools  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/7c20ac83-31ff-40ea-a1a0-06c2acefb2ef" width=400/>
+
+## image ops
+- `Blur`: Blur an image using a Gaussian filter.
 - `Deglaze Image`: taken from [FN16](https://github.com/Fannovel16/FN16-ComfyUI-nodes/blob/main/DeglazeImage.py),
+- `Denoise`: Denoise the input image,
+- `Image Compare`: Compare two images and return a difference image
+- `Image Premultiply`: Premultiply image with mask
+- `Image Remove Background Rembg`: [RemBG](https://github.com/danielgatis/rembg) powered background removal.  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/e69253b4-c03c-45e9-92b5-aa46fb887be8" width=320/>
+- `Image Resize Factor`: Extracted mostly from [WAS Node Suite](https://github.com/WASasquatch/was-node-suite-comfyui), with a few edits (most notably multiple image support) and less features.
+- `Mask To Image`: Converts a mask (alpha) to an RGB image with a color and background
+- `Save Image Grid`: Save all the images in the input batch as a grid of images.
+
+## latent utils
+- `Latent Lerp`: Linear interpolation (blend) between two latent 
+
+## textures
+- `Model Patch Seamless`: Use the [seamless diffusion "hack"](https://gitlab.com/-/snippets/2395088) to patch any model to infere seamless images, check the [examples](https://github.com/melMass/comfy_mtb/wiki/Examples) to see how to use all those textures node together  
+  <img width=500 src="https://user-images.githubusercontent.com/7041726/272970506-9db516b5-45d2-4389-b904-b3a94660f24c.png"/>
+- `DeepBump`: Normal & height maps generation from single pictures  
+  <img width=500 src="https://user-images.githubusercontent.com/7041726/272970715-7e4477f6-8e18-4839-9864-83d07d6690a1.png"/>
+- `Image Tile Offset`: Mimics an old photoshop technique to check for seamless textures by offsetting tiles of the image.  
+  <img width=600 src="https://github.com/melMass/comfy_mtb/assets/7041726/cbcc51fb-922f-433f-acf1-c6c6c2a7ffc4" />
+
+## misc utils
+- `Any To String`: Tries to take any input and convert it to a string.
+- `Concat Images`: Takes two image stream and merge them as a batch of images supported by other Comfy pipelines.
+- `Image Resize Factor`: **Deprecated**, I since discovered the builtin image resize.
+- `Text To Image`: Utils to convert text to image using a font
+- `Styles Loader`: Load csv files and populate a dropdown from the rows (à la A111)  
+  <img src="https://github.com/melMass/comfy_mtb/assets/7041726/02fe3211-18ee-4e54-a029-931388f5fde8" width=320/>
 - `Smart Step`: A very basic node to get step percent to use in KSampler advanced,
+- `Qr Code`: Basic QR Code generator
+- `Save Tensors`: Debug node that will probably be removed in the future
+- `Int to Number`: Supplement for WASSuite number nodes
+- `Smart Step`: A very basic tool to control the steps (start/stop) of the `KAdvancedSampler` using percentage
+- `Load Image From Url`: Load an image from the given URL
 
 
-### Comfy Resources
+## Optional nodes
+
+These nodes are still bundled in mtb, but moving forward (>0.2.0) they won't
+be setup by the install script and their dependencies won't install either.
+The reason is mostly that they all have a better alternatives available and tensorflow on windows was not a fun experience and since Python 3.11 not an experience at all.
+
+For linux and mac users though these nodes didn't cause any issue and I personally still use them, these are the extra requirements needed:
+
+```console
+.venv/python -m pip install tensorflow facexlib insightface basicsr
+```
+
+### face detection / swapping
+> **Warning**
+> Those nodes were among the first to be implemented they do work, but on windows the installation is still not properly handled for everyone  
+> As alternatives you can use [reactor](https://github.com/Gourieff/comfyui-reactor-node) for face swap and [facerestore](https://github.com/Haidra-Org/hordelib/tree/main/hordelib/nodes/facerestore) for restoration  
+> You can check [this video](https://www.youtube.com/watch?v=FShlpMxbU0E) for a tutorial by Ferniclestix using these alternatives  
+
+- `Face Swap`: Face swap using deepinsight/insightface models (this node used to be called `Roop` in early versions, it does the same, roop is *just* an app that uses those model)  
+  <img  width=320 src="https://user-images.githubusercontent.com/7041726/260261217-54e33446-183f-4dda-88b3-d38a1e6de980.gif"/>
+- `Load Face Swap Model`: Load an insightface model for face swapping
+- `Restore Face`: Using [GFPGan](https://github.com/TencentARC/GFPGAN) to restore faces, works great in conjunction with `Face Swap` and supports Comfy native upscalers for the `bg_upscaler`
+  
+### image interpolation (animation)
+> **Warning**
+> The FILM nodes will be deprecated at some point after 0.2.0, [Fannovel16](https://github.com/Fannovel16/ComfyUI-Frame-Interpolation)'s interpolation nodes implement it and they rely on a pytorch implementation of FILM
+> which solves the issues related to the ones included in mtb. They will probably remain available if your system meet the requirements and ignored otherwise.
+
+<details><summary>Why?</summary>
+  
+> **Windows only issue**: This requires tensorflow-gpu that is unfortunately not a thing anymore on Windows since 2.10.1 (unless you use a complex WSL passthrough setup but it's still not "Windows")  
+> Using this old version is quite clunky and require some patching that install.py does automatically, but the main issue is that no wheels are available for python > 3.10
+> Comfy-nightly is already using Python 11 so installing this old tf version won't work there.  
+> You can in any case install the normal up to date tensorflow but that will run on CPU and is much MUCH slower for FILM inference.
+</details>
+
+- `Load Film Model`: Loads a [FILM](https://github.com/google-research/frame-interpolation) model
+- `Film Interpolation`: Process input frames using [FILM](https://github.com/google-research/frame-interpolation)  
+  <img width=400 src="https://github.com/melMass/comfy_mtb/assets/7041726/3afd1647-6634-4b92-a34b-51432e6a9834"/>  
+  <img width=400 src="https://user-images.githubusercontent.com/7041726/260259079-c0f04a63-960c-43a7-ba78-a45cd5ac7514.gif"/>
+- `Export to Prores (experimental)`: Exports the input frames to a ProRes 4444 mov file. This is using ffmpeg stdin to send raw numpy arrays, used with `Film Interpolation` and very simple for now but could be expanded upon.
+
+# Comfy Resources
+
+**Misc**
+
+- [Slick ComfyUI by NoCrypt](https://colab.research.google.com/drive/1ZMvLWEiYITmBJngtqeIQToeNuiydwI0z#scrollTo=1fWMaexXS188): A colab notebook with batteries included!
 
 **Guides**:
 - [Official Examples (eng)](https://comfyanonymous.github.io/ComfyUI_examples/)

__init__.py

@@ -1,16 +1,78 @@
-import traceback
-from .log import log, blue_text, get_summary, get_label
-from .utils import here
-import importlib
+#!/usr/bin/env python3
+# -*- coding:utf-8 -*-
+###
+# File: __init__.py
+# Project: comfy_mtb
+# Author: Mel Massadian
+# Copyright (c) 2023 Mel Massadian
+#
+###
 import os
 
+# todo: don't override this if the user has that setup already
+os.environ["TF_FORCE_GPU_ALLOW_GROWTH"] = "true"
+os.environ["TF_GPU_ALLOCATOR"] = "cuda_malloc_async"
+
+import ast
+import contextlib
+import importlib
+import json
+import logging
+import os
+import shutil
+import traceback
+from importlib import reload
+
+from aiohttp import web
+from server import PromptServer
+
+import nodes
+
+from .endpoint import endlog
+from .log import blue_text, cyan_text, get_label, get_summary, log
+from .utils import comfy_dir, here
+
 NODE_CLASS_MAPPINGS = {}
+NODE_DISPLAY_NAME_MAPPINGS = {}
 NODE_CLASS_MAPPINGS_DEBUG = {}
+WEB_DIRECTORY = "./web"
+
+__version__ = "0.2.0"
+
+
+def extract_nodes_from_source(filename):
+    source_code = ""
+
+    with open(filename, "r", encoding="utf8") as file:
+        source_code = file.read()
+
+    nodes = []
+
+    try:
+        parsed = ast.parse(source_code)
+        for node in ast.walk(parsed):
+            if isinstance(node, ast.Assign) and len(node.targets) == 1:
+                target = node.targets[0]
+                if isinstance(target, ast.Name) and target.id == "__nodes__":
+                    value = ast.get_source_segment(source_code, node.value)
+                    node_value = ast.parse(value).body[0].value
+                    if isinstance(node_value, (ast.List, ast.Tuple)):
+                        nodes.extend(
+                            element.id
+                            for element in node_value.elts
+                            if isinstance(element, ast.Name)
+                        )
+                    break
+    except SyntaxError:
+        log.error("Failed to parse")
+    return nodes
 
 
 def load_nodes():
     errors = []
     nodes = []
+    nodes_failed = []
+
     for filename in (here / "nodes").iterdir():
         if filename.suffix == ".py":
             module_name = filename.stem
@@ -21,17 +83,20 @@ def load_nodes():
                 )
                 _nodes = getattr(module, "__nodes__")
                 nodes.extend(_nodes)
-
                 log.debug(f"Imported {module_name} nodes")
 
             except AttributeError:
                 pass  # wip nodes
             except Exception:
                 error_message = traceback.format_exc().splitlines()[-1]
-                errors.append(f"Failed to import {module_name} because {error_message}")
+                errors.append(
+                    f"Failed to import module {module_name} because {error_message}"
+                )
+                # Read __nodes__ variable from the source file
+                nodes_failed.extend(extract_nodes_from_source(filename))
 
     if errors:
-        log.error(
+        log.debug(
             f"Some nodes failed to load:\n\t"
             + "\n\t".join(errors)
             + "\n\n"
@@ -39,36 +104,238 @@ def load_nodes():
             + "If you think this is a bug, please report it on the github page (https://github.com/melMass/comfy_mtb/issues)"
         )
 
-    return nodes
+    return (nodes, nodes_failed)
 
 
 # - REGISTER WEB EXTENSIONS
-web_extensions_root = utils.comfy_dir / "web" / "extensions"
+web_extensions_root = comfy_dir / "web" / "extensions"
 web_mtb = web_extensions_root / "mtb"
 
-if web_mtb.exists():
-    log.debug(f"Web extensions folder found at {web_mtb}")
-elif web_extensions_root.exists():
-    os.symlink((here / "web"), web_mtb.as_posix())
-else:
-    log.error(
-        f"Comfy root probably not found automatically, please copy the folder {web_mtb} manually in the web/extensions folder of ComfyUI"
-    )
+if web_mtb.exists() and hasattr(nodes, "EXTENSION_WEB_DIRS"):
+    try:
+        if web_mtb.is_symlink():
+            web_mtb.unlink()
+        else:
+            shutil.rmtree(web_mtb)
+    except Exception as e:
+        log.warning(
+            f"Failed to remove web mtb directory: {e}\nPlease manually remove it from disk ({web_mtb}) and restart the server."
+        )
+
 
 # - REGISTER NODES
-nodes = load_nodes()
+nodes, failed = load_nodes()
 for node_class in nodes:
     class_name = node_class.__name__
-    class_name = node_class.__name__
-    node_name = f"{get_label(class_name)} (mtb)"
-    NODE_CLASS_MAPPINGS[node_name] = node_class
-    NODE_CLASS_MAPPINGS_DEBUG[node_name] = node_class.__doc__
+    node_label = f"{get_label(class_name)} (mtb)"
+    NODE_CLASS_MAPPINGS[node_label] = node_class
+    NODE_DISPLAY_NAME_MAPPINGS[class_name] = node_label
+    NODE_CLASS_MAPPINGS_DEBUG[node_label] = node_class.__doc__
+    # TODO: I removed this, I find it more convenient to write without spaces, but it breaks every of my workflows
+    # TODO (cont): and until I find a way to automate the conversion, I'll leave it like this
 
+    if os.environ.get("MTB_EXPORT"):
+        with open(here / "node_list.json", "w") as f:
+            f.write(
+                json.dumps(
+                    {
+                        k: NODE_CLASS_MAPPINGS_DEBUG[k]
+                        for k in sorted(NODE_CLASS_MAPPINGS_DEBUG.keys())
+                    },
+                    indent=4,
+                )
+            )
 
 log.debug(
     f"Loaded the following nodes:\n\t"
     + "\n\t".join(
-        f"{k}: {blue_text(get_summary(doc)) if doc else '-'}"
+        f"{cyan_text(k)}: {blue_text(get_summary(doc)) if doc else '-'}"
         for k, doc in NODE_CLASS_MAPPINGS_DEBUG.items()
     )
 )
+
+log.info(f"loaded {cyan_text(len(nodes))} nodes successfuly")
+if failed:
+    with contextlib.suppress(Exception):
+        base_url, port = utils.get_server_info()
+        log.info(
+            f"Some nodes ({len(failed)}) could not be loaded. This can be ignored, but go to http://{base_url}:{port}/mtb if you want more information."
+        )
+
+
+# - ENDPOINT
+
+
+if hasattr(PromptServer, "instance"):
+    restore_deps = ["basicsr"]
+    onnx_deps = ["onnxruntime"]
+    swap_deps = ["insightface"] + onnx_deps
+    node_dependency_mapping = {
+        "QrCode": ["qrcode"],
+        "DeepBump": onnx_deps,
+        "FaceSwap": swap_deps,
+        "LoadFaceSwapModel": swap_deps,
+        "LoadFaceAnalysisModel": restore_deps,
+    }
+
+    PromptServer.instance.app.router.add_static(
+        "/mtb-assets/", path=(here / "html").as_posix()
+    )
+
+    @PromptServer.instance.routes.get("/mtb/manage")
+    async def manage(request):
+        from . import endpoint
+
+        reload(endpoint)
+
+        endlog.debug("Initializing Manager")
+        if "text/html" in request.headers.get("Accept", ""):
+            csv_editor = endpoint.csv_editor()
+
+            tabview = endpoint.render_tab_view(Styles=csv_editor)
+            return web.Response(
+                text=endpoint.render_base_template("MTB", tabview),
+                content_type="text/html",
+            )
+
+        return web.json_response(
+            {
+                "message": "manage only has a POST api for now",
+            }
+        )
+
+    @PromptServer.instance.routes.get("/mtb/status")
+    async def get_full_library(request):
+        from . import endpoint
+
+        reload(endpoint)
+
+        endlog.debug("Getting node registration status")
+        # Check if the request prefers HTML content
+        if "text/html" in request.headers.get("Accept", ""):
+            # # Return an HTML page
+            html_response = endpoint.render_table(
+                NODE_CLASS_MAPPINGS_DEBUG, title="Registered"
+            )
+            html_response += endpoint.render_table(
+                {
+                    k: {"dependencies": node_dependency_mapping.get(k)}
+                    if node_dependency_mapping.get(k)
+                    else "-"
+                    for k in failed
+                },
+                title="Failed to load",
+            )
+
+            return web.Response(
+                text=endpoint.render_base_template("MTB", html_response),
+                content_type="text/html",
+            )
+
+        return web.json_response(
+            {
+                "registered": NODE_CLASS_MAPPINGS_DEBUG,
+                "failed": failed,
+            }
+        )
+
+    @PromptServer.instance.routes.post("/mtb/debug")
+    async def set_debug(request):
+        json_data = await request.json()
+        enabled = json_data.get("enabled")
+        if enabled:
+            os.environ["MTB_DEBUG"] = "true"
+            log.setLevel(logging.DEBUG)
+            log.debug("Debug mode set from API (/mtb/debug POST route)")
+
+        elif "MTB_DEBUG" in os.environ:
+            # del os.environ["MTB_DEBUG"]
+            os.environ.pop("MTB_DEBUG")
+            log.setLevel(logging.INFO)
+
+        return web.json_response(
+            {"message": f"Debug mode {'set' if enabled else 'unset'}"}
+        )
+
+    @PromptServer.instance.routes.get("/mtb")
+    async def get_home(request):
+        from . import endpoint
+
+        reload(endpoint)
+        # Check if the request prefers HTML content
+        if "text/html" in request.headers.get("Accept", ""):
+            # # Return an HTML page
+            html_response = """
+            <div class="flex-container menu">
+                <a href="/mtb/manage">manage</a>
+                <a href="/mtb/debug">debug</a>
+                <a href="/mtb/status">status</a>
+            </div>            
+            """
+            return web.Response(
+                text=endpoint.render_base_template("MTB", html_response),
+                content_type="text/html",
+            )
+
+        # Return JSON for other requests
+        return web.json_response({"message": "Welcome to MTB!"})
+
+    @PromptServer.instance.routes.get("/mtb/debug")
+    async def get_debug(request):
+        from . import endpoint
+
+        reload(endpoint)
+        enabled = "MTB_DEBUG" in os.environ
+        # Check if the request prefers HTML content
+        if "text/html" in request.headers.get("Accept", ""):
+            # # Return an HTML page
+            html_response = f"""
+                <h1>MTB Debug Status: {'Enabled' if enabled else 'Disabled'}</h1>
+            """
+            return web.Response(
+                text=endpoint.render_base_template("Debug", html_response),
+                content_type="text/html",
+            )
+
+        # Return JSON for other requests
+        return web.json_response({"enabled": enabled})
+
+    @PromptServer.instance.routes.get("/mtb/actions")
+    async def no_route(request):
+        from . import endpoint
+
+        if "text/html" in request.headers.get("Accept", ""):
+            html_response = """
+            <h1>Actions has no get for now...</h1>
+            """
+            return web.Response(
+                text=endpoint.render_base_template("Actions", html_response),
+                content_type="text/html",
+            )
+        return web.json_response({"message": "actions has no get for now"})
+
+    @PromptServer.instance.routes.post("/mtb/actions")
+    async def do_action(request):
+        from . import endpoint
+
+        reload(endpoint)
+
+        return await endpoint.do_action(request)
+
+    @PromptServer.instance.routes.get("/mtb/audio")
+    async def get_audio(request):
+        from . import endpoint
+
+        reload(endpoint)
+
+        return await endpoint.get_audio(request)
+
+
+# - WAS Dictionary
+MANIFEST = {
+    "name": "MTB Nodes",  # The title that will be displayed on Node Class menu,. and Node Class view
+    "version": (0, 1, 0),  # Version of the custom_node or sub module
+    "author": "Mel Massadian",  # Author or organization of the custom_node or sub module
+    "project": "https://github.com/melMass/comfy_mtb",  # The address that the `name` value will link to on Node Class Views
+    "description": "Set of nodes that enhance your animation workflow and provide a range of useful tools including features such as manipulating bounding boxes, perform color corrections, swap faces in images, interpolate frames for smooth animation, export to ProRes format, apply various image operations, work with latent spaces, generate QR codes, and create normal and height maps for textures.",
+}

endpoint.py

@@ -0,0 +1,389 @@
+import csv
+
+from aiohttp import web
+
+from .log import mklog
+from .utils import (
+    audioInputDir,
+    backup_file,
+    comfy_dir,
+    here,
+    import_install,
+    reqs_map,
+    run_command,
+    styles_dir,
+)
+
+endlog = mklog("mtb endpoint")
+
+# - ACTIONS
+import platform
+import sys
+from pathlib import Path
+
+import_install("requirements")
+
+
+def ACTIONS_loadAudio(args):
+    if not audioInputDir.exists():
+        audioInputDir.mkdir()
+
+    endlog.debug(f"Received Load Audio request for {args}")
+
+    if not args.file:
+        return web.Response(status=400)
+
+    filename = args.filename
+    if not filename:
+        return web.Response(status=400)
+
+    target = audioInputDir / filename
+    if target.exists():
+        target.unlink()
+
+    with target.open("wb") as f:
+        f.write(args.file.read())
+
+    return {"name": filename}
+
+
+def ACTIONS_installDependency(dependency_names=None):
+    if dependency_names is None:
+        return {"error": "No dependency name provided"}
+    endlog.debug(f"Received Install Dependency request for {dependency_names}")
+    # reqs = []
+    resolved_names = [reqs_map.get(name, name) for name in dependency_names]
+    try:
+        run_command([Path(sys.executable), "-m", "pip", "install"] + resolved_names)
+        return {"success": True}
+
+    except Exception as e:
+        return {"error": f"Failed to install dependencies: {e}"}
+
+    # if platform.system() == "Windows":
+    #     reqs = list(requirements.parse((here / "reqs_windows.txt").read_text()))
+    # else:
+    #     reqs = list(requirements.parse((here / "reqs.txt").read_text()))
+    # print([x.specs for x in reqs])
+    # print(
+    #     "\n".join([f"{x.line} {''.join(x.specs[0] if x.specs else '')}" for x in reqs])
+    # )
+    # for dependency_name in dependency_names:
+    #     for req in reqs:
+    #         if req.name == dependency_name:
+    #             endlog.debug(f"Dependency {dependency_name} installed")
+    #             break
+
+
+def ACTIONS_getStyles(style_name=None):
+    from .nodes.conditions import StylesLoader
+
+    styles = StylesLoader.options
+    match_list = ["name"]
+    if styles:
+        filtered_styles = {
+            key: value
+            for key, value in styles.items()
+            if not key.startswith("__") and key not in match_list
+        }
+        if style_name:
+            return filtered_styles.get(style_name, {"error": "Style not found"})
+        return filtered_styles
+    return {"error": "No styles found"}
+
+
+def ACTIONS_saveStyle(data):
+    # endlog.debug(f"Received Save Styles for {data.keys()}")
+    # endlog.debug(data)
+
+    styles = [f.name for f in styles_dir.iterdir() if f.suffix == ".csv"]
+    target = None
+    rows = []
+    for fp, content in data.items():
+        if fp in styles:
+            endlog.debug(f"Overwriting {fp}")
+            target = styles_dir / fp
+            rows = content
+            break
+
+    if not target:
+        endlog.warning(f"Could not determine the target file for {data.keys()}")
+        return {"error": "Could not determine the target file for the style"}
+
+    backup_file(target)
+
+    with target.open("w", newline="", encoding="utf-8") as file:
+        csv_writer = csv.writer(file, quoting=csv.QUOTE_ALL)
+        for row in rows:
+            csv_writer.writerow(row)
+
+
+async def do_action(request) -> web.Response:
+    endlog.debug("Init action request")
+    request_data = await request.post()
+    name = request_data.get("name")
+    args = request_data.get("args")
+
+    endlog.debug(f"Received action request: {name} {args}")
+
+    method_name = f"ACTIONS_{name}"
+    method = globals().get(method_name)
+
+    if callable(method):
+        result = method(args) if args else method()
+        endlog.debug(f"Action result: {result}")
+        return web.json_response({"result": result}, status=200)
+
+    available_methods = [
+        attr[len("ACTIONS_") :] for attr in globals() if attr.startswith("ACTIONS_")
+    ]
+
+    return web.json_response(
+        {"error": "Invalid method name.", "available_methods": available_methods},
+        status=400,
+    )
+
+
+async def get_audio(request):
+    name = request.rel_url.query.get("filename")
+    if not name:
+        return web.json_response(
+            {"error": "No filename provided as url query."}, status=400
+        )
+
+    target = audioInputDir / name
+    if not target.exists():
+        return web.json_response(
+            {"error": f"File {name} (in {audioInputDir}) not found..."}, status=404
+        )
+
+    return web.FileResponse(
+        target, headers={"Content-Disposition": f'filename="{name}"'}
+    )
+
+
+# - HTML UTILS
+
+
+def dependencies_button(name, dependencies):
+    deps = ",".join([f"'{x}'" for x in dependencies])
+    return f"""
+        <button class="dependency-button" onclick="window.mtb_action('installDependency',[{deps}])">Install {name} deps</button>
+        """
+
+
+def csv_editor():
+    inputs = [f for f in styles_dir.iterdir() if f.suffix == ".csv"]
+    # rows = {f.stem: list(csv.reader(f.read_text("utf8"))) for f in styles}
+
+    style_files = {}
+    for file in inputs:
+        with open(file, "r", encoding="utf8") as f:
+            parsed = csv.reader(f)
+            style_files[file.name] = []
+            for row in parsed:
+                endlog.debug(f"Adding style {row[0]}")
+                style_files[file.name].append((row[0], row[1], row[2]))
+
+    html_out = """
+            <div id="style-editor">
+             <h1>Style Editor</h1>
+            
+            """
+    for current, styles in style_files.items():
+        current_out = f"<h3>{current}</h3>"
+        table_rows = []
+        for index, style in enumerate(styles):
+            table_rows += (
+                (["<tr>"] + [f"<th>{cell}</th>" for cell in style] + ["</tr>"])
+                if index == 0
+                else (
+                    ["<tr>"]
+                    + [
+                        f"<td><input type='text' value='{cell}'></td>"
+                        if i == 0
+                        else f"<td><textarea name='Text1' cols='40' rows='5'>{cell}</textarea></td>"
+                        for i, cell in enumerate(style)
+                    ]
+                    + ["</tr>"]
+                )
+            )
+        current_out += (
+            f"<table data-id='{current}' data-filename='{current}'>"
+            + "".join(table_rows)
+            + "</table>"
+        )
+        current_out += f"<button data-id='{current}' onclick='saveTableData(this.getAttribute(\"data-id\"))'>Save {current}</button>"
+
+        html_out += add_foldable_region(current, current_out)
+
+    html_out += "</div>"
+    html_out += """<script src='/mtb-assets/js/saveTableData.js'></script>"""
+
+    return html_out
+
+
+def render_tab_view(**kwargs):
+    tab_headers = []
+    tab_contents = []
+
+    for idx, (tab_name, content) in enumerate(kwargs.items()):
+        active_class = "active" if idx == 0 else ""
+        tab_headers.append(
+            f"<button class='tablinks {active_class}' onclick=\"openTab(event, '{tab_name}')\">{tab_name}</button>"
+        )
+        tab_contents.append(
+            f"<div id='{tab_name}' class='tabcontent {active_class}'>{content}</div>"
+        )
+
+    headers_str = "\n".join(tab_headers)
+    contents_str = "\n".join(tab_contents)
+
+    return f"""
+<div class='tab-container'>
+    <div class='tab'>
+        {headers_str}
+    </div>
+    {contents_str}
+    </div>
+    <script src='/mtb-assets/js/tabSwitch.js'></script>
+    """
+
+
+def add_foldable_region(title, content):
+    symbol_id = f"{title}-symbol"
+    return f"""
+    <div class='foldable'>
+        <div class='foldable-title' onclick="toggleFoldable('{title}', '{symbol_id}')">
+            <span id='{symbol_id}' class='foldable-symbol'>&#9655;</span>
+            {title}
+        </div>
+        <div id='{title}' class='foldable-content'>
+            {content}
+        </div>
+    </div>
+    <script src='/mtb-assets/js/foldable.js'></script>
+    """
+
+
+def add_split_pane(left_content, right_content, vertical=True):
+    orientation = "vertical" if vertical else "horizontal"
+    return f"""
+    <div class="split-pane {orientation}">
+        <div id="leftPane">
+            {left_content}
+        </div>
+        <div id="resizer"></div>
+        <div id="rightPane">
+            {right_content}
+        </div>
+    </div>
+    <script>
+        initSplitPane({str(vertical).lower()});
+    </script>
+    <script src='/mtb-assets/js/splitPane.js'></script>
+    """
+
+
+def add_dropdown(title, options):
+    option_str = "\n".join([f"<option value='{opt}'>{opt}</option>" for opt in options])
+    return f"""
+    <select>
+        <option disabled selected>{title}</option>
+        {option_str}
+    </select>
+    """
+
+
+def render_table(table_dict, sort=True, title=None):
+    table_dict = sorted(
+        table_dict.items(), key=lambda item: item[0]
+    )  # Sort the dictionary by keys
+
+    table_rows = ""
+    for name, item in table_dict:
+        if isinstance(item, dict):
+            if "dependencies" in item:
+                table_rows += f"<tr><td>{name}</td><td>"
+                table_rows += f"{dependencies_button(name,item['dependencies'])}"
+
+                table_rows += "</td></tr>"
+            else:
+                table_rows += f"<tr><td>{name}</td><td>{render_table(item)}</td></tr>"
+        # elif isinstance(item, str):
+        #     table_rows += f"<tr><td>{name}</td><td>{item}</td></tr>"
+        else:
+            table_rows += f"<tr><td>{name}</td><td>{item}</td></tr>"
+
+    return f"""
+        <div class="table-container">
+        {"" if title is None else f"<h1>{title}</h1>"}
+        <table>
+            <thead>
+                <tr>
+                    <th>Name</th>
+                    <th>Description</th>
+                </tr>
+            </thead>
+            <tbody>
+                {table_rows}
+            </tbody>
+        </table>      
+        </div>
+        """
+
+
+def render_base_template(title, content):
+    github_icon_svg = """<svg xmlns="http://www.w3.org/2000/svg" fill="whitesmoke" height="3em" viewBox="0 0 496 512"><path d="M165.9 397.4c0 2-2.3 3.6-5.2 3.6-3.3.3-5.6-1.3-5.6-3.6 0-2 2.3-3.6 5.2-3.6 3-.3 5.6 1.3 5.6 3.6zm-31.1-4.5c-.7 2 1.3 4.3 4.3 4.9 2.6 1 5.6 0 6.2-2s-1.3-4.3-4.3-5.2c-2.6-.7-5.5.3-6.2 2.3zm44.2-1.7c-2.9.7-4.9 2.6-4.6 4.9.3 2 2.9 3.3 5.9 2.6 2.9-.7 4.9-2.6 4.6-4.6-.3-1.9-3-3.2-5.9-2.9zM244.8 8C106.1 8 0 113.3 0 252c0 110.9 69.8 205.8 169.5 239.2 12.8 2.3 17.3-5.6 17.3-12.1 0-6.2-.3-40.4-.3-61.4 0 0-70 15-84.7-29.8 0 0-11.4-29.1-27.8-36.6 0 0-22.9-15.7 1.6-15.4 0 0 24.9 2 38.6 25.8 21.9 38.6 58.6 27.5 72.9 20.9 2.3-16 8.8-27.1 16-33.7-55.9-6.2-112.3-14.3-112.3-110.5 0-27.5 7.6-41.3 23.6-58.9-2.6-6.5-11.1-33.3 2.6-67.9 20.9-6.5 69 27 69 27 20-5.6 41.5-8.5 62.8-8.5s42.8 2.9 62.8 8.5c0 0 48.1-33.6 69-27 13.7 34.7 5.2 61.4 2.6 67.9 16 17.7 25.8 31.5 25.8 58.9 0 96.5-58.9 104.2-114.8 110.5 9.2 7.9 17 22.9 17 46.4 0 33.7-.3 75.4-.3 83.6 0 6.5 4.6 14.4 17.3 12.1C428.2 457.8 496 362.9 496 252 496 113.3 383.5 8 244.8 8zM97.2 352.9c-1.3 1-1 3.3.7 5.2 1.6 1.6 3.9 2.3 5.2 1 1.3-1 1-3.3-.7-5.2-1.6-1.6-3.9-2.3-5.2-1zm-10.8-8.1c-.7 1.3.3 2.9 2.3 3.9 1.6 1 3.6.7 4.3-.7.7-1.3-.3-2.9-2.3-3.9-2-.6-3.6-.3-4.3.7zm32.4 35.6c-1.6 1.3-1 4.3 1.3 6.2 2.3 2.3 5.2 2.6 6.5 1 1.3-1.3.7-4.3-1.3-6.2-2.2-2.3-5.2-2.6-6.5-1zm-11.4-14.7c-1.6 1-1.6 3.6 0 5.9 1.6 2.3 4.3 3.3 5.6 2.3 1.6-1.3 1.6-3.9 0-6.2-1.4-2.3-4-3.3-5.6-2z"/></svg>"""
+    return f"""
+    <!DOCTYPE html>
+    <html>
+    <head>
+        <title>{title}</title>
+        <link rel="stylesheet" href="/mtb-assets/style.css"/>
+    </head>
+    <script type="module">
+        import {{ api }} from '/scripts/api.js'
+        const mtb_action = async (action, args) =>{{
+            console.log(`Sending ${{action}} with args: ${{args}}`)
+            }}
+        window.mtb_action = async (action, args) =>{{
+            console.log(`Sending ${{action}} with args: ${{args}} to the API`)
+            const res = await api.fetchApi('/actions', {{
+                method: 'POST',
+                body: JSON.stringify({{
+                  name: action,
+                  args,
+                }}),
+            }})
+
+              const output = await res.json()
+              console.debug(`Received ${{action}} response:`, output)
+              if (output?.result?.error){{
+                  alert(`An error occured: {{output?.result?.error}}`)
+              }}
+              return output?.result
+        }}
+    </script>
+    <body>
+        <header>
+        <a href="/">Back to Comfy</a>
+        <div class="mtb_logo">
+            <img src="https://repository-images.githubusercontent.com/649047066/a3eef9a7-20dd-4ef9-b839-884502d4e873" alt="Comfy MTB Logo" height="70" width="128">
+            <span class="title">Comfy MTB</span></div>
+            <a style="width:128px;text-align:center" href="https://www.github.com/melmass/comfy_mtb">
+                {github_icon_svg}
+            </a>
+        </header>
+
+        <main>
+            {content}
+        </main>
+
+        <footer>
+            <!-- Shared footer content here -->
+        </footer>
+    </body>
+    
+    </html>
+    """

errors.py

@@ -0,0 +1,7 @@
+class ModelNotFound(Exception):
+    def __init__(self, model_name, *args, **kwargs):
+        super().__init__(
+            f"The model {model_name} could not be found, make sure to download it using ComfyManager first.\nrepository: https://github.com/ltdrdata/ComfyUI-Manager",
+            *args,
+            **kwargs,
+        )

examples/05-seamless_texture.json

@@ -0,0 +1,905 @@
+{
+  "last_node_id": 97,
+  "last_link_id": 179,
+  "nodes": [
+    {
+      "id": 6,
+      "type": "CLIPTextEncode",
+      "pos": [
+        -1165.8749246009997,
+        30
+      ],
+      "size": [
+        422.84503173828125,
+        164.31304931640625
+      ],
+      "flags": {},
+      "order": 4,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "clip",
+          "type": "CLIP",
+          "link": 3
+        }
+      ],
+      "outputs": [
+        {
+          "name": "CONDITIONING",
+          "type": "CONDITIONING",
+          "links": [
+            4,
+            158
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "CLIPTextEncode"
+      },
+      "widgets_values": [
+        "Closeup texture of rocks"
+      ],
+      "color": "#432",
+      "bgcolor": "#653",
+      "shape": 1
+    },
+    {
+      "id": 7,
+      "type": "CLIPTextEncode",
+      "pos": [
+        -1175.8749246009997,
+        250
+      ],
+      "size": [
+        425.27801513671875,
+        180.6060791015625
+      ],
+      "flags": {},
+      "order": 5,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "clip",
+          "type": "CLIP",
+          "link": 5
+        }
+      ],
+      "outputs": [
+        {
+          "name": "CONDITIONING",
+          "type": "CONDITIONING",
+          "links": [
+            6
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "CLIPTextEncode"
+      },
+      "widgets_values": [
+        "((drawing, cartoon, painting, sketch, blur, depth of field, dof))"
+      ],
+      "color": "#432",
+      "bgcolor": "#653",
+      "shape": 1
+    },
+    {
+      "id": 89,
+      "type": "Reroute",
+      "pos": [
+        350,
+        803
+      ],
+      "size": [
+        75,
+        26
+      ],
+      "flags": {},
+      "order": 9,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "",
+          "type": "*",
+          "link": 176
+        }
+      ],
+      "outputs": [
+        {
+          "name": "",
+          "type": "IMAGE",
+          "links": [
+            167
+          ]
+        }
+      ],
+      "properties": {
+        "showOutputText": false,
+        "horizontal": false
+      }
+    },
+    {
+      "id": 4,
+      "type": "CheckpointLoaderSimple",
+      "pos": [
+        -1740,
+        236
+      ],
+      "size": [
+        315,
+        98
+      ],
+      "flags": {},
+      "order": 0,
+      "mode": 0,
+      "outputs": [
+        {
+          "name": "MODEL",
+          "type": "MODEL",
+          "links": [
+            170
+          ],
+          "slot_index": 0
+        },
+        {
+          "name": "CLIP",
+          "type": "CLIP",
+          "links": [
+            3,
+            5
+          ],
+          "slot_index": 1
+        },
+        {
+          "name": "VAE",
+          "type": "VAE",
+          "links": [],
+          "slot_index": 2
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "CheckpointLoaderSimple"
+      },
+      "widgets_values": [
+        "revAnimated_v122.safetensors"
+      ],
+      "shape": 1
+    },
+    {
+      "id": 63,
+      "type": "SaveImage",
+      "pos": [
+        1315,
+        18
+      ],
+      "size": [
+        539.2050170898438,
+        617.2159423828125
+      ],
+      "flags": {},
+      "order": 13,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 115
+        }
+      ],
+      "title": "Normal",
+      "properties": {},
+      "widgets_values": [
+        "Normal"
+      ],
+      "shape": 1
+    },
+    {
+      "id": 67,
+      "type": "SaveImage",
+      "pos": [
+        2095,
+        22
+      ],
+      "size": [
+        539.2050170898438,
+        617.2159423828125
+      ],
+      "flags": {},
+      "order": 16,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 119
+        }
+      ],
+      "title": "Curvature",
+      "properties": {},
+      "widgets_values": [
+        "Curvature"
+      ],
+      "shape": 1
+    },
+    {
+      "id": 69,
+      "type": "SaveImage",
+      "pos": [
+        1560,
+        1290
+      ],
+      "size": [
+        539.2050170898438,
+        617.2159423828125
+      ],
+      "flags": {},
+      "order": 17,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 121
+        }
+      ],
+      "title": "Depth",
+      "properties": {},
+      "widgets_values": [
+        "Height"
+      ],
+      "shape": 1
+    },
+    {
+      "id": 91,
+      "type": "Model Patch Seamless (mtb)",
+      "pos": [
+        -1150,
+        -146
+      ],
+      "size": [
+        430.8000183105469,
+        78
+      ],
+      "flags": {},
+      "order": 3,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "model",
+          "type": "MODEL",
+          "link": 170
+        }
+      ],
+      "outputs": [
+        {
+          "name": "Original Model (passthrough)",
+          "type": "MODEL",
+          "links": null,
+          "shape": 3
+        },
+        {
+          "name": "Patched Model",
+          "type": "MODEL",
+          "links": [
+            169
+          ],
+          "shape": 3,
+          "slot_index": 1
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "Model Patch Seamless (mtb)"
+      },
+      "widgets_values": [
+        true
+      ],
+      "color": "#232",
+      "bgcolor": "#353"
+    },
+    {
+      "id": 93,
+      "type": "PreviewImage",
+      "pos": [
+        1115,
+        -597
+      ],
+      "size": [
+        451.3526306152344,
+        478.3444519042969
+      ],
+      "flags": {},
+      "order": 12,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 179
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "PreviewImage"
+      }
+    },
+    {
+      "id": 43,
+      "type": "VAELoader",
+      "pos": [
+        -598.2757622278747,
+        577.3595309932109
+      ],
+      "size": [
+        387.48089599609375,
+        70.60645294189453
+      ],
+      "flags": {},
+      "order": 1,
+      "mode": 0,
+      "outputs": [
+        {
+          "name": "VAE",
+          "type": "VAE",
+          "links": [
+            174
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "VAELoader"
+      },
+      "widgets_values": [
+        "vae-ft-mse-840000-ema-pruned.safetensors"
+      ],
+      "shape": 1
+    },
+    {
+      "id": 97,
+      "type": "Image Tile Offset (mtb)",
+      "pos": [
+        617,
+        -598
+      ],
+      "size": [
+        315,
+        58
+      ],
+      "flags": {},
+      "order": 10,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "image",
+          "type": "IMAGE",
+          "link": 178
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            179
+          ],
+          "shape": 3,
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "Image Tile Offset (mtb)"
+      },
+      "widgets_values": [
+        2
+      ],
+      "color": "#232",
+      "bgcolor": "#353"
+    },
+    {
+      "id": 96,
+      "type": "Vae Decode (mtb)",
+      "pos": [
+        -52,
+        40
+      ],
+      "size": [
+        315,
+        126
+      ],
+      "flags": {},
+      "order": 7,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "samples",
+          "type": "LATENT",
+          "link": 173
+        },
+        {
+          "name": "vae",
+          "type": "VAE",
+          "link": 174
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            175,
+            176,
+            178
+          ],
+          "shape": 3,
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "Vae Decode (mtb)"
+      },
+      "widgets_values": [
+        true,
+        false,
+        512
+      ],
+      "color": "#232",
+      "bgcolor": "#353"
+    },
+    {
+      "id": 46,
+      "type": "SaveImage",
+      "pos": [
+        533,
+        25
+      ],
+      "size": [
+        539.2050170898438,
+        617.2159423828125
+      ],
+      "flags": {},
+      "order": 8,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 175
+        }
+      ],
+      "title": "Albedo",
+      "properties": {},
+      "widgets_values": [
+        "Albedo"
+      ],
+      "shape": 1
+    },
+    {
+      "id": 74,
+      "type": "EmptyLatentImage",
+      "pos": [
+        -1075.8749246009997,
+        480
+      ],
+      "size": [
+        315,
+        106
+      ],
+      "flags": {},
+      "order": 2,
+      "mode": 0,
+      "outputs": [
+        {
+          "name": "LATENT",
+          "type": "LATENT",
+          "links": [
+            132
+          ],
+          "shape": 3,
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "EmptyLatentImage"
+      },
+      "widgets_values": [
+        768,
+        768,
+        1
+      ],
+      "color": "#323",
+      "bgcolor": "#535",
+      "shape": 1
+    },
+    {
+      "id": 62,
+      "type": "Deep Bump (mtb)",
+      "pos": [
+        727,
+        801
+      ],
+      "size": [
+        315,
+        130
+      ],
+      "flags": {},
+      "order": 11,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "image",
+          "type": "IMAGE",
+          "link": 167
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            115,
+            118,
+            122
+          ],
+          "shape": 3,
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "Deep Bump (mtb)"
+      },
+      "widgets_values": [
+        "Color to Normals",
+        "SMALL",
+        "SMALLEST",
+        true
+      ],
+      "color": "#232",
+      "bgcolor": "#353",
+      "shape": 1
+    },
+    {
+      "id": 66,
+      "type": "Deep Bump (mtb)",
+      "pos": [
+        1626,
+        808
+      ],
+      "size": [
+        315,
+        130
+      ],
+      "flags": {},
+      "order": 14,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "image",
+          "type": "IMAGE",
+          "link": 118
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            119
+          ],
+          "shape": 3,
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "Deep Bump (mtb)"
+      },
+      "widgets_values": [
+        "Normals to Curvature",
+        "SMALL",
+        "SMALLEST",
+        true
+      ],
+      "color": "#232",
+      "bgcolor": "#353",
+      "shape": 1
+    },
+    {
+      "id": 68,
+      "type": "Deep Bump (mtb)",
+      "pos": [
+        1185,
+        1288
+      ],
+      "size": [
+        315,
+        130
+      ],
+      "flags": {},
+      "order": 15,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "image",
+          "type": "IMAGE",
+          "link": 122
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            121
+          ],
+          "shape": 3,
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "Deep Bump (mtb)"
+      },
+      "widgets_values": [
+        "Normals to Height",
+        "SMALL",
+        "SMALLEST",
+        true
+      ],
+      "color": "#232",
+      "bgcolor": "#353",
+      "shape": 1
+    },
+    {
+      "id": 3,
+      "type": "KSampler",
+      "pos": [
+        -518.2757622278748,
+        47.359530993211024
+      ],
+      "size": [
+        315,
+        474
+      ],
+      "flags": {},
+      "order": 6,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "model",
+          "type": "MODEL",
+          "link": 169
+        },
+        {
+          "name": "positive",
+          "type": "CONDITIONING",
+          "link": 4
+        },
+        {
+          "name": "negative",
+          "type": "CONDITIONING",
+          "link": 6
+        },
+        {
+          "name": "latent_image",
+          "type": "LATENT",
+          "link": 132
+        }
+      ],
+      "outputs": [
+        {
+          "name": "LATENT",
+          "type": "LATENT",
+          "links": [
+            173
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "KSampler"
+      },
+      "widgets_values": [
+        1001,
+        "fixed",
+        28,
+        8,
+        "dpmpp_2m",
+        "normal",
+        1
+      ],
+      "color": "#222",
+      "bgcolor": "#000",
+      "shape": 1
+    }
+  ],
+  "links": [
+    [
+      3,
+      4,
+      1,
+      6,
+      0,
+      "CLIP"
+    ],
+    [
+      4,
+      6,
+      0,
+      3,
+      1,
+      "CONDITIONING"
+    ],
+    [
+      5,
+      4,
+      1,
+      7,
+      0,
+      "CLIP"
+    ],
+    [
+      6,
+      7,
+      0,
+      3,
+      2,
+      "CONDITIONING"
+    ],
+    [
+      115,
+      62,
+      0,
+      63,
+      0,
+      "IMAGE"
+    ],
+    [
+      118,
+      62,
+      0,
+      66,
+      0,
+      "IMAGE"
+    ],
+    [
+      119,
+      66,
+      0,
+      67,
+      0,
+      "IMAGE"
+    ],
+    [
+      121,
+      68,
+      0,
+      69,
+      0,
+      "IMAGE"
+    ],
+    [
+      122,
+      62,
+      0,
+      68,
+      0,
+      "IMAGE"
+    ],
+    [
+      132,
+      74,
+      0,
+      3,
+      3,
+      "LATENT"
+    ],
+    [
+      158,
+      6,
+      0,
+      86,
+      0,
+      "*"
+    ],
+    [
+      167,
+      89,
+      0,
+      62,
+      0,
+      "IMAGE"
+    ],
+    [
+      169,
+      91,
+      1,
+      3,
+      0,
+      "MODEL"
+    ],
+    [
+      170,
+      4,
+      0,
+      91,
+      0,
+      "MODEL"
+    ],
+    [
+      173,
+      3,
+      0,
+      96,
+      0,
+      "LATENT"
+    ],
+    [
+      174,
+      43,
+      0,
+      96,
+      1,
+      "VAE"
+    ],
+    [
+      175,
+      96,
+      0,
+      46,
+      0,
+      "IMAGE"
+    ],
+    [
+      176,
+      96,
+      0,
+      89,
+      0,
+      "*"
+    ],
+    [
+      178,
+      96,
+      0,
+      97,
+      0,
+      "IMAGE"
+    ],
+    [
+      179,
+      97,
+      0,
+      93,
+      0,
+      "IMAGE"
+    ]
+  ],
+  "groups": [
+    {
+      "title": "Seamless Diffusion",
+      "bounding": [
+        -1752,
+        -392,
+        1658,
+        1102
+      ],
+      "color": "#3f789e",
+      "font_size": 76,
+      "locked": false
+    },
+    {
+      "title": "Seamless Check",
+      "bounding": [
+        421,
+        -795,
+        1374,
+        763
+      ],
+      "color": "#3f789e",
+      "font_size": 76,
+      "locked": false
+    }
+  ],
+  "config": {},
+  "extra": {},
+  "version": 0.4
+}

examples/README.md

@@ -0,0 +1,12 @@
+# Examples
+All the examples use the [RevAnimated model 1.22](https://civitai.com/models/7371?modelVersionId=46846)
+## 01 Faceswap
+
+This example showcase the `Face Swap` & `Restore Face` nodes to replace the character with Georges Lucas's face.
+The face reference image is using the `Load Image From Url` node to avoid bundling input images.
+
+<img src="https://github.com/melMass/comfy_mtb/assets/7041726/272af7d6-f01c-478e-a82f-926e772d7209" width=500/>
+
+## 02 FILM interpolation
+This example showcase the FILM interpolation implementation. Here we do text replacement on the condition of two distinct images sharing the same model, input latent & seed to get relatively close images.  
+<img src="https://github.com/melMass/comfy_mtb/assets/7041726/4c28dd87-89fc-4d27-910a-0a1fcf28cdc0" width=500/>

html/js/foldable.js

@@ -0,0 +1,20 @@
+/**
+ * File: foldable.js
+ * Project: comfy_mtb
+ * Author: Mel Massadian
+ *
+ * Copyright (c) 2023 Mel Massadian
+ *
+ */
+
+function toggleFoldable(elementId, symbolId) {
+  const content = document.getElementById(elementId)
+  const symbol = document.getElementById(symbolId)
+  if (content.style.display === 'none' || content.style.display === '') {
+    content.style.display = 'flex'
+    symbol.innerHTML = '▽' // Down arrow
+  } else {
+    content.style.display = 'none'
+    symbol.innerHTML = '▷' // Right arrow
+  }
+}

html/js/saveTableData.js

@@ -0,0 +1,54 @@
+/**
+ * File: saveTableData.js
+ * Project: comfy_mtb
+ * Author: Mel Massadian
+ *
+ * Copyright (c) 2023 Mel Massadian
+ *
+ */
+
+function saveTableData(identifier) {
+  const table = document.querySelector(
+    `#style-editor table[data-id='${identifier}']`
+  )
+
+  let currentData = []
+  const rows = table.querySelectorAll('tr')
+  const filename = table.getAttribute('data-id')
+
+  rows.forEach((row, rowIndex) => {
+    const rowData = []
+    const cells =
+      rowIndex === 0
+        ? row.querySelectorAll('th')
+        : row.querySelectorAll('td input, td textarea')
+
+    cells.forEach((cell) => {
+      rowData.push(rowIndex === 0 ? cell.textContent : cell.value)
+    })
+
+    currentData.push(rowData)
+  })
+
+  let tablesData = {}
+  tablesData[filename] = currentData
+
+  console.debug('Sending styles to manage endpoint:', tablesData)
+  fetch('/mtb/actions', {
+    method: 'POST',
+    headers: {
+      'Content-Type': 'application/json',
+    },
+    body: JSON.stringify({
+      name: 'saveStyle',
+      args: tablesData,
+    }),
+  })
+    .then((response) => response.json())
+    .then((data) => {
+      console.debug('Success:', data)
+    })
+    .catch((error) => {
+      console.error('Error:', error)
+    })
+}

html/js/splitPane.js

@@ -0,0 +1,34 @@
+/**
+ * File: splitPane.js
+ * Project: comfy_mtb
+ * Author: Mel Massadian
+ *
+ * Copyright (c) 2023 Mel Massadian
+ *
+ */
+
+function initSplitPane(vertical) {
+  let resizer = document.getElementById('resizer')
+  let left = document.getElementById('leftPane')
+  let right = document.getElementById('rightPane')
+  resizer.addEventListener('mousedown', function (e) {
+    document.addEventListener('mousemove', onMouseMove)
+    document.addEventListener('mouseup', function () {
+      document.removeEventListener('mousemove', onMouseMove)
+    })
+  })
+
+  const onMouseMove = (e) => {
+    if (vertical) {
+      let leftWidth = e.clientX
+      let rightWidth = window.innerWidth - e.clientX
+      left.style.width = leftWidth + 'px'
+      right.style.width = rightWidth + 'px'
+    } else {
+      let topHeight = e.clientY
+      let bottomHeight = window.innerHeight - e.clientY
+      left.style.height = topHeight + 'px'
+      right.style.height = bottomHeight + 'px'
+    }
+  }
+}

html/js/tabSwitch.js

@@ -0,0 +1,22 @@
+/**
+ * File: tabSwitch.js
+ * Project: comfy_mtb
+ * Author: Mel Massadian
+ *
+ * Copyright (c) 2023 Mel Massadian
+ *
+ */
+
+function openTab(evt, tabName) {
+  var i, tabcontent, tablinks
+  tabcontent = document.getElementsByClassName('tabcontent')
+  for (i = 0; i < tabcontent.length; i++) {
+    tabcontent[i].style.display = 'none'
+  }
+  tablinks = document.getElementsByClassName('tablinks')
+  for (i = 0; i < tablinks.length; i++) {
+    tablinks[i].className = tablinks[i].className.replace(' active', '')
+  }
+  document.getElementById(tabName).style.display = 'block'
+  evt.currentTarget.className += ' active'
+}

html/style.css

@@ -0,0 +1,228 @@
+html {
+    height: 100%;
+    margin: 0;
+    padding: 0;
+    background-color: rgb(33, 33, 33);
+    color: whitesmoke;
+}
+
+a {
+    color: whitesmoke;
+
+}
+
+.table-container {
+    width: 70%;
+    height: 100%;
+    overflow: auto;
+}
+
+table {
+    width: 100%;
+    border-collapse: collapse;
+}
+
+th,
+td {
+    padding: 10px;
+    text-align: left;
+}
+
+th {
+    background-color: rgb(45, 45, 45);
+    /* Light gray background for header row */
+    font-weight: bold;
+}
+
+tr:nth-child(even) {
+    background-color: rgb(45, 45, 45);
+    /* Alternate row background color */
+}
+
+tr:hover {
+    background-color: #797979;
+    /* Highlight color on hover */
+}
+
+td:nth-child(2) {
+    /* Applies to the second column (Description) */
+    width: 80%;
+    /* Adjust the width as needed */
+    word-wrap: break-word;
+    /* Allow long words to be broken and wrapped to the next line */
+}
+
+.mtb_logo {
+    display: flex;
+    flex-direction: column;
+    align-items: center;
+}
+
+/* Styling for WebKit-based browsers (Chrome, Edge) */
+.table-container::-webkit-scrollbar {
+    width: 10px;
+    /* Set the width of the scrollbar */
+}
+
+.table-container::-webkit-scrollbar-thumb {
+    background-color: #797979;
+    /* Color of the scrollbar thumb */
+}
+
+/* Styling for Firefox */
+.table-container {
+    scrollbar-width: thin;
+    /* Set the width of the scrollbar */
+}
+
+.table-container::-webkit-scrollbar-thumb {
+    background-color: #797979;
+    /* Color of the scrollbar thumb */
+}
+
+/* Optionally, you can also style the scrollbar track (background) */
+.table-container::-webkit-scrollbar-track {
+    background-color: #f2f2f2;
+}
+
+
+
+body {
+    margin: 0;
+    padding: 0;
+    font-family: monospace;
+    height: 100%;
+    background-color: rgb(33, 33, 33);
+
+}
+
+.title {
+    font-size: 2.5em;
+    font-weight: 700;
+
+}
+
+header {
+    display: flex;
+    align-items: center;
+    vertical-align: middle;
+    justify-content: space-between;
+    background-color: rgb(12, 12, 12);
+    padding: 1em;
+    margin: 0;
+}
+
+main {
+    display: flex;
+    align-items: center;
+    vertical-align: middle;
+    justify-content: center;
+    padding: 1em;
+    margin: 0;
+    /* height: 80%; */
+}
+
+.flex-container {
+    display: flex;
+    flex-direction: column;
+}
+
+.menu {
+    font-size: 3em;
+    text-align: center;
+}
+
+input, button, textarea {
+    background-color: rgba(0,0,0,0.5);
+    color: white;
+    border: none;
+}
+
+button:hover {
+    background-color: rgba(0,0,0,0.3);
+
+}
+button {
+    padding: 14px 16px;
+
+}
+/* -STYLES EDITOR */
+
+#style-editor {
+    display: flex;
+    flex-direction: column;
+    width:100%;
+    
+}
+
+#style-editor > table {
+    /* background-color: red; */
+    width:100%;
+}
+#style-editor input, #style-editor textarea {
+    /* background-color: blue; */
+    width:100%;
+}
+
+#style-editor td{
+    width: 33.33%;
+}
+
+/* -TABS */
+
+.tab {
+    overflow: hidden;
+    width: 100%;
+    display: flex;
+    flex-direction: row;
+
+}
+
+.tab-container{
+    width: 100%;
+    display: flex;
+    flex-direction: column;
+}
+
+.tab button {
+    background-color: transparent;
+    color:white;
+    float: left;
+    border: none;
+    outline: none;
+    cursor: pointer;
+    padding: 14px 16px;
+    transition: 0.3s;
+    width:100%;
+    font-size: 1.5em;
+}
+
+.tab button.active {
+    background-color: #2e2e2e;
+}
+
+.tabcontent {
+    display: none;
+}
+
+.tabcontent.active {
+    display: block;
+}
+
+
+
+.foldable-title {
+    cursor: pointer;
+    font-weight: bold;
+    user-select: none;
+}
+
+.foldable-symbol {
+    margin-right: 10px;
+}
+
+.foldable-content {
+    display: none;
+    flex-direction: column;
+    margin-left: 20px;
+}

install.py

@@ -0,0 +1,421 @@
+import argparse
+import ast
+import os
+import platform
+import shlex
+import stat
+import subprocess
+import sys
+from contextlib import contextmanager
+from importlib import import_module
+from pathlib import Path
+
+import requests
+
+# region constants
+here = Path(__file__).parent
+executable = Path(sys.executable)
+
+# - detect mode
+mode = None
+if os.environ.get("COLAB_GPU"):
+    mode = "colab"
+elif "python_embeded" in str(executable):
+    mode = "embeded"
+elif ".venv" in str(executable):
+    mode = "venv"
+
+
+if mode is None:
+    mode = "unknown"
+
+repo_url = "https://github.com/melmass/comfy_mtb.git"
+repo_owner = "melmass"
+repo_name = "comfy_mtb"
+short_platform = {
+    "windows": "win_amd64",
+    "linux": "linux_x86_64",
+}
+current_platform = platform.system().lower()
+pip_map = {
+    "onnxruntime-gpu": "onnxruntime",
+    "opencv-contrib": "cv2",
+    "tb-nightly": "tensorboard",
+    "protobuf": "google.protobuf",
+    "qrcode[pil]": "qrcode",
+    "requirements-parser": "requirements"
+    # Add more mappings as needed
+}
+
+# endregion
+
+# region ansi
+# ANSI escape sequences for text styling
+ANSI_FORMATS = {
+    "reset": "\033[0m",
+    "bold": "\033[1m",
+    "dim": "\033[2m",
+    "italic": "\033[3m",
+    "underline": "\033[4m",
+    "blink": "\033[5m",
+    "reverse": "\033[7m",
+    "strike": "\033[9m",
+}
+
+ANSI_COLORS = {
+    "black": "\033[30m",
+    "red": "\033[31m",
+    "green": "\033[32m",
+    "yellow": "\033[33m",
+    "blue": "\033[34m",
+    "magenta": "\033[35m",
+    "cyan": "\033[36m",
+    "white": "\033[37m",
+    "bright_black": "\033[30;1m",
+    "bright_red": "\033[31;1m",
+    "bright_green": "\033[32;1m",
+    "bright_yellow": "\033[33;1m",
+    "bright_blue": "\033[34;1m",
+    "bright_magenta": "\033[35;1m",
+    "bright_cyan": "\033[36;1m",
+    "bright_white": "\033[37;1m",
+    "bg_black": "\033[40m",
+    "bg_red": "\033[41m",
+    "bg_green": "\033[42m",
+    "bg_yellow": "\033[43m",
+    "bg_blue": "\033[44m",
+    "bg_magenta": "\033[45m",
+    "bg_cyan": "\033[46m",
+    "bg_white": "\033[47m",
+    "bg_bright_black": "\033[40;1m",
+    "bg_bright_red": "\033[41;1m",
+    "bg_bright_green": "\033[42;1m",
+    "bg_bright_yellow": "\033[43;1m",
+    "bg_bright_blue": "\033[44;1m",
+    "bg_bright_magenta": "\033[45;1m",
+    "bg_bright_cyan": "\033[46;1m",
+    "bg_bright_white": "\033[47;1m",
+}
+
+
+def apply_format(text, *formats):
+    """Apply ANSI escape sequences for the specified formats to the given text."""
+    formatted_text = text
+    for format in formats:
+        formatted_text = f"{ANSI_FORMATS.get(format, '')}{formatted_text}{ANSI_FORMATS.get('reset', '')}"
+    return formatted_text
+
+
+def apply_color(text, color=None, background=None):
+    """Apply ANSI escape sequences for the specified color and background to the given text."""
+    formatted_text = text
+    if color:
+        formatted_text = f"{ANSI_COLORS.get(color, '')}{formatted_text}{ANSI_FORMATS.get('reset', '')}"
+    if background:
+        formatted_text = f"{ANSI_COLORS.get(background, '')}{formatted_text}{ANSI_FORMATS.get('reset', '')}"
+    return formatted_text
+
+
+def print_formatted(text, *formats, color=None, background=None, **kwargs):
+    """Print the given text with the specified formats, color, and background."""
+    formatted_text = apply_format(text, *formats)
+    formatted_text = apply_color(formatted_text, color, background)
+    file = kwargs.get("file", sys.stdout)
+    header = "[mtb install] "
+
+    # Handle console encoding for Unicode characters (utf-8)
+    encoded_header = header.encode(sys.stdout.encoding, errors="replace").decode(
+        sys.stdout.encoding
+    )
+    encoded_text = formatted_text.encode(sys.stdout.encoding, errors="replace").decode(
+        sys.stdout.encoding
+    )
+
+    print(
+        " " * len(encoded_header)
+        if kwargs.get("no_header")
+        else apply_color(apply_format(encoded_header, "bold"), color="yellow"),
+        encoded_text,
+        file=file,
+    )
+
+
+# endregion
+
+
+# region utils
+def run_command(cmd, ignored_lines_start=None):
+    if ignored_lines_start is None:
+        ignored_lines_start = []
+
+    if isinstance(cmd, str):
+        shell_cmd = cmd
+    elif isinstance(cmd, list):
+        shell_cmd = " ".join(
+            arg.as_posix() if isinstance(arg, Path) else shlex.quote(str(arg))
+            for arg in cmd
+        )
+    else:
+        raise ValueError(
+            "Invalid 'cmd' argument. It must be a string or a list of arguments."
+        )
+
+    try:
+        _run_command(shell_cmd, ignored_lines_start)
+    except subprocess.CalledProcessError as e:
+        print(f"Command failed with return code: {e.returncode}", file=sys.stderr)
+        print(e.stderr.strip(), file=sys.stderr)
+
+    except KeyboardInterrupt:
+        print("Command execution interrupted.")
+
+
+def _run_command(shell_cmd, ignored_lines_start):
+    print_formatted(f"Running {shell_cmd}", "bold")
+    result = subprocess.run(
+        shell_cmd,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
+        text=True,
+        shell=True,
+        check=True,
+    )
+
+    stdout_lines = result.stdout.strip().split("\n")
+    stderr_lines = result.stderr.strip().split("\n")
+
+    # Print stdout, skipping ignored lines
+    for line in stdout_lines:
+        if not any(line.startswith(ign) for ign in ignored_lines_start):
+            print(line)
+
+    # Print stderr
+    for line in stderr_lines:
+        print(line, file=sys.stderr)
+
+    print("Command executed successfully!")
+
+
+def is_pipe():
+    if not sys.stdin.isatty():
+        return False
+    if sys.platform == "win32":
+        try:
+            import msvcrt
+
+            return msvcrt.get_osfhandle(0) != -1
+        except ImportError:
+            return False
+    else:
+        try:
+            mode = os.fstat(0).st_mode
+            return (
+                stat.S_ISFIFO(mode)
+                or stat.S_ISREG(mode)
+                or stat.S_ISBLK(mode)
+                or stat.S_ISSOCK(mode)
+            )
+        except OSError:
+            return False
+
+
+@contextmanager
+def suppress_std():
+    with open(os.devnull, "w") as devnull:
+        old_stdout = sys.stdout
+        old_stderr = sys.stderr
+        sys.stdout = devnull
+        sys.stderr = devnull
+
+        try:
+            yield
+        finally:
+            sys.stdout = old_stdout
+            sys.stderr = old_stderr
+
+
+# Get the version from __init__.py
+def get_local_version():
+    init_file = os.path.join(os.path.dirname(__file__), "__init__.py")
+    if os.path.isfile(init_file):
+        with open(init_file, "r") as f:
+            tree = ast.parse(f.read())
+            for node in ast.walk(tree):
+                if isinstance(node, ast.Assign):
+                    for target in node.targets:
+                        if (
+                            isinstance(target, ast.Name)
+                            and target.id == "__version__"
+                            and isinstance(node.value, ast.Str)
+                        ):
+                            return node.value.s
+    return None
+
+
+def download_file(url, file_name):
+    with requests.get(url, stream=True) as response:
+        response.raise_for_status()
+        total_size = int(response.headers.get("content-length", 0))
+        with open(file_name, "wb") as file, tqdm(
+            desc=file_name.stem,
+            total=total_size,
+            unit="B",
+            unit_scale=True,
+            unit_divisor=1024,
+        ) as progress_bar:
+            for chunk in response.iter_content(chunk_size=8192):
+                file.write(chunk)
+                progress_bar.update(len(chunk))
+
+
+def try_import(requirement):
+    dependency = requirement.name.strip()
+    import_name = pip_map.get(dependency, dependency)
+    installed = False
+
+    pip_name = dependency
+    pip_spec = "".join(specs[0]) if (specs := requirement.specs) else ""
+    try:
+        with suppress_std():
+            import_module(import_name)
+        print_formatted(
+            f"\t✅ Package {pip_name} already installed (import name: '{import_name}').",
+            "bold",
+            color="green",
+            no_header=True,
+        )
+        installed = True
+    except ImportError:
+        print_formatted(
+            f"\t⛔ Package {pip_name} is missing (import name: '{import_name}').",
+            "bold",
+            color="red",
+            no_header=True,
+        )
+
+    return (installed, pip_name, pip_spec, import_name)
+
+
+def import_or_install(requirement, dry=False):
+    installed, pip_name, pip_spec, import_name = try_import(requirement)
+
+    pip_install_name = pip_name + pip_spec
+
+    if not installed:
+        print_formatted(f"Installing package {pip_name}...", "italic", color="yellow")
+        if dry:
+            print_formatted(
+                f"Dry-run: Package {pip_install_name} would be installed (import name: '{import_name}').",
+                color="yellow",
+            )
+        else:
+            try:
+                run_command([executable, "-m", "pip", "install", pip_install_name])
+                print_formatted(
+                    f"Package {pip_install_name} installed successfully using pip package name  (import name: '{import_name}')",
+                    "bold",
+                    color="green",
+                )
+            except subprocess.CalledProcessError as e:
+                print_formatted(
+                    f"Failed to install package {pip_install_name} using pip package name  (import name: '{import_name}'). Error: {str(e)}",
+                    "bold",
+                    color="red",
+                )
+
+
+def get_github_assets(tag=None):
+    if tag:
+        tag_url = (
+            f"https://api.github.com/repos/{repo_owner}/{repo_name}/releases/tags/{tag}"
+        )
+    else:
+        tag_url = (
+            f"https://api.github.com/repos/{repo_owner}/{repo_name}/releases/latest"
+        )
+    response = requests.get(tag_url)
+    if response.status_code == 404:
+        # print_formatted(
+        #     f"Tag version '{apply_color(version,'cyan')}' not found for {owner}/{repo} repository."
+        # )
+        print_formatted("Error retrieving the release assets.", color="red")
+        sys.exit()
+
+    tag_data = response.json()
+    tag_name = tag_data["name"]
+
+    return tag_data, tag_name
+
+
+# endregion
+
+
+try:
+    from tqdm import tqdm
+except ImportError:
+    print_formatted("Installing tqdm...", "italic", color="yellow")
+    run_command([executable, "-m", "pip", "install", "--upgrade", "tqdm"])
+    from tqdm import tqdm
+
+
+def main():
+    if len(sys.argv) == 1:
+        print_formatted(
+            "mtb doesn't need an install script anymore.", "italic", color="yellow"
+        )
+        return
+    if all(arg not in ("-p", "--path") for arg in sys.argv):
+        print(
+            "This script is only used for and edge case of remote installs on some cloud providers, unrecognized arguments:",
+            sys.argv[1:],
+        )
+        return
+
+    # Parse command-line arguments
+    parser = argparse.ArgumentParser(description="Comfy_mtb install script")
+    parser.add_argument(
+        "--path",
+        "-p",
+        type=str,
+        help="Path to clone the repository to (i.e the absolute path to ComfyUI/custom_nodes)",
+    )
+
+    print_formatted("mtb install", "bold", color="yellow")
+
+    args = parser.parse_args()
+
+    print_formatted(f"Detected environment: {apply_color(mode,'cyan')}")
+
+    if args.path:
+        clone_dir = Path(args.path)
+        if not clone_dir.exists():
+            print_formatted(
+                "The path provided does not exist on disk... It must be pointing to ComfyUI's custom_nodes directory"
+            )
+            sys.exit()
+
+        else:
+            repo_dir = clone_dir / repo_name
+            if not repo_dir.exists():
+                print_formatted(f"Cloning to {repo_dir}...", "italic", color="yellow")
+                run_command(["git", "clone", "--recursive", repo_url, repo_dir])
+            else:
+                print_formatted(
+                    f"Directory {repo_dir} already exists, we will update it..."
+                )
+                run_command(["git", "pull", "-C", repo_dir])
+        here = clone_dir
+        full = True
+
+    print_formatted("Checking environment...", "italic", color="yellow")
+    missing_deps = []
+    install_cmd = [executable, "-m", "pip", "install", "-r", "requirements.txt"]
+    run_command(install_cmd)
+
+    print_formatted(
+        "✅ Successfully installed all dependencies.", "italic", color="green"
+    )
+
+
+if __name__ == "__main__":
+    main()

log.py

@@ -1,9 +1,20 @@
 import logging
+import os
 import re
 
+base_log_level = logging.DEBUG if os.environ.get("MTB_DEBUG") else logging.INFO
+
+
+# Custom object that discards the output
+class NullWriter:
+    def write(self, text):
+        pass
+
 
 class Formatter(logging.Formatter):
     grey = "\x1b[38;20m"
+    cyan = "\x1b[36;20m"
+    purple = "\x1b[35;20m"
     yellow = "\x1b[33;20m"
     red = "\x1b[31;20m"
     bold_red = "\x1b[31;1m"
@@ -12,8 +23,8 @@ class Formatter(logging.Formatter):
     format = "[%(name)s] | %(levelname)s -> %(message)s"
 
     FORMATS = {
-        logging.DEBUG: grey + format + reset,
-        logging.INFO: grey + format + reset,
+        logging.DEBUG: purple + format + reset,
+        logging.INFO: cyan + format + reset,
         logging.WARNING: yellow + format + reset,
         logging.ERROR: red + format + reset,
         logging.CRITICAL: bold_red + format + reset,
@@ -25,21 +36,26 @@ class Formatter(logging.Formatter):
         return formatter.format(record)
 
 
-def mklog(name, level=logging.DEBUG):
+def mklog(name, level=base_log_level):
     logger = logging.getLogger(name)
     logger.setLevel(level)
-    # create console handler with a higher log level
+
+    for handler in logger.handlers:
+        logger.removeHandler(handler)
+
     ch = logging.StreamHandler()
-    ch.setLevel(logging.DEBUG)
-
+    ch.setLevel(level)
     ch.setFormatter(Formatter())
-
     logger.addHandler(ch)
+
+    # Disable log propagation
+    logger.propagate = False
+
     return logger
 
 
 # - The main app logger
-log = mklog(__package__)
+log = mklog(__package__, base_log_level)
 
 
 def log_user(arg):
@@ -54,6 +70,12 @@ def blue_text(text):
     return f"\033[94m{text}\033[0m"
 
 
+def cyan_text(text):
+    return f"\033[96m{text}\033[0m"
+
+
 def get_label(label):
+    if label.startswith("MTB_"):
+        label = label[4:]
     words = re.findall(r"(?:^|[A-Z])[a-z]*", label)
     return " ".join(words).strip()

node_list.json

@@ -0,0 +1,61 @@
+{
+    "Animation Builder (mtb)": "Convenient way to manage basic animation maths at the core of many of my workflows",
+    "Any To String (mtb)": "Tries to take any input and convert it to a string",
+    "Batch Float (mtb)": "Generates a batch of float values with interpolation",
+    "Batch Float Assemble (mtb)": "Assembles mutiple batches of floats into a single stream (batch)",
+    "Batch Float Fill (mtb)": "Fills a batch float with a single value until it reaches the target length",
+    "Batch Make (mtb)": "Simply duplicates the input frame as a batch",
+    "Batch Merge (mtb)": "Merges multiple image batches with different frame counts",
+    "Batch Shake (mtb)": "Applies a shaking effect to batches of images.",
+    "Batch Shape (mtb)": "Generates a batch of 2D shapes with optional shading (experimental)",
+    "Batch Transform (mtb)": "Transform a batch of images using a batch of keyframes",
+    "Bbox (mtb)": "The bounding box (BBOX) custom type used by other nodes",
+    "Bbox From Mask (mtb)": "From a mask extract the bounding box",
+    "Blur (mtb)": "Blur an image using a Gaussian filter.",
+    "Color Correct (mtb)": "Various color correction methods",
+    "Colored Image (mtb)": "Constant color image of given size",
+    "Concat Images (mtb)": "Add images to batch",
+    "Crop (mtb)": "Crops an image and an optional mask to a given bounding box\n\n    The bounding box can be given as a tuple of (x, y, width, height) or as a BBOX type\n    The BBOX input takes precedence over the tuple input\n    ",
+    "Debug (mtb)": "Experimental node to debug any Comfy values, support for more types and widgets is planned",
+    "Deep Bump (mtb)": "Normal & height maps generation from single pictures",
+    "Export With Ffmpeg (mtb)": "Export with FFmpeg (Experimental)",
+    "Face Swap (mtb)": "Face swap using deepinsight/insightface models",
+    "Film Interpolation (mtb)": "Google Research FILM frame interpolation for large motion",
+    "Fit Number (mtb)": "Fit the input float using a source and target range",
+    "Float To Number (mtb)": "Node addon for the WAS Suite. Converts a \"comfy\" FLOAT to a NUMBER.",
+    "Get Batch From History (mtb)": "Very experimental node to load images from the history of the server.\n\n    Queue items without output are ignored in the count.",
+    "Image Compare (mtb)": "Compare two images and return a difference image",
+    "Image Premultiply (mtb)": "Premultiply image with mask",
+    "Image Remove Background Rembg (mtb)": "Removes the background from the input using Rembg.",
+    "Image Resize Factor (mtb)": "Extracted mostly from WAS Node Suite, with a few edits (most notably multiple image support) and less features.",
+    "Image Tile Offset (mtb)": "Mimics an old photoshop technique to check for seamless textures",
+    "Int To Bool (mtb)": "Basic int to bool conversion",
+    "Int To Number (mtb)": "Node addon for the WAS Suite. Converts a \"comfy\" INT to a NUMBER.",
+    "Interpolate Clip Sequential (mtb)": null,
+    "Latent Lerp (mtb)": "Linear interpolation (blend) between two latent vectors",
+    "Load Face Analysis Model (mtb)": "Loads a face analysis model",
+    "Load Face Enhance Model (mtb)": "Loads a GFPGan or RestoreFormer model for face enhancement.",
+    "Load Face Swap Model (mtb)": "Loads a faceswap model",
+    "Load Film Model (mtb)": "Loads a FILM model",
+    "Load Image From Url (mtb)": "Load an image from the given URL",
+    "Load Image Sequence (mtb)": "Load an image sequence from a folder. The current frame is used to determine which image to load.\n\n    Usually used in conjunction with the `Primitive` node set to increment to load a sequence of images from a folder.\n    Use -1 to load all matching frames as a batch.\n    ",
+    "Mask To Image (mtb)": "Converts a mask (alpha) to an RGB image with a color and background",
+    "Math Expression (mtb)": "Node to evaluate a simple math expression string",
+    "Model Patch Seamless (mtb)": "Uses the stable diffusion 'hack' to infer seamless images by setting the model layers padding mode to circular (experimental)",
+    "Qr Code (mtb)": "Basic QR Code generator",
+    "Restore Face (mtb)": "Uses GFPGan to restore faces",
+    "Save Gif (mtb)": "Save the images from the batch as a GIF",
+    "Save Image Grid (mtb)": "Save all the images in the input batch as a grid of images.",
+    "Save Image Sequence (mtb)": "Save an image sequence to a folder. The current frame is used to determine which image to save.\n\n    This is merely a wrapper around the `save_images` function with formatting for the output folder and filename.\n    ",
+    "Save Tensors (mtb)": "Save torch tensors (image, mask or latent) to disk, useful to debug things outside comfy",
+    "Sharpen (mtb)": "Sharpens an image using a Gaussian kernel.",
+    "Smart Step (mtb)": "Utils to control the steps start/stop of the KAdvancedSampler in percentage",
+    "Stack Images (mtb)": "Stack the input images horizontally or vertically",
+    "String Replace (mtb)": "Basic string replacement",
+    "Styles Loader (mtb)": "Load csv files and populate a dropdown from the rows (\u00e0 la A111)",
+    "Text To Image (mtb)": "Utils to convert text to image using a font\n\n\n    The tool looks for any .ttf file in the Comfy folder hierarchy.\n    ",
+    "Transform Image (mtb)": "Save torch tensors (image, mask or latent) to disk, useful to debug things outside comfy\n\n\n    it return a tensor representing the transformed images with the same shape as the input tensor\n    ",
+    "Uncrop (mtb)": "Uncrops an image to a given bounding box\n\n    The bounding box can be given as a tuple of (x, y, width, height) or as a BBOX type\n    The BBOX input takes precedence over the tuple input",
+    "Unsplash Image (mtb)": "Unsplash Image given a keyword and a size",
+    "Vae Decode (mtb)": "Wrapper for the 2 core decoders but also adding the sd seamless hack, taken from: FlyingFireCo/tiled_ksampler"
+}

nodes/animation.py

@@ -0,0 +1,44 @@
+from ..log import log
+
+
+class AnimationBuilder:
+    """Convenient way to manage basic animation maths at the core of many of my workflows"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "total_frames": ("INT", {"default": 100, "min": 0}),
+                # "fps": ("INT", {"default": 12, "min": 0}),
+                "scale_float": ("FLOAT", {"default": 1.0, "min": 0.0}),
+                "loop_count": ("INT", {"default": 1, "min": 0}),
+                "raw_iteration": ("INT", {"default": 0, "min": 0}),
+                "raw_loop": ("INT", {"default": 0, "min": 0}),
+            },
+        }
+
+    RETURN_TYPES = ("INT", "FLOAT", "INT", "BOOLEAN")
+    RETURN_NAMES = ("frame", "0-1 (scaled)", "count", "loop_ended")
+    CATEGORY = "mtb/animation"
+    FUNCTION = "build_animation"
+
+    def build_animation(
+        self,
+        total_frames=100,
+        # fps=12,
+        scale_float=1.0,
+        loop_count=1,  # set in js
+        raw_iteration=0,  # set in js
+        raw_loop=0,  # set in js
+    ):
+        frame = raw_iteration % (total_frames)
+        scaled = (frame / (total_frames - 1)) * scale_float
+        # if frame == 0:
+        #     log.debug("Reseting history")
+        #     PromptServer.instance.prompt_queue.wipe_history()
+        log.debug(f"frame: {frame}/{total_frames}  scaled: {scaled}")
+
+        return (frame, scaled, raw_loop, (frame == (total_frames - 1)))
+
+
+__nodes__ = [AnimationBuilder]

nodes/batch.py

@@ -0,0 +1,676 @@
+from io import BytesIO
+
+import cv2
+import torchaudio
+import numpy as np
+import torch
+from PIL import Image
+
+from ..log import log
+from ..utils import apply_easing, pil2tensor
+from .transform import TransformImage
+
+try:
+    import librosa
+except ImportError:
+    log.warning("librosa not installed. Batch Audio features will not be available.")
+
+
+def hex_to_rgb(hex_color, bgr=False):
+    hex_color = hex_color.lstrip("#")
+    if bgr:
+        return tuple(int(hex_color[i : i + 2], 16) for i in (4, 2, 0))
+
+    return tuple(int(hex_color[i : i + 2], 16) for i in (0, 2, 4))
+
+
+class BatchMake:
+    """Simply duplicates the input frame as a batch"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "count": ("INT", {"default": 1}),
+            }
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "generate_batch"
+    CATEGORY = "mtb/batch"
+
+    def generate_batch(self, image: torch.Tensor, count):
+        if len(image.shape) == 3:
+            image = image.unsqueeze(0)
+
+        return (image.repeat(count, 1, 1, 1),)
+
+
+class BatchShape:
+    """Generates a batch of 2D shapes with optional shading (experimental)"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "count": ("INT", {"default": 1}),
+                "shape": (
+                    ["Box", "Circle", "Diamond"],
+                    {"default": "Box"},
+                ),
+                "image_width": ("INT", {"default": 512}),
+                "image_height": ("INT", {"default": 512}),
+                "shape_size": ("INT", {"default": 100}),
+                "color": ("COLOR", {"default": "#ffffff"}),
+                "bg_color": ("COLOR", {"default": "#000000"}),
+                "shade_color": ("COLOR", {"default": "#000000"}),
+                "shadex": ("FLOAT", {"default": 0.0}),
+                "shadey": ("FLOAT", {"default": 0.0}),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "generate_shapes"
+    CATEGORY = "mtb/batch"
+
+    def generate_shapes(
+        self,
+        count,
+        shape,
+        image_width,
+        image_height,
+        shape_size,
+        color,
+        bg_color,
+        shade_color,
+        shadex,
+        shadey,
+    ):
+        print(f"COLOR: {color}")
+        print(f"BG_COLOR: {bg_color}")
+        print(f"SHADE_COLOR: {shade_color}")
+
+        # Parse color input to BGR tuple for OpenCV
+        color = hex_to_rgb(color)
+        bg_color = hex_to_rgb(bg_color)
+        shade_color = hex_to_rgb(shade_color)
+        res = []
+        for x in range(count):
+            # Initialize an image canvas
+            canvas = np.full((image_height, image_width, 3), bg_color, dtype=np.uint8)
+            mask = np.zeros((image_height, image_width), dtype=np.uint8)
+
+            # Compute the center point of the shape
+            center = (image_width // 2, image_height // 2)
+
+            if shape == "Box":
+                half_size = shape_size // 2
+                top_left = (center[0] - half_size, center[1] - half_size)
+                bottom_right = (center[0] + half_size, center[1] + half_size)
+                cv2.rectangle(mask, top_left, bottom_right, 255, -1)
+            elif shape == "Circle":
+                cv2.circle(mask, center, shape_size // 2, 255, -1)
+            elif shape == "Diamond":
+                pts = np.array(
+                    [
+                        [center[0], center[1] - shape_size // 2],
+                        [center[0] + shape_size // 2, center[1]],
+                        [center[0], center[1] + shape_size // 2],
+                        [center[0] - shape_size // 2, center[1]],
+                    ]
+                )
+                cv2.fillPoly(mask, [pts], 255)
+
+            # Color the shape
+            canvas[mask == 255] = color
+
+            # Apply shading effects to a separate shading canvas
+            shading = np.zeros_like(canvas, dtype=np.float32)
+            shading[:, :, 0] = shadex * np.linspace(0, 1, image_width)
+            shading[:, :, 1] = shadey * np.linspace(0, 1, image_height).reshape(-1, 1)
+            shading_canvas = cv2.addWeighted(
+                canvas.astype(np.float32), 1, shading, 1, 0
+            ).astype(np.uint8)
+
+            # Apply shading only to the shape area using the mask
+            canvas[mask == 255] = shading_canvas[mask == 255]
+            res.append(canvas)
+
+        return (pil2tensor(res),)
+
+
+class BatchFloatFill:
+    """Fills a batch float with a single value until it reaches the target length"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "floats": ("FLOATS",),
+                "direction": (["head", "tail"], {"default": "tail"}),
+                "value": ("FLOAT", {"default": 0.0}),
+                "count": ("INT", {"default": 1}),
+            }
+        }
+
+    FUNCTION = "fill_floats"
+    RETURN_TYPES = ("FLOATS",)
+    CATEGORY = "mtb/batch"
+
+    def fill_floats(self, floats, direction, value, count):
+        size = len(floats)
+        if size > count:
+            raise ValueError(f"Size ({size}) is less then target count ({count})")
+
+        rem = count - size
+        if direction == "tail":
+            floats = floats + [value] * rem
+        else:
+            floats = [value] * rem + floats
+        return (floats,)
+
+
+class BatchFloatAssemble:
+    """Assembles mutiple batches of floats into a single stream (batch)"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {"required": {"reverse": ("BOOLEAN", {"default": False})}}
+
+    FUNCTION = "assemble_floats"
+    RETURN_TYPES = ("FLOATS",)
+    CATEGORY = "mtb/batch"
+
+    def assemble_floats(self, reverse, **kwargs):
+        res = []
+        if reverse:
+            for x in reversed(kwargs.values()):
+                res += x
+        else:
+            for x in kwargs.values():
+                res += x
+
+        return (res,)
+
+
+class BatchFloat:
+    """Generates a batch of float values with interpolation"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "mode": (
+                    ["Single", "Steps"],
+                    {"default": "Steps"},
+                ),
+                "count": ("INT", {"default": 1}),
+                "min": ("FLOAT", {"default": 0.0}),
+                "max": ("FLOAT", {"default": 1.0}),
+                "easing": (
+                    [
+                        "Linear",
+                        "Sine In",
+                        "Sine Out",
+                        "Sine In/Out",
+                        "Quart In",
+                        "Quart Out",
+                        "Quart In/Out",
+                        "Cubic In",
+                        "Cubic Out",
+                        "Cubic In/Out",
+                        "Circ In",
+                        "Circ Out",
+                        "Circ In/Out",
+                        "Back In",
+                        "Back Out",
+                        "Back In/Out",
+                        "Elastic In",
+                        "Elastic Out",
+                        "Elastic In/Out",
+                        "Bounce In",
+                        "Bounce Out",
+                        "Bounce In/Out",
+                    ],
+                    {"default": "Linear"},
+                ),
+            }
+        }
+
+    FUNCTION = "set_floats"
+    RETURN_TYPES = ("FLOATS",)
+    CATEGORY = "mtb/batch"
+
+    def set_floats(self, mode, count, min, max, easing):
+        keyframes = []
+        if mode == "Single":
+            keyframes = [min] * count
+            return (keyframes,)
+
+        for i in range(count):
+            normalized_step = i / (count - 1)
+            eased_step = apply_easing(normalized_step, easing)
+            eased_value = min + (max - min) * eased_step
+            keyframes.append(eased_value)
+
+        return (keyframes,)
+
+
+class BatchMerge:
+    """Merges multiple image batches with different frame counts"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "fusion_mode": (["add", "multiply", "average"], {"default": "average"}),
+                "fill": (["head", "tail"], {"default": "tail"}),
+            }
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "merge_batches"
+    CATEGORY = "mtb/batch"
+
+    def merge_batches(self, fusion_mode, fill, **kwargs):
+        images = kwargs.values()
+        max_frames = max(img.shape[0] for img in images)
+
+        adjusted_images = []
+        for img in images:
+            frame_count = img.shape[0]
+            if frame_count < max_frames:
+                fill_frame = img[0] if fill == "head" else img[-1]
+                fill_frames = fill_frame.repeat(max_frames - frame_count, 1, 1, 1)
+                adjusted_batch = (
+                    torch.cat((fill_frames, img), dim=0)
+                    if fill == "head"
+                    else torch.cat((img, fill_frames), dim=0)
+                )
+            else:
+                adjusted_batch = img
+            adjusted_images.append(adjusted_batch)
+
+        # Merge the adjusted batches
+        merged_image = None
+        for img in adjusted_images:
+            if merged_image is None:
+                merged_image = img
+            else:
+                if fusion_mode == "add":
+                    merged_image += img
+                elif fusion_mode == "multiply":
+                    merged_image *= img
+                elif fusion_mode == "average":
+                    merged_image = (merged_image + img) / 2
+
+        return (merged_image,)
+
+
+class Batch2dTransform:
+    """Transform a batch of images using a batch of keyframes"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "border_handling": (
+                    ["edge", "constant", "reflect", "symmetric"],
+                    {"default": "edge"},
+                ),
+                "constant_color": ("COLOR", {"default": "#000000"}),
+            },
+            "optional": {
+                "x": ("FLOATS",),
+                "y": ("FLOATS",),
+                "zoom": ("FLOATS",),
+                "angle": ("FLOATS",),
+                "shear": ("FLOATS",),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "transform_batch"
+    CATEGORY = "mtb/batch"
+
+    def get_num_elements(self, param) -> int:
+        if isinstance(param, torch.Tensor):
+            return torch.numel(param)
+        elif isinstance(param, list):
+            return len(param)
+
+        return 0
+
+    def transform_batch(
+        self,
+        image: torch.Tensor,
+        border_handling,
+        constant_color,
+        x=None,
+        y=None,
+        zoom=None,
+        angle=None,
+        shear=None,
+    ):
+        if all(
+            self.get_num_elements(param) <= 0 for param in [x, y, zoom, angle, shear]
+        ):
+            raise ValueError("At least one transform parameter must be provided")
+
+        keyframes = {"x": [], "y": [], "zoom": [], "angle": [], "shear": []}
+
+        default_vals = {"x": 0, "y": 0, "zoom": 1.0, "angle": 0, "shear": 0}
+
+        if self.get_num_elements(x) > 0:
+            keyframes["x"] = x
+        if self.get_num_elements(y) > 0:
+            keyframes["y"] = y
+        if self.get_num_elements(zoom) > 0:
+            keyframes["zoom"] = zoom
+        if self.get_num_elements(angle) > 0:
+            keyframes["angle"] = angle
+        if self.get_num_elements(shear) > 0:
+            keyframes["shear"] = shear
+
+        for name, values in keyframes.items():
+            count = len(values)
+            if count > 0 and count != image.shape[0]:
+                raise ValueError(
+                    f"Length of {name} values ({count}) must match number of images ({image.shape[0]})"
+                )
+            if count == 0:
+                keyframes[name] = [default_vals[name]] * image.shape[0]
+
+        transformer = TransformImage()
+        res = [
+            transformer.transform(
+                image[i].unsqueeze(0),
+                keyframes["x"][i],
+                keyframes["y"][i],
+                keyframes["zoom"][i],
+                keyframes["angle"][i],
+                keyframes["shear"][i],
+                border_handling,
+                constant_color,
+            )[0]
+            for i in range(image.shape[0])
+        ]
+        return (torch.cat(res, dim=0),)
+
+
+DEFAULT_INTERPOLANT = lambda t: t * t * t * (t * (t * 6 - 15) + 10)
+
+
+class BatchShake:
+    """Applies a shaking effect to batches of images."""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "images": ("IMAGE",),
+                "position_amount_x": ("FLOAT", {"default": 1.0}),
+                "position_amount_y": ("FLOAT", {"default": 1.0}),
+                "rotation_amount": ("FLOAT", {"default": 10.0}),
+                "frequency": ("FLOAT", {"default": 1.0, "min": 0.005}),
+                "frequency_divider": ("FLOAT", {"default": 1.0, "min": 0.005}),
+                "octaves": ("INT", {"default": 1, "min": 1}),
+                "seed": ("INT", {"default": 0}),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE", "FLOATS", "FLOATS", "FLOATS")
+    RETURN_NAMES = ("image", "pos_x", "pos_y", "rot")
+    FUNCTION = "apply_shake"
+    CATEGORY = "mtb/batch"
+
+    # def interpolant(self, t):
+    # return t * t * t * (t * (t * 6 - 15) + 10)
+
+    def generate_perlin_noise_2d(
+        self, shape, res, tileable=(False, False), interpolant=None
+    ):
+        """Generate a 2D numpy array of perlin noise.
+
+        Args:
+            shape: The shape of the generated array (tuple of two ints).
+                This must be a multple of res.
+            res: The number of periods of noise to generate along each
+                axis (tuple of two ints). Note shape must be a multiple of
+                res.
+            tileable: If the noise should be tileable along each axis
+                (tuple of two bools). Defaults to (False, False).
+            interpolant: The interpolation function, defaults to
+                t*t*t*(t*(t*6 - 15) + 10).
+
+        Returns:
+            A numpy array of shape shape with the generated noise.
+
+        Raises:
+            ValueError: If shape is not a multiple of res.
+        """
+        interpolant = interpolant or DEFAULT_INTERPOLANT
+        delta = (res[0] / shape[0], res[1] / shape[1])
+        d = (shape[0] // res[0], shape[1] // res[1])
+        grid = (
+            np.mgrid[0 : res[0] : delta[0], 0 : res[1] : delta[1]].transpose(1, 2, 0)
+            % 1
+        )
+        # Gradients
+        angles = 2 * np.pi * np.random.rand(res[0] + 1, res[1] + 1)
+        gradients = np.dstack((np.cos(angles), np.sin(angles)))
+        if tileable[0]:
+            gradients[-1, :] = gradients[0, :]
+        if tileable[1]:
+            gradients[:, -1] = gradients[:, 0]
+        gradients = gradients.repeat(d[0], 0).repeat(d[1], 1)
+        g00 = gradients[: -d[0], : -d[1]]
+        g10 = gradients[d[0] :, : -d[1]]
+        g01 = gradients[: -d[0], d[1] :]
+        g11 = gradients[d[0] :, d[1] :]
+        # Ramps
+        n00 = np.sum(np.dstack((grid[:, :, 0], grid[:, :, 1])) * g00, 2)
+        n10 = np.sum(np.dstack((grid[:, :, 0] - 1, grid[:, :, 1])) * g10, 2)
+        n01 = np.sum(np.dstack((grid[:, :, 0], grid[:, :, 1] - 1)) * g01, 2)
+        n11 = np.sum(np.dstack((grid[:, :, 0] - 1, grid[:, :, 1] - 1)) * g11, 2)
+        # Interpolation
+        t = interpolant(grid)
+        n0 = n00 * (1 - t[:, :, 0]) + t[:, :, 0] * n10
+        n1 = n01 * (1 - t[:, :, 0]) + t[:, :, 0] * n11
+        return np.sqrt(2) * ((1 - t[:, :, 1]) * n0 + t[:, :, 1] * n1)
+
+    def generate_fractal_noise_2d(
+        self,
+        shape,
+        res,
+        octaves=1,
+        persistence=0.5,
+        lacunarity=2,
+        tileable=(True, True),
+        interpolant=None,
+    ):
+        """Generate a 2D numpy array of fractal noise.
+
+        Args:
+            shape: The shape of the generated array (tuple of two ints).
+                This must be a multiple of lacunarity**(octaves-1)*res.
+            res: The number of periods of noise to generate along each
+                axis (tuple of two ints). Note shape must be a multiple of
+                (lacunarity**(octaves-1)*res).
+            octaves: The number of octaves in the noise. Defaults to 1.
+            persistence: The scaling factor between two octaves.
+            lacunarity: The frequency factor between two octaves.
+            tileable: If the noise should be tileable along each axis
+                (tuple of two bools). Defaults to (True,True).
+            interpolant: The, interpolation function, defaults to
+                t*t*t*(t*(t*6 - 15) + 10).
+
+        Returns:
+            A numpy array of fractal noise and of shape shape generated by
+            combining several octaves of perlin noise.
+
+        Raises:
+            ValueError: If shape is not a multiple of
+                (lacunarity**(octaves-1)*res).
+        """
+        interpolant = interpolant or DEFAULT_INTERPOLANT
+
+        noise = np.zeros(shape)
+        frequency = 1
+        amplitude = 1
+        for _ in range(octaves):
+            noise += amplitude * self.generate_perlin_noise_2d(
+                shape, (frequency * res[0], frequency * res[1]), tileable, interpolant
+            )
+            frequency *= lacunarity
+            amplitude *= persistence
+        return noise
+
+    def fbm(self, x, y, octaves):
+        # noise_2d = self.generate_fractal_noise_2d((256, 256), (8, 8), octaves)
+        # Now, extract a single noise value based on x and y, wrapping indices if necessary
+        x_idx = int(x) % 256
+        y_idx = int(y) % 256
+        return self.noise_pattern[x_idx, y_idx]
+
+    def apply_shake(
+        self,
+        images,
+        position_amount_x,
+        position_amount_y,
+        rotation_amount,
+        frequency,
+        frequency_divider,
+        octaves,
+        seed,
+    ):
+        # Rehash
+        np.random.seed(seed)
+        self.position_offset = np.random.uniform(-1e3, 1e3, 3)
+        self.rotation_offset = np.random.uniform(-1e3, 1e3, 3)
+        self.noise_pattern = self.generate_perlin_noise_2d(
+            (512, 512), (32, 32), (True, True)
+        )
+
+        # Assuming frame count is derived from the first dimension of images tensor
+        frame_count = images.shape[0]
+
+        frequency = frequency / frequency_divider
+
+        # Generate shaking parameters for each frame
+        x_translations = []
+        y_translations = []
+        rotations = []
+
+        for frame_num in range(frame_count):
+            time = frame_num * frequency
+            x_idx = (self.position_offset[0] + frame_num) % 256
+            y_idx = (self.position_offset[1] + frame_num) % 256
+
+            np_position = np.array(
+                [
+                    self.fbm(x_idx, time, octaves),
+                    self.fbm(y_idx, time, octaves),
+                ]
+            )
+
+            # np_position = np.array(
+            #     [
+            #         self.fbm(self.position_offset[0] + frame_num, time, octaves),
+            #         self.fbm(self.position_offset[1] + frame_num, time, octaves),
+            #     ]
+            # )
+            # np_rotation = self.fbm(self.rotation_offset[2] + frame_num, time, octaves)
+
+            rot_idx = (self.rotation_offset[2] + frame_num) % 256
+            np_rotation = self.fbm(rot_idx, time, octaves)
+
+            x_translations.append(np_position[0] * position_amount_x)
+            y_translations.append(np_position[1] * position_amount_y)
+            rotations.append(np_rotation * rotation_amount)
+
+        # Convert lists to tensors
+        # x_translations = torch.tensor(x_translations, dtype=torch.float32)
+        # y_translations = torch.tensor(y_translations, dtype=torch.float32)
+        # rotations = torch.tensor(rotations, dtype=torch.float32)
+
+        # Create an instance of Batch2dTransform
+        transform = Batch2dTransform()
+
+        log.debug(
+            f"Applying shaking with parameters: \nposition {position_amount_x}, {position_amount_y}\nrotation {rotation_amount}\nfrequency {frequency}\noctaves {octaves}"
+        )
+
+        # Apply shaking transformations to images
+        shaken_images = transform.transform_batch(
+            images,
+            border_handling="edge",  # Assuming edge handling as default
+            constant_color="#000000",  # Assuming black as default constant color
+            x=x_translations,
+            y=y_translations,
+            angle=rotations,
+        )[0]
+
+        return (shaken_images, x_translations, y_translations, rotations)
+
+
+class BatchFloatsFromSound:
+    """Extracts a list of floats based on audio frequency band peaks."""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "audio": ("AUDIO",),
+                "sensitivity": ("FLOAT", {"default": 1.0}),
+                "low_freq": ("FLOAT", {"default": 100.0}),
+                "high_freq": ("FLOAT", {"default": 2000.0}),
+                "hop_length": ("INT", {"default": 512}),
+            },
+        }
+
+    RETURN_TYPES = ("FLOATS",)
+    RETURN_NAMES = ("float_data",)
+    FUNCTION = "process_audio"
+    CATEGORY = "mtb/audio"
+
+    def process_audio(
+        self,
+        audio,
+        sensitivity=1.0,
+        low_freq=100,
+        high_freq=2000,
+        hop_length=512,
+    ):
+        # audio_data, _ = librosa.load(audio_file_path, sr=sample_rate)
+
+        # audio_data_tensor = audio.squeeze(1)  # Remove the channel dimension if present
+        # audio_tensor = audio_data_tensor.float()
+        audio_data = audio.to(device=torchaudio.transforms.Spectrogram().window.device)
+
+        hop_length = 512
+        stft = torchaudio.transforms.Spectrogram()(audio_data)
+        freqs = torchaudio.transforms.FrequencyMasking(low_freq, high_freq)(stft)
+        band_energy = torch.sum(freqs, dim=1)
+
+        min_val = torch.min(band_energy)
+        max_val = torch.max(band_energy)
+        normalized_peaks = (band_energy - min_val) / (max_val - min_val)
+        scaled_peaks = normalized_peaks * sensitivity
+
+        return (scaled_peaks.tolist(),)
+
+
+__nodes__ = [
+    BatchFloat,
+    Batch2dTransform,
+    BatchFloatsFromSound,
+    BatchShape,
+    BatchMake,
+    BatchFloatAssemble,
+    BatchFloatFill,
+    BatchMerge,
+    BatchShake,
+]

nodes/conditions.py

@@ -1,18 +1,98 @@
-from ..utils import pil2tensor
-from ..utils import here
-from ..log import log
-import folder_paths
-from pathlib import Path
-import shutil
 import csv
+import shutil
+from pathlib import Path
+
+import folder_paths
+
+from ..log import log
+from ..utils import here
+
+
+class InterpolateClipSequential:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "base_text": ("STRING", {"multiline": True}),
+                "text_to_replace": ("STRING", {"default": ""}),
+                "clip": ("CLIP",),
+                "interpolation_strength": (
+                    "FLOAT",
+                    {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01},
+                ),
+            }
+        }
+
+    RETURN_TYPES = ("CONDITIONING",)
+    FUNCTION = "interpolate_encodings_sequential"
+
+    CATEGORY = "mtb/conditioning"
+
+    def interpolate_encodings_sequential(
+        self, base_text, text_to_replace, clip, interpolation_strength, **replacements
+    ):
+        log.debug(f"Received interpolation_strength: {interpolation_strength}")
+
+        # - Ensure interpolation strength is within [0, 1]
+        interpolation_strength = max(0.0, min(1.0, interpolation_strength))
+
+        # - Check if replacements were provided
+        if not replacements:
+            raise ValueError("At least one replacement should be provided.")
+
+        num_replacements = len(replacements)
+        log.debug(f"Number of replacements: {num_replacements}")
+
+        segment_length = 1.0 / num_replacements
+        log.debug(f"Calculated segment_length: {segment_length}")
+
+        # - Find the segment that the interpolation_strength falls into
+        segment_index = min(
+            int(interpolation_strength // segment_length), num_replacements - 1
+        )
+        log.debug(f"Segment index: {segment_index}")
+
+        # - Calculate the local strength within the segment
+        local_strength = (
+            interpolation_strength - (segment_index * segment_length)
+        ) / segment_length
+        log.debug(f"Local strength: {local_strength}")
+
+        # - If it's the first segment, interpolate between base_text and the first replacement
+        if segment_index == 0:
+            replacement_text = list(replacements.values())[0]
+            log.debug("Using the base text a the base blend")
+            # -  Start with the base_text condition
+            tokens = clip.tokenize(base_text)
+            cond_from, pooled_from = clip.encode_from_tokens(tokens, return_pooled=True)
+        else:
+            base_replace = list(replacements.values())[segment_index - 1]
+            log.debug(f"Using {base_replace} a the base blend")
+
+            # - Start with the base_text condition replaced by the closest replacement
+            tokens = clip.tokenize(base_text.replace(text_to_replace, base_replace))
+            cond_from, pooled_from = clip.encode_from_tokens(tokens, return_pooled=True)
+
+            replacement_text = list(replacements.values())[segment_index]
+
+        interpolated_text = base_text.replace(text_to_replace, replacement_text)
+        tokens = clip.tokenize(interpolated_text)
+        cond_to, pooled_to = clip.encode_from_tokens(tokens, return_pooled=True)
+
+        # - Linearly interpolate between the two conditions
+        interpolated_condition = (
+            1.0 - local_strength
+        ) * cond_from + local_strength * cond_to
+        interpolated_pooled = (
+            1.0 - local_strength
+        ) * pooled_from + local_strength * pooled_to
+
+        return ([[interpolated_condition, {"pooled_output": interpolated_pooled}]],)
 
 
 class SmartStep:
     """Utils to control the steps start/stop of the KAdvancedSampler in percentage"""
 
-    def __init__(self):
-        pass
-
     @classmethod
     def INPUT_TYPES(cls):
         return {
@@ -35,7 +115,7 @@ class SmartStep:
     RETURN_TYPES = ("INT", "INT", "INT")
     RETURN_NAMES = ("step", "start", "end")
     FUNCTION = "do_step"
-    CATEGORY = "conditioning"
+    CATEGORY = "mtb/conditioning"
 
     def do_step(self, step, start_percent, end_percent):
         start = int(step * start_percent / 100)
@@ -62,37 +142,41 @@ class StylesLoader:
 
     options = {}
 
-    def __init__(self):
-        pass
-
     @classmethod
     def INPUT_TYPES(cls):
-        input_dir = Path(folder_paths.base_path) / "styles"
-        if not input_dir.exists():
-            install_default_styles()
+        if not cls.options:
+            input_dir = Path(folder_paths.base_path) / "styles"
+            if not input_dir.exists():
+                install_default_styles()
+
+            if not (files := [f for f in input_dir.iterdir() if f.suffix == ".csv"]):
+                log.warn(
+                    "No styles found in the styles folder, place at least one csv file in the styles folder at the root of ComfyUI (for instance ComfyUI/styles/mystyle.csv)"
+                )
+
+            for file in files:
+                with open(file, "r", encoding="utf8") as f:
+                    parsed = csv.reader(f)
+                    for row in parsed:
+                        log.debug(f"Adding style {row[0]}")
+                        try:
+                            cls.options[row[0]] = (row[1], row[2])
+                        except Exception:
+                            log.warning(
+                                f"There was an error while parsing {file}, make sure it respects A1111 format, i.e 3 columns name, positive, negative"
+                            )
+                            continue
+
+        else:
+            log.debug(f"Using cached styles (count: {len(cls.options)})")
 
-        if not (files := [f for f in input_dir.iterdir() if f.suffix == ".csv"]):
-            log.error(
-                "No styles found in the styles folder, place at least one csv file in the styles folder"
-            )
-            return {
-                "required": {
-                    "style_name": (["error"],),
-                }
-            }
-        for file in files:
-            with open(file, "r", encoding="utf8") as f:
-                parsed = csv.reader(f)
-                for row in parsed:
-                    log.debug(f"Adding style {row[0]}")
-                    cls.options[row[0]] = (row[1], row[2])
         return {
             "required": {
                 "style_name": (list(cls.options.keys()),),
             }
         }
 
-    CATEGORY = "conditioning"
+    CATEGORY = "mtb/conditioning"
 
     RETURN_TYPES = ("STRING", "STRING")
     RETURN_NAMES = ("positive", "negative")
@@ -102,100 +186,4 @@ class StylesLoader:
         return (self.options[style_name][0], self.options[style_name][1])
 
 
-class TextToImage:
-    """Utils to convert text to image using a font
-
-
-    The tool looks for any .ttf file in the Comfy folder hierarchy.
-    """
-
-    fonts = {}
-
-    def __init__(self):
-        pass
-
-    @classmethod
-    def INPUT_TYPES(cls):
-        fonts = list(Path(folder_paths.base_path).glob("**/*.ttf"))
-        if not fonts:
-            log.error(
-                "No fonts found in the fonts folder, place at least one ttf file in the fonts folder"
-            )
-            return {
-                "required": {
-                    "font": (["error"],),
-                }
-            }
-        for font in fonts:
-            log.debug(f"Adding font {font}")
-            cls.fonts[font.stem] = font.as_posix()
-
-        return {
-            "required": {
-                "text": (
-                    "STRING",
-                    {"default": "Hello world!"},
-                ),
-                "font": ((sorted(cls.fonts.keys())),),
-                "wrap": (
-                    "INT",
-                    {"default": 120, "min": 0, "max": 8096, "step": 1},
-                ),
-                "font_size": (
-                    "INT",
-                    {"default": 12, "min": 1, "max": 100, "step": 1},
-                ),
-                "width": (
-                    "INT",
-                    {"default": 512, "min": 1, "max": 1000, "step": 1},
-                ),
-                "height": (
-                    "INT",
-                    {"default": 512, "min": 1, "max": 8096, "step": 1},
-                ),
-                # "position": (["INT"], {"default": 0, "min": 0, "max": 100, "step": 1}),
-                "color": (
-                    "COLOR",
-                    {"default": "black"},
-                ),
-                "background": (
-                    "COLOR",
-                    {"default": "white"},
-                ),
-            }
-        }
-
-    RETURN_TYPES = ("IMAGE",)
-    RETURN_NAMES = ("image",)
-    FUNCTION = "text_to_image"
-    CATEGORY = "utils"
-
-    def text_to_image(
-        self, text, font, wrap, font_size, width, height, color, background
-    ):
-        from PIL import Image, ImageDraw, ImageFont
-        import textwrap
-
-        font = self.fonts[font]
-        font = ImageFont.truetype(font, font_size)
-        if wrap == 0:
-            wrap = width / font_size
-        lines = textwrap.wrap(text, width=wrap)
-        log.debug(f"Lines: {lines}")
-        line_height = font.getsize("hg")[1]
-        img_height = height  # line_height * len(lines)
-        img_width = width  # max(font.getsize(line)[0] for line in lines)
-
-        img = Image.new("RGBA", (img_width, img_height), background)
-        draw = ImageDraw.Draw(img)
-        y_text = 0
-        for line in lines:
-            width, height = font.getsize(line)
-            draw.text((0, y_text), line, color, font=font)
-            y_text += height
-
-        # img.save(os.path.join(folder_paths.base_path, f'{str(uuid.uuid4())}.png'))
-        return (pil2tensor(img),)
-
-
-__nodes__ = [SmartStep, TextToImage, StylesLoader]
+__nodes__ = [SmartStep, StylesLoader, InterpolateClipSequential]

nodes/crop.py

@@ -1,18 +1,19 @@
-import torch
-from ..utils import tensor2pil, pil2tensor
-from PIL import Image, ImageFilter, ImageDraw
 import numpy as np
+import torch
+from PIL import Image, ImageChops, ImageDraw, ImageFilter
+
+from ..log import log
+from ..utils import np2tensor, pil2tensor, tensor2np, tensor2pil
 
 
-class BoundingBox:
+class Bbox:
     """The bounding box (BBOX) custom type used by other nodes"""
-    def __init__(self):
-        pass
 
     @classmethod
     def INPUT_TYPES(cls):
         return {
             "required": {
+                # "bbox": ("BBOX",),
                 "x": ("INT", {"default": 0, "max": 10000000, "min": 0, "step": 1}),
                 "y": ("INT", {"default": 0, "max": 10000000, "min": 0, "step": 1}),
                 "width": (
@@ -28,16 +29,15 @@ class BoundingBox:
 
     RETURN_TYPES = ("BBOX",)
     FUNCTION = "do_crop"
-    CATEGORY = "image/crop"
+    CATEGORY = "mtb/crop"
 
-    def do_crop(self, x, y, width, height):
-        return (x, y, width, height)
+    def do_crop(self, x, y, width, height):  # bbox
+        return ((x, y, width, height),)
+        # return bbox
 
 
-class BBoxFromMask:
+class BboxFromMask:
     """From a mask extract the bounding box"""
-    def __init__(self):
-        pass
 
     @classmethod
     def INPUT_TYPES(cls):
@@ -59,13 +59,25 @@ class BBoxFromMask:
         "image (optional)",
     )
     FUNCTION = "extract_bounding_box"
-    CATEGORY = "image/crop"
+    CATEGORY = "mtb/crop"
 
     def extract_bounding_box(self, mask: torch.Tensor, image=None):
+        # if image != None:
+        #     if mask.size(0) != image.size(0):
+        #         if mask.size(0) != 1:
+        #             log.error(
+        #                 f"Batch count mismatch for mask and image, it can either be 1 mask for X images, or X masks for X images (mask: {mask.shape} | image: {image.shape})"
+        #             )
 
-        mask = tensor2pil(mask)
+        #             raise Exception(
+        #                 f"Batch count mismatch for mask and image, it can either be 1 mask for X images, or X masks for X images (mask: {mask.shape} | image: {image.shape})"
+        #             )
+
+        _mask = tensor2pil(1.0 - mask)[0]
+
+        # we invert it
+        alpha_channel = np.array(_mask)
 
-        alpha_channel = np.array(mask)
         non_zero_indices = np.nonzero(alpha_channel)
 
         min_x, max_x = np.min(non_zero_indices[1]), np.max(non_zero_indices[1])
@@ -74,11 +86,16 @@ class BBoxFromMask:
         # Create a bounding box tuple
         if image != None:
             # Convert the image to a NumPy array
-            image = image.numpy()
-            # Crop the image from the bounding box
-            image = image[:, min_y:max_y, min_x:max_x]
-            image = torch.from_numpy(image)
+            imgs = tensor2np(image)
+            out = []
+            for img in imgs:
+                # Crop the image from the bounding box
+                img = img[min_y:max_y, min_x:max_x, :]
+                log.debug(f"Cropped image to shape {img.shape}")
+                out.append(img)
 
+            image = np2tensor(out)
+            log.debug(f"Cropped images shape: {image.shape}")
         bounding_box = (min_x, min_y, max_x - min_x, max_y - min_y)
         return (
             bounding_box,
@@ -92,8 +109,6 @@ class Crop:
     The bounding box can be given as a tuple of (x, y, width, height) or as a BBOX type
     The BBOX input takes precedence over the tuple input
     """
-    def __init__(self):
-        pass
 
     @classmethod
     def INPUT_TYPES(cls):
@@ -120,12 +135,11 @@ class Crop:
     RETURN_TYPES = ("IMAGE", "MASK", "BBOX")
     FUNCTION = "do_crop"
 
-    CATEGORY = "image/crop"
+    CATEGORY = "mtb/crop"
 
     def do_crop(
         self, image: torch.Tensor, mask=None, x=0, y=0, width=256, height=256, bbox=None
     ):
-
         image = image.numpy()
         if mask:
             mask = mask.numpy()
@@ -144,13 +158,43 @@ class Crop:
         )
 
 
+# def calculate_intersection(rect1, rect2):
+#     x_left = max(rect1[0], rect2[0])
+#     y_top = max(rect1[1], rect2[1])
+#     x_right = min(rect1[2], rect2[2])
+#     y_bottom = min(rect1[3], rect2[3])
+
+#     return (x_left, y_top, x_right, y_bottom)
+
+
+def bbox_check(bbox, target_size=None):
+    if not target_size:
+        return bbox
+
+    new_bbox = (
+        bbox[0],
+        bbox[1],
+        min(target_size[0] - bbox[0], bbox[2]),
+        min(target_size[1] - bbox[1], bbox[3]),
+    )
+    if new_bbox != bbox:
+        log.warn(f"BBox too big, constrained to {new_bbox}")
+
+    return new_bbox
+
+
+def bbox_to_region(bbox, target_size=None):
+    bbox = bbox_check(bbox, target_size)
+
+    # to region
+    return (bbox[0], bbox[1], bbox[0] + bbox[2], bbox[1] + bbox[3])
+
+
 class Uncrop:
     """Uncrops an image to a given bounding box
 
     The bounding box can be given as a tuple of (x, y, width, height) or as a BBOX type
     The BBOX input takes precedence over the tuple input"""
-    def __init__(self):
-        pass
 
     @classmethod
     def INPUT_TYPES(cls):
@@ -169,7 +213,7 @@ class Uncrop:
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "do_crop"
 
-    CATEGORY = "image/crop"
+    CATEGORY = "mtb/crop"
 
     def do_crop(self, image, crop_image, bbox, border_blending):
         def inset_border(image, border_width=20, border_color=(0)):
@@ -182,41 +226,63 @@ class Uncrop:
             )
             return bordered_image
 
-        image = tensor2pil(image)
-        crop_img = tensor2pil(crop_image)
-        crop_img = crop_img.convert("RGB")
+        single = image.size(0) == 1
+        if image.size(0) != crop_image.size(0):
+            if not single:
+                raise ValueError(
+                    "The Image batch count is greater than 1, but doesn't match the crop_image batch count. If using batches they should either match or only crop_image must be greater than 1"
+                )
 
-        # uncrop the image based on the bounding box
-        bb_x, bb_y, bb_width, bb_height = bbox
+        images = tensor2pil(image)
+        crop_imgs = tensor2pil(crop_image)
+        out_images = []
+        for i, crop in enumerate(crop_imgs):
+            if single:
+                img = images[0]
+            else:
+                img = images[i]
 
-        if border_blending > 1.0:
-            border_blending = 1.0
-        elif border_blending < 0.0:
-            border_blending = 0.0
+            # uncrop the image based on the bounding box
+            bb_x, bb_y, bb_width, bb_height = bbox
 
-        blend_ratio = (max(crop_img.size) / 2) * float(border_blending)
+            paste_region = bbox_to_region((bb_x, bb_y, bb_width, bb_height), img.size)
+            # log.debug(f"Paste region: {paste_region}")
+            # new_region = adjust_paste_region(img.size, paste_region)
+            # log.debug(f"Adjusted paste region: {new_region}")
+            # # Check if the adjusted paste region is different from the original
 
-        blend = image.convert("RGBA")
-        mask = Image.new("L", image.size, 0)
+            crop_img = crop.convert("RGB")
 
-        mask_block = Image.new("L", (bb_width, bb_height), 255)
-        mask_block = inset_border(mask_block, int(blend_ratio / 2), (0))
+            log.debug(f"Crop image size: {crop_img.size}")
+            log.debug(f"Image size: {img.size}")
 
-        mask.paste(mask_block, (bb_x, bb_y, bb_x + bb_width, bb_y + bb_height))
-        blend.paste(crop_img, (bb_x, bb_y, bb_x + bb_width, bb_y + bb_height))
+            if border_blending > 1.0:
+                border_blending = 1.0
+            elif border_blending < 0.0:
+                border_blending = 0.0
 
-        mask = mask.filter(ImageFilter.BoxBlur(radius=blend_ratio / 4))
-        mask = mask.filter(ImageFilter.GaussianBlur(radius=blend_ratio / 4))
+            blend_ratio = (max(crop_img.size) / 2) * float(border_blending)
 
-        blend.putalpha(mask)
-        image = Image.alpha_composite(image.convert("RGBA"), blend)
+            blend = img.convert("RGBA")
+            mask = Image.new("L", img.size, 0)
 
-        return (pil2tensor(image.convert("RGB")),)
+            mask_block = Image.new("L", (bb_width, bb_height), 255)
+            mask_block = inset_border(mask_block, int(blend_ratio / 2), (0))
+
+            mask.paste(mask_block, paste_region)
+            log.debug(f"Blend size: {blend.size} | kind {blend.mode}")
+            log.debug(f"Crop image size: {crop_img.size} | kind {crop_img.mode}")
+            log.debug(f"BBox: {paste_region}")
+            blend.paste(crop_img, paste_region)
+
+            mask = mask.filter(ImageFilter.BoxBlur(radius=blend_ratio / 4))
+            mask = mask.filter(ImageFilter.GaussianBlur(radius=blend_ratio / 4))
+
+            blend.putalpha(mask)
+            img = Image.alpha_composite(img.convert("RGBA"), blend)
+            out_images.append(img.convert("RGB"))
+
+        return (pil2tensor(out_images),)
 
 
-__nodes__ = [
-    BBoxFromMask,
-    BoundingBox,
-    Crop,
-    Uncrop
-]
+__nodes__ = [BboxFromMask, Bbox, Crop, Uncrop]

nodes/debug.py

@@ -0,0 +1,174 @@
+import base64
+import io
+from pathlib import Path
+from typing import Optional
+
+import folder_paths
+import torch
+
+from ..log import log
+from ..utils import tensor2pil
+
+
+# region processors
+def process_tensor(tensor):
+    log.debug(f"Tensor: {tensor.shape}")
+
+    image = tensor2pil(tensor)
+    b64_imgs = []
+    for im in image:
+        buffered = io.BytesIO()
+        im.save(buffered, format="PNG")
+        b64_imgs.append(
+            "data:image/png;base64,"
+            + base64.b64encode(buffered.getvalue()).decode("utf-8")
+        )
+
+    return {"b64_images": b64_imgs}
+
+
+def process_list(anything):
+    text = []
+    if not anything:
+        return {"text": []}
+
+    first_element = anything[0]
+    if (
+        isinstance(first_element, list)
+        and first_element
+        and isinstance(first_element[0], torch.Tensor)
+    ):
+        text.append(
+            f"List of List of Tensors: {first_element[0].shape} (x{len(anything)})"
+        )
+
+    elif isinstance(first_element, torch.Tensor):
+        text.append(f"List of Tensors: {first_element.shape} (x{len(anything)})")
+
+    return {"text": text}
+
+
+def process_dict(anything):
+    text = []
+    if "samples" in anything:
+        is_empty = "(empty)" if torch.count_nonzero(anything["samples"]) == 0 else ""
+        text.append(f"Latent Samples: {anything['samples'].shape} {is_empty}")
+
+    return {"text": text}
+
+
+def process_bool(anything):
+    return {"text": ["True" if anything else "False"]}
+
+
+def process_text(anything):
+    return {"text": [str(anything)]}
+
+
+# endregion
+
+
+class Debug:
+    """Experimental node to debug any Comfy values, support for more types and widgets is planned"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {"output_to_console": ("BOOLEAN", {"default": False})},
+        }
+
+    RETURN_TYPES = ()
+    FUNCTION = "do_debug"
+    CATEGORY = "mtb/debug"
+    OUTPUT_NODE = True
+
+    def do_debug(self, output_to_console, **kwargs):
+        output = {
+            "ui": {"b64_images": [], "text": []},
+            # "result": ("A"),
+        }
+
+        processors = {
+            torch.Tensor: process_tensor,
+            list: process_list,
+            dict: process_dict,
+            bool: process_bool,
+        }
+        if output_to_console:
+            print("bouh!")
+
+        for anything in kwargs.values():
+            processor = processors.get(type(anything), process_text)
+            processed_data = processor(anything)
+
+            for ui_key, ui_value in processed_data.items():
+                output["ui"][ui_key].extend(ui_value)
+            # log.debug(
+            #     f"Processed input {k}, found {len(processed_data.get('b64_images', []))} images and {len(processed_data.get('text', []))} text items."
+            # )
+
+        return output
+
+
+class SaveTensors:
+    """Save torch tensors (image, mask or latent) to disk, useful to debug things outside comfy"""
+
+    def __init__(self):
+        self.output_dir = folder_paths.get_output_directory()
+        self.type = "mtb/debug"
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "filename_prefix": ("STRING", {"default": "ComfyPickle"}),
+            },
+            "optional": {
+                "image": ("IMAGE",),
+                "mask": ("MASK",),
+                "latent": ("LATENT",),
+            },
+        }
+
+    FUNCTION = "save"
+    OUTPUT_NODE = True
+    RETURN_TYPES = ()
+    CATEGORY = "mtb/debug"
+
+    def save(
+        self,
+        filename_prefix,
+        image: Optional[torch.Tensor] = None,
+        mask: Optional[torch.Tensor] = None,
+        latent: Optional[torch.Tensor] = None,
+    ):
+        (
+            full_output_folder,
+            filename,
+            counter,
+            subfolder,
+            filename_prefix,
+        ) = folder_paths.get_save_image_path(filename_prefix, self.output_dir)
+        full_output_folder = Path(full_output_folder)
+        if image is not None:
+            image_file = f"{filename}_image_{counter:05}.pt"
+            torch.save(image, full_output_folder / image_file)
+            # np.save(full_output_folder/ image_file, image.cpu().numpy())
+
+        if mask is not None:
+            mask_file = f"{filename}_mask_{counter:05}.pt"
+            torch.save(mask, full_output_folder / mask_file)
+            # np.save(full_output_folder/ mask_file, mask.cpu().numpy())
+
+        if latent is not None:
+            # for latent we must use pickle
+            latent_file = f"{filename}_latent_{counter:05}.pt"
+            torch.save(latent, full_output_folder / latent_file)
+            # pickle.dump(latent, open(full_output_folder/ latent_file, "wb"))
+
+            # np.save(full_output_folder/ latent_file, latent[""].cpu().numpy())
+
+        return f"{filename_prefix}_{counter:05}"
+
+
+__nodes__ = [Debug, SaveTensors]

nodes/deep_bump.py

@@ -1,23 +1,38 @@
-import onnxruntime as ort
+import tempfile
+from pathlib import Path
+
 import numpy as np
-import pathlib
 import onnxruntime as ort
-import numpy as np
-from .. import utils as utils_inference
-from ..log import log
+import torch
+from PIL import Image
+
+from ..errors import ModelNotFound
+from ..log import mklog
+from ..utils import get_model_path, tensor2pil, tiles_infer, tiles_merge, tiles_split
 
 # Disable MS telemetry
 ort.disable_telemetry_events()
+log = mklog(__name__)
 
 
 # - COLOR to NORMALS
-def color_to_normals(color_img, overlap, progress_callback):
+def color_to_normals(color_img, overlap, progress_callback, save_temp=False):
     """Computes a normal map from the given color map. 'color_img' must be a numpy array
     in C,H,W format (with C as RGB). 'overlap' must be one of 'SMALL', 'MEDIUM', 'LARGE'.
     """
+    temp_dir = Path(tempfile.mkdtemp()) if save_temp else None
 
     # Remove alpha & convert to grayscale
-    img = np.mean(color_img[:3], axis=0, keepdimss=True)
+    img = np.mean(color_img[:3], axis=0, keepdims=True)
+
+    if temp_dir:
+        Image.fromarray((img[0] * 255).astype(np.uint8)).save(
+            temp_dir / "grayscale_img.png"
+        )
+
+    log.debug(
+        f"Converting color image to grayscale by taking the mean over color channels: {img.shape}"
+    )
 
     # Split image in tiles
     log.debug("DeepBump Color → Normals : tilling")
@@ -28,32 +43,56 @@ def color_to_normals(color_img, overlap, progress_callback):
         "LARGE": tile_size // 2,
     }
     stride_size = tile_size - overlaps[overlap]
-    tiles, paddings = utils_inference.tiles_split(
+    tiles, paddings = tiles_split(
         img, (tile_size, tile_size), (stride_size, stride_size)
     )
+    if temp_dir:
+        for i, tile in enumerate(tiles):
+            Image.fromarray((tile[0] * 255).astype(np.uint8)).save(
+                temp_dir / f"tile_{i}.png"
+            )
 
     # Load model
     log.debug("DeepBump Color → Normals : loading model")
-    addon_path = str(pathlib.Path(__file__).parent.absolute())
-    ort_session = ort.InferenceSession(f"{addon_path}/models/deepbump256.onnx")
+    model = get_model_path("deepbump", "deepbump256.onnx")
+    if not model or not model.exists():
+        raise ModelNotFound(f"deepbump ({model})")
+
+    ort_session = ort.InferenceSession(model)
 
     # Predict normal map for each tile
     log.debug("DeepBump Color → Normals : generating")
-    pred_tiles = utils_inference.tiles_infer(
-        tiles, ort_session, progress_callback=progress_callback
-    )
+    pred_tiles = tiles_infer(tiles, ort_session, progress_callback=progress_callback)
+
+    if temp_dir:
+        for i, pred_tile in enumerate(pred_tiles):
+            Image.fromarray((pred_tile.transpose(1, 2, 0) * 255).astype(np.uint8)).save(
+                temp_dir / f"pred_tile_{i}.png"
+            )
 
     # Merge tiles
     log.debug("DeepBump Color → Normals : merging")
-    pred_img = utils_inference.tiles_merge(
+    pred_img = tiles_merge(
         pred_tiles,
         (stride_size, stride_size),
         (3, img.shape[1], img.shape[2]),
         paddings,
     )
 
+    if temp_dir:
+        Image.fromarray((pred_img.transpose(1, 2, 0) * 255).astype(np.uint8)).save(
+            temp_dir / "merged_img.png"
+        )
+
     # Normalize each pixel to unit vector
-    pred_img = utils_inference.normalize(pred_img)
+    pred_img = normalize(pred_img)
+
+    if temp_dir:
+        Image.fromarray((pred_img.transpose(1, 2, 0) * 255).astype(np.uint8)).save(
+            temp_dir / "final_img.png"
+        )
+
+        log.debug(f"Debug images saved in {temp_dir}")
 
     return pred_img
 
@@ -241,9 +280,6 @@ def normals_to_height(normals_img, seamless, progress_callback):
 class DeepBump:
     """Normal & height maps generation from single pictures"""
 
-    def __init__(self):
-        pass
-
     @classmethod
     def INPUT_TYPES(cls):
         return {
@@ -264,14 +300,14 @@ class DeepBump:
                         "LARGEST",
                     ],
                 ),
-                "normals_to_height_seamless": (["TRUE", "FALSE"],),
+                "normals_to_height_seamless": ("BOOLEAN", {"default": True}),
             },
         }
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "apply"
 
-    CATEGORY = "image processing"
+    CATEGORY = "mtb/textures"
 
     def apply(
         self,
@@ -279,29 +315,40 @@ class DeepBump:
         mode="Color to Normals",
         color_to_normals_overlap="SMALL",
         normals_to_curvature_blur_radius="SMALL",
-        normals_to_height_seamless="TRUE",
+        normals_to_height_seamless=True,
     ):
-        image = utils_inference.tensor2pil(image)
+        images = tensor2pil(image)
+        out_images = []
 
-        in_img = np.transpose(image, (2, 0, 1)) / 255
+        for image in images:
+            log.debug(f"Input image shape: {image}")
 
-        log.debug(f"Input image shape: {in_img.shape}")
+            in_img = np.transpose(image, (2, 0, 1)) / 255
+            log.debug(f"transposed for deep image shape: {in_img.shape}")
+            out_img = None
 
-        # Apply processing
-        if mode == "Color to Normals":
-            out_img = color_to_normals(in_img, color_to_normals_overlap, None)
-        if mode == "Normals to Curvature":
-            out_img = normals_to_curvature(
-                in_img, normals_to_curvature_blur_radius, None
-            )
-        if mode == "Normals to Height":
-            out_img = normals_to_height(
-                in_img, normals_to_height_seamless == "TRUE", None
-            )
+            # Apply processing
+            if mode == "Color to Normals":
+                out_img = color_to_normals(in_img, color_to_normals_overlap, None)
+            if mode == "Normals to Curvature":
+                out_img = normals_to_curvature(
+                    in_img, normals_to_curvature_blur_radius, None
+                )
+            if mode == "Normals to Height":
+                out_img = normals_to_height(in_img, normals_to_height_seamless, None)
 
-        out_img = (np.transpose(out_img, (1, 2, 0)) * 255).astype(np.uint8)
-
-        return (utils_inference.pil2tensor(out_img),)
+            if out_img is not None:
+                log.debug(f"Output image shape: {out_img.shape}")
+                out_images.append(
+                    torch.from_numpy(
+                        np.transpose(out_img, (1, 2, 0)).astype(np.float32)
+                    ).unsqueeze(0)
+                )
+            else:
+                log.error("No out img... This should not happen")
+        for outi in out_images:
+            log.debug(f"Shape fed to utils: {outi.shape}")
+        return (torch.cat(out_images, dim=0),)
 
 
 __nodes__ = [DeepBump]

nodes/faceenhance.py

@@ -0,0 +1,261 @@
+import os
+from pathlib import Path
+from typing import Tuple
+
+import comfy
+import comfy.utils
+import cv2
+import folder_paths
+import numpy as np
+import torch
+from comfy import model_management
+from gfpgan import GFPGANer
+from PIL import Image
+
+from ..log import NullWriter, log
+from ..utils import get_model_path, np2tensor, pil2tensor, tensor2np
+
+
+class LoadFaceEnhanceModel:
+    """Loads a GFPGan or RestoreFormer model for face enhancement."""
+
+    def __init__(self) -> None:
+        pass
+
+    @classmethod
+    def get_models_root(cls):
+        fr = get_model_path("face_restore")
+        # fr = Path(folder_paths.models_dir) / "face_restore"
+        if fr.exists():
+            return (fr, None)
+
+        um = get_model_path("upscale_models")
+        return (fr, um) if um.exists() else (None, None)
+
+    @classmethod
+    def get_models(cls):
+        fr_models_path, um_models_path = cls.get_models_root()
+
+        if fr_models_path is None and um_models_path is None:
+            log.warning("Face restoration models not found.")
+            return []
+        if not fr_models_path.exists():
+            log.warning(
+                f"No Face Restore checkpoints found at {fr_models_path} (if you've used mtb before these checkpoints were saved in upscale_models before)"
+            )
+            log.warning(
+                "For now we fallback to upscale_models but this will be removed in a future version"
+            )
+            if um_models_path.exists():
+                return [
+                    x
+                    for x in um_models_path.iterdir()
+                    if x.name.endswith(".pth")
+                    and ("GFPGAN" in x.name or "RestoreFormer" in x.name)
+                ]
+            return []
+
+        return [
+            x
+            for x in fr_models_path.iterdir()
+            if x.name.endswith(".pth")
+            and ("GFPGAN" in x.name or "RestoreFormer" in x.name)
+        ]
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "model_name": (
+                    [x.name for x in cls.get_models()],
+                    {"default": "None"},
+                ),
+                "upscale": ("INT", {"default": 1}),
+            },
+            "optional": {"bg_upsampler": ("UPSCALE_MODEL", {"default": None})},
+        }
+
+    RETURN_TYPES = ("FACEENHANCE_MODEL",)
+    RETURN_NAMES = ("model",)
+    FUNCTION = "load_model"
+    CATEGORY = "mtb/facetools"
+
+    def load_model(self, model_name, upscale=2, bg_upsampler=None):
+        basic = "RestoreFormer" not in model_name
+
+        fr_root, um_root = self.get_models_root()
+
+        if bg_upsampler is not None:
+            log.warning(
+                f"Upscale value overridden to {bg_upsampler.scale} from bg_upsampler"
+            )
+            upscale = bg_upsampler.scale
+            bg_upsampler = BGUpscaleWrapper(bg_upsampler)
+
+        sys.stdout = NullWriter()
+        model = GFPGANer(
+            model_path=(
+                (fr_root if fr_root.exists() else um_root) / model_name
+            ).as_posix(),
+            upscale=upscale,
+            arch="clean" if basic else "RestoreFormer",  # or original for v1.0 only
+            channel_multiplier=2,  # 1 for v1.0 only
+            bg_upsampler=bg_upsampler,
+        )
+
+        sys.stdout = sys.__stdout__
+        return (model,)
+
+
+class BGUpscaleWrapper:
+    def __init__(self, upscale_model) -> None:
+        self.upscale_model = upscale_model
+
+    def enhance(self, img: Image.Image, outscale=2):
+        device = model_management.get_torch_device()
+        self.upscale_model.to(device)
+
+        tile = 128 + 64
+        overlap = 8
+
+        imgt = np2tensor(img)
+        imgt = imgt.movedim(-1, -3).to(device)
+
+        steps = imgt.shape[0] * comfy.utils.get_tiled_scale_steps(
+            imgt.shape[3], imgt.shape[2], tile_x=tile, tile_y=tile, overlap=overlap
+        )
+
+        log.debug(f"Steps: {steps}")
+
+        pbar = comfy.utils.ProgressBar(steps)
+
+        s = comfy.utils.tiled_scale(
+            imgt,
+            lambda a: self.upscale_model(a),
+            tile_x=tile,
+            tile_y=tile,
+            overlap=overlap,
+            upscale_amount=self.upscale_model.scale,
+            pbar=pbar,
+        )
+
+        self.upscale_model.cpu()
+        s = torch.clamp(s.movedim(-3, -1), min=0, max=1.0)
+        return (tensor2np(s)[0],)
+
+
+import sys
+
+
+class RestoreFace:
+    """Uses GFPGan to restore faces"""
+
+    def __init__(self) -> None:
+        pass
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "restore"
+    CATEGORY = "mtb/facetools"
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "model": ("FACEENHANCE_MODEL",),
+                # Input are aligned faces
+                "aligned": ("BOOLEAN", {"default": False}),
+                # Only restore the center face
+                "only_center_face": ("BOOLEAN", {"default": False}),
+                # Adjustable weights
+                "weight": ("FLOAT", {"default": 0.5}),
+                "save_tmp_steps": ("BOOLEAN", {"default": True}),
+            }
+        }
+
+    def do_restore(
+        self,
+        image: torch.Tensor,
+        model: GFPGANer,
+        aligned,
+        only_center_face,
+        weight,
+        save_tmp_steps,
+    ) -> torch.Tensor:
+        pimage = tensor2np(image)[0]
+        width, height = pimage.shape[1], pimage.shape[0]
+        source_img = cv2.cvtColor(np.array(pimage), cv2.COLOR_RGB2BGR)
+
+        sys.stdout = NullWriter()
+        cropped_faces, restored_faces, restored_img = model.enhance(
+            source_img,
+            has_aligned=aligned,
+            only_center_face=only_center_face,
+            paste_back=True,
+            # TODO: weight has no effect in 1.3 and 1.4 (only tested these for now...)
+            weight=weight,
+        )
+        sys.stdout = sys.__stdout__
+        log.warning(f"Weight value has no effect for now. (value: {weight})")
+
+        if save_tmp_steps:
+            self.save_intermediate_images(cropped_faces, restored_faces, height, width)
+        output = None
+        if restored_img is not None:
+            output = Image.fromarray(cv2.cvtColor(restored_img, cv2.COLOR_BGR2RGB))
+            # imwrite(restored_img, save_restore_path)
+
+        return pil2tensor(output)
+
+    def restore(
+        self,
+        image: torch.Tensor,
+        model: GFPGANer,
+        aligned=False,
+        only_center_face=False,
+        weight=0.5,
+        save_tmp_steps=True,
+    ) -> Tuple[torch.Tensor]:
+        out = [
+            self.do_restore(
+                image[i], model, aligned, only_center_face, weight, save_tmp_steps
+            )
+            for i in range(image.size(0))
+        ]
+
+        return (torch.cat(out, dim=0),)
+
+    def get_step_image_path(self, step, idx):
+        (
+            full_output_folder,
+            filename,
+            counter,
+            _subfolder,
+            _filename_prefix,
+        ) = folder_paths.get_save_image_path(
+            f"{step}_{idx:03}",
+            folder_paths.temp_directory,
+        )
+        file = f"{filename}_{counter:05}_.png"
+
+        return os.path.join(full_output_folder, file)
+
+    def save_intermediate_images(self, cropped_faces, restored_faces, height, width):
+        for idx, (cropped_face, restored_face) in enumerate(
+            zip(cropped_faces, restored_faces)
+        ):
+            face_id = idx + 1
+            file = self.get_step_image_path("cropped_faces", face_id)
+            cv2.imwrite(file, cropped_face)
+
+            file = self.get_step_image_path("cropped_faces_restored", face_id)
+            cv2.imwrite(file, restored_face)
+
+            file = self.get_step_image_path("cropped_faces_compare", face_id)
+
+            # save comparison image
+            cmp_img = np.concatenate((cropped_face, restored_face), axis=1)
+            cv2.imwrite(file, cmp_img)
+
+
+__nodes__ = [RestoreFace, LoadFaceEnhanceModel]

nodes/faceswap.py

@@ -1,25 +1,96 @@
+# Optional face enhance nodes
 # region imports
-from ifnude import detect
+import sys
 from pathlib import Path
-from PIL import Image
-from typing import List, Set, Tuple
+from typing import List, Optional, Set, Union
+
+import comfy.model_management as model_management
 import cv2
-import folder_paths
-import glob
 import insightface
 import numpy as np
 import onnxruntime
-import os
-import tempfile
 import torch
+from insightface.model_zoo.inswapper import INSwapper
+from PIL import Image
 
-from ..utils import pil2tensor, tensor2pil
-from ..log import mklog
+from ..errors import ModelNotFound
+from ..log import NullWriter, mklog
+from ..utils import download_antelopev2, get_model_path, pil2tensor, tensor2pil
 
 # endregion
 
-logger = mklog(__name__)
-providers = onnxruntime.get_available_providers()
+log = mklog(__name__)
+
+
+class LoadFaceAnalysisModel:
+    """Loads a face analysis model"""
+
+    models = []
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "faceswap_model": (
+                    ["antelopev2", "buffalo_l", "buffalo_m", "buffalo_sc"],
+                    {"default": "buffalo_l"},
+                ),
+            },
+        }
+
+    RETURN_TYPES = ("FACE_ANALYSIS_MODEL",)
+    FUNCTION = "load_model"
+    CATEGORY = "mtb/facetools"
+
+    def load_model(self, faceswap_model: str):
+        if faceswap_model == "antelopev2":
+            download_antelopev2()
+
+        face_analyser = insightface.app.FaceAnalysis(
+            name=faceswap_model,
+            root=get_model_path("insightface"),
+        )
+        return (face_analyser,)
+
+
+class LoadFaceSwapModel:
+    """Loads a faceswap model"""
+
+    @staticmethod
+    def get_models() -> List[Path]:
+        models_path = get_model_path("insightface").iterdir()
+        return [x for x in models_path if x.suffix in [".onnx", ".pth"]]
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "faceswap_model": (
+                    [x.name for x in cls.get_models()],
+                    {"default": "None"},
+                ),
+            },
+        }
+
+    RETURN_TYPES = ("FACESWAP_MODEL",)
+    FUNCTION = "load_model"
+    CATEGORY = "mtb/facetools"
+
+    def load_model(self, faceswap_model: str):
+        model_path = get_model_path("insightface", faceswap_model)
+        if not model_path or not model_path.exists():
+            raise ModelNotFound(f"{faceswap_model} ({model_path})")
+
+        log.info(f"Loading model {model_path}")
+        return (
+            INSwapper(
+                model_path,
+                onnxruntime.InferenceSession(
+                    path_or_bytes=model_path,
+                    providers=onnxruntime.get_available_providers(),
+                ),
+            ),
+        )
 
 
 # region roop node
@@ -32,13 +103,6 @@ class FaceSwap:
     def __init__(self) -> None:
         pass
 
-    @staticmethod
-    def get_models() -> List[Path]:
-        models_path = os.path.join(folder_paths.models_dir, "insightface/*")
-        models = glob.glob(models_path)
-        models = [Path(x) for x in models if x.endswith(".onnx") or x.endswith(".pth")]
-        return models
-
     @classmethod
     def INPUT_TYPES(cls):
         return {
@@ -46,39 +110,39 @@ class FaceSwap:
                 "image": ("IMAGE",),
                 "reference": ("IMAGE",),
                 "faces_index": ("STRING", {"default": "0"}),
-                "faceswap_model": (
-                    [x.name for x in cls.get_models()],
-                    {"default": "None"},
-                ),
+                "faceanalysis_model": ("FACE_ANALYSIS_MODEL", {"default": "None"}),
+                "faceswap_model": ("FACESWAP_MODEL", {"default": "None"}),
             },
-            "optional": {"debug": (["true", "false"], {"default": "false"})},
+            "optional": {},
         }
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "swap"
-    CATEGORY = "face"
+    CATEGORY = "mtb/facetools"
 
     def swap(
         self,
         image: torch.Tensor,
         reference: torch.Tensor,
         faces_index: str,
-        faceswap_model: str,
-        debug: str,
+        faceanalysis_model,
+        faceswap_model,
     ):
         def do_swap(img):
-            img = tensor2pil(img)
-            ref = tensor2pil(reference)
+            model_management.throw_exception_if_processing_interrupted()
+            img = tensor2pil(img)[0]
+            ref = tensor2pil(reference)[0]
             face_ids = {
                 int(x) for x in faces_index.strip(",").split(",") if x.isnumeric()
             }
-            model = self.getFaceSwapModel(faceswap_model)
-            swapped = swap_face(ref, img, model, face_ids)
+            sys.stdout = NullWriter()
+            swapped = swap_face(faceanalysis_model, ref, img, faceswap_model, face_ids)
+            sys.stdout = sys.__stdout__
             return pil2tensor(swapped)
 
         batch_count = image.size(0)
 
-        logger.info(f"Running insightface swap (batch size: {batch_count})")
+        log.info(f"Running insightface swap (batch size: {batch_count})")
 
         if reference.size(0) != 1:
             raise ValueError("Reference image must have batch size 1")
@@ -86,38 +150,28 @@ class FaceSwap:
             image = do_swap(image)
 
         else:
-            image = [do_swap(image[i]) for i in range(batch_count)]
-            image = torch.cat(image, dim=0)
+            image_batch = [do_swap(image[i]) for i in range(batch_count)]
+            image = torch.cat(image_batch, dim=0)
 
         return (image,)
 
-    def getFaceSwapModel(self, model_path: str):
-        model_path = os.path.join(folder_paths.models_dir, "insightface", model_path)
-        if self.model_path is None or self.model_path != model_path:
-            logger.info(f"Loading model {model_path}")
-            self.model_path = model_path
-            self.model = insightface.model_zoo.get_model(
-                model_path, providers=providers
-            )
-        else:
-            logger.info("Using cached model")
-
-        logger.info("Model loaded")
-        return self.model
-
 
 # endregion
 
 
 # region face swap utils
-def get_face_single(img_data: np.ndarray, face_index=0, det_size=(640, 640)):
-    face_analyser = insightface.app.FaceAnalysis(name="buffalo_l", providers=providers)
+def get_face_single(
+    face_analyser, img_data: np.ndarray, face_index=0, det_size=(640, 640)
+):
     face_analyser.prepare(ctx_id=0, det_size=det_size)
     face = face_analyser.get(img_data)
 
     if len(face) == 0 and det_size[0] > 320 and det_size[1] > 320:
+        log.debug("No face ed, trying again with smaller image")
         det_size_half = (det_size[0] // 2, det_size[1] // 2)
-        return get_face_single(img_data, face_index=face_index, det_size=det_size_half)
+        return get_face_single(
+            face_analyser, img_data, face_index=face_index, det_size=det_size_half
+        )
 
     try:
         return sorted(face, key=lambda x: x.bbox[0])[face_index]
@@ -125,59 +179,45 @@ def get_face_single(img_data: np.ndarray, face_index=0, det_size=(640, 640)):
         return None
 
 
-def convert_to_sd(img) -> Tuple[bool, str]:
-    chunks = detect(img)
-    shapes = [chunk["score"] > 0.7 for chunk in chunks]
-    return [any(shapes), tempfile.NamedTemporaryFile(delete=False, suffix=".png")]
-
-
 def swap_face(
-    source_img: Image.Image,
-    target_img: Image.Image,
-    face_swapper_model=None,
-    faces_index: Set[int] = None,
+    face_analyser,
+    source_img: Union[Image.Image, List[Image.Image]],
+    target_img: Union[Image.Image, List[Image.Image]],
+    face_swapper_model,
+    faces_index: Optional[Set[int]] = None,
 ) -> Image.Image:
     if faces_index is None:
         faces_index = {0}
-    logger.info(f"Swapping faces: {faces_index}")
+    log.debug(f"Swapping faces: {faces_index}")
     result_image = target_img
-    converted = convert_to_sd(target_img)
-    scale, fn = converted[0], converted[1]
-    if face_swapper_model is not None and not scale:
-        if isinstance(source_img, str):  # source_img is a base64 string
-            import base64, io
 
-            if (
-                "base64," in source_img
-            ):  # check if the base64 string has a data URL scheme
-                base64_data = source_img.split("base64,")[-1]
-                img_bytes = base64.b64decode(base64_data)
-            else:
-                # if no data URL scheme, just decode
-                img_bytes = base64.b64decode(source_img)
-            source_img = Image.open(io.BytesIO(img_bytes))
-        source_img = cv2.cvtColor(np.array(source_img), cv2.COLOR_RGB2BGR)
-        target_img = cv2.cvtColor(np.array(target_img), cv2.COLOR_RGB2BGR)
-        source_face = get_face_single(source_img, face_index=0)
+    if face_swapper_model is not None:
+        cv_source_img = cv2.cvtColor(np.array(source_img), cv2.COLOR_RGB2BGR)
+        cv_target_img = cv2.cvtColor(np.array(target_img), cv2.COLOR_RGB2BGR)
+        source_face = get_face_single(face_analyser, cv_source_img, face_index=0)
         if source_face is not None:
-            result = target_img
+            result = cv_target_img
 
             for face_num in faces_index:
-                target_face = get_face_single(target_img, face_index=face_num)
+                target_face = get_face_single(
+                    face_analyser, cv_target_img, face_index=face_num
+                )
                 if target_face is not None:
+                    sys.stdout = NullWriter()
                     result = face_swapper_model.get(result, target_face, source_face)
+                    sys.stdout = sys.__stdout__
                 else:
-                    logger.warning(f"No target face found for {face_num}")
+                    log.warning(f"No target face found for {face_num}")
 
             result_image = Image.fromarray(cv2.cvtColor(result, cv2.COLOR_BGR2RGB))
         else:
-            logger.warning("No source face found")
+            log.warning("No source face found")
     else:
-        logger.error("No face swap model provided")
+        log.error("No face swap model provided")
     return result_image
 
 
 # endregion face swap utils
 
 
-__nodes__ = [FaceSwap]
+__nodes__ = [FaceSwap, LoadFaceSwapModel, LoadFaceAnalysisModel]

nodes/fun.py

@@ -1,66 +0,0 @@
-import qrcode
-from ..utils import pil2tensor
-from PIL import Image
-
-
-class QrCode:
-    """Basic QR Code generator"""
-
-    def __init__(self):
-        pass
-
-    @classmethod
-    def INPUT_TYPES(cls):
-        return {
-            "required": {
-                "url": ("STRING", {"default": "https://www.github.com"}),
-                "width": (
-                    "INT",
-                    {"default": 256, "max": 8096, "min": 0, "step": 1},
-                ),
-                "height": (
-                    "INT",
-                    {"default": 256, "max": 8096, "min": 0, "step": 1},
-                ),
-                "error_correct": (("L", "M", "Q", "H"), {"default": "L"}),
-                "box_size": ("INT", {"default": 10, "max": 8096, "min": 0, "step": 1}),
-                "border": ("INT", {"default": 4, "max": 8096, "min": 0, "step": 1}),
-                "invert": (("True", "False"), {"default": "False"}),
-            }
-        }
-
-    RETURN_TYPES = ("IMAGE",)
-    FUNCTION = "do_qr"
-    CATEGORY = "fun"
-
-    def do_qr(self, url, width, height, error_correct, box_size, border, invert):
-        if error_correct == "L" or error_correct not in ["M", "Q", "H"]:
-            error_correct = qrcode.constants.ERROR_CORRECT_L
-        elif error_correct == "M":
-            error_correct = qrcode.constants.ERROR_CORRECT_M
-        elif error_correct == "Q":
-            error_correct = qrcode.constants.ERROR_CORRECT_Q
-        else:
-            error_correct = qrcode.constants.ERROR_CORRECT_H
-
-        qr = qrcode.QRCode(
-            version=1,
-            error_correction=error_correct,
-            box_size=box_size,
-            border=border,
-        )
-        qr.add_data(url)
-        qr.make(fit=True)
-
-        back_color = (255, 255, 255) if invert == "True" else (0, 0, 0)
-        fill_color = (0, 0, 0) if invert == "True" else (255, 255, 255)
-
-        code = img = qr.make_image(back_color=back_color, fill_color=fill_color)
-
-        # that we now resize without filtering
-        code = code.resize((width, height), Image.NEAREST)
-
-        return (pil2tensor(code),)
-
-
-__nodes__ = [QrCode]

nodes/generate.py

@@ -0,0 +1,286 @@
+import qrcode
+from ..utils import pil2tensor
+from ..utils import comfy_dir
+from typing import cast
+from PIL import Image
+from ..log import log
+
+# class MtbExamples:
+#     """MTB Example Images"""
+
+#     def __init__(self):
+#         pass
+
+#     @classmethod
+#     @lru_cache(maxsize=1)
+#     def get_root(cls):
+#         return here / "examples" / "samples"
+
+#     @classmethod
+#     def INPUT_TYPES(cls):
+#         input_dir = cls.get_root()
+#         files = [f.name for f in input_dir.iterdir() if f.is_file()]
+#         return {
+#             "required": {"image": (sorted(files),)},
+#         }
+
+#     RETURN_TYPES = ("IMAGE", "MASK")
+#     FUNCTION = "do_mtb_examples"
+#     CATEGORY = "fun"
+
+#     def do_mtb_examples(self, image, index):
+#         image_path = (self.get_root() / image).as_posix()
+
+#         i = Image.open(image_path)
+#         i = ImageOps.exif_transpose(i)
+#         image = i.convert("RGB")
+#         image = np.array(image).astype(np.float32) / 255.0
+#         image = torch.from_numpy(image)[None,]
+#         if "A" in i.getbands():
+#             mask = np.array(i.getchannel("A")).astype(np.float32) / 255.0
+#             mask = 1.0 - torch.from_numpy(mask)
+#         else:
+#             mask = torch.zeros((64, 64), dtype=torch.float32, device="cpu")
+#         return (image, mask)
+
+#     @classmethod
+#     def IS_CHANGED(cls, image):
+#         image_path = (cls.get_root() / image).as_posix()
+
+#         m = hashlib.sha256()
+#         with open(image_path, "rb") as f:
+#             m.update(f.read())
+#         return m.digest().hex()
+
+
+class UnsplashImage:
+    """Unsplash Image given a keyword and a size"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "width": ("INT", {"default": 512, "max": 8096, "min": 0, "step": 1}),
+                "height": ("INT", {"default": 512, "max": 8096, "min": 0, "step": 1}),
+                "random_seed": ("INT", {"default": 0, "max": 1e5, "min": 0, "step": 1}),
+            },
+            "optional": {
+                "keyword": ("STRING", {"default": "nature"}),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "do_unsplash_image"
+    CATEGORY = "mtb/generate"
+
+    def do_unsplash_image(self, width, height, random_seed, keyword=None):
+        import requests
+        import io
+
+        base_url = "https://source.unsplash.com/random/"
+
+        if width and height:
+            base_url += f"/{width}x{height}"
+
+        if keyword:
+            keyword = keyword.replace(" ", "%20")
+            base_url += f"?{keyword}&{random_seed}"
+        else:
+            base_url += f"?&{random_seed}"
+        try:
+            log.debug(f"Getting unsplash image from {base_url}")
+            response = requests.get(base_url)
+            response.raise_for_status()
+
+            image = Image.open(io.BytesIO(response.content))
+            return (
+                pil2tensor(
+                    image,
+                ),
+            )
+
+        except requests.exceptions.RequestException as e:
+            print("Error retrieving image:", e)
+            return (None,)
+
+
+class QrCode:
+    """Basic QR Code generator"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "url": ("STRING", {"default": "https://www.github.com"}),
+                "width": (
+                    "INT",
+                    {"default": 256, "max": 8096, "min": 0, "step": 1},
+                ),
+                "height": (
+                    "INT",
+                    {"default": 256, "max": 8096, "min": 0, "step": 1},
+                ),
+                "error_correct": (("L", "M", "Q", "H"), {"default": "L"}),
+                "box_size": ("INT", {"default": 10, "max": 8096, "min": 0, "step": 1}),
+                "border": ("INT", {"default": 4, "max": 8096, "min": 0, "step": 1}),
+                "invert": (("BOOLEAN",), {"default": False}),
+            }
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "do_qr"
+    CATEGORY = "mtb/generate"
+
+    def do_qr(self, url, width, height, error_correct, box_size, border, invert):
+        log.warning(
+            "This node will soon be deprecated, there are much better alternatives like https://github.com/coreyryanhanson/comfy-qr"
+        )
+        if error_correct == "L" or error_correct not in ["M", "Q", "H"]:
+            error_correct = qrcode.constants.ERROR_CORRECT_L
+        elif error_correct == "M":
+            error_correct = qrcode.constants.ERROR_CORRECT_M
+        elif error_correct == "Q":
+            error_correct = qrcode.constants.ERROR_CORRECT_Q
+        else:
+            error_correct = qrcode.constants.ERROR_CORRECT_H
+
+        qr = qrcode.QRCode(
+            version=1,
+            error_correction=error_correct,
+            box_size=box_size,
+            border=border,
+        )
+        qr.add_data(url)
+        qr.make(fit=True)
+
+        back_color = (255, 255, 255) if invert else (0, 0, 0)
+        fill_color = (0, 0, 0) if invert else (255, 255, 255)
+
+        code = img = qr.make_image(back_color=back_color, fill_color=fill_color)
+
+        # that we now resize without filtering
+        code = code.resize((width, height), Image.NEAREST)
+
+        return (pil2tensor(code),)
+
+
+def bbox_dim(bbox):
+    left, upper, right, lower = bbox
+    width = right - left
+    height = lower - upper
+    return width, height
+
+
+class TextToImage:
+    """Utils to convert text to image using a font
+
+
+    The tool looks for any .ttf file in the Comfy folder hierarchy.
+    """
+
+    fonts = {}
+
+    def __init__(self):
+        # - This is executed when the graph is executed, we could conditionaly reload fonts there
+        pass
+
+    @classmethod
+    def CACHE_FONTS(cls):
+        font_extensions = ["*.ttf", "*.otf", "*.woff", "*.woff2", "*.eot"]
+        fonts = []
+
+        for extension in font_extensions:
+            fonts.extend(comfy_dir.glob(f"**/{extension}"))
+
+        if not fonts:
+            log.warn(
+                "> No fonts found in the comfy folder, place at least one font file somewhere in ComfyUI's hierarchy"
+            )
+        else:
+            log.debug(f"> Found {len(fonts)} fonts")
+
+        for font in fonts:
+            log.debug(f"Adding font {font}")
+            cls.fonts[font.stem] = font.as_posix()
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        if not cls.fonts:
+            cls.CACHE_FONTS()
+        else:
+            log.debug(f"Using cached fonts (count: {len(cls.fonts)})")
+        return {
+            "required": {
+                "text": (
+                    "STRING",
+                    {"default": "Hello world!"},
+                ),
+                "font": ((sorted(cls.fonts.keys())),),
+                "wrap": (
+                    "INT",
+                    {"default": 120, "min": 0, "max": 8096, "step": 1},
+                ),
+                "font_size": (
+                    "INT",
+                    {"default": 12, "min": 1, "max": 2500, "step": 1},
+                ),
+                "width": (
+                    "INT",
+                    {"default": 512, "min": 1, "max": 8096, "step": 1},
+                ),
+                "height": (
+                    "INT",
+                    {"default": 512, "min": 1, "max": 8096, "step": 1},
+                ),
+                # "position": (["INT"], {"default": 0, "min": 0, "max": 100, "step": 1}),
+                "color": (
+                    "COLOR",
+                    {"default": "black"},
+                ),
+                "background": (
+                    "COLOR",
+                    {"default": "white"},
+                ),
+            }
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("image",)
+    FUNCTION = "text_to_image"
+    CATEGORY = "mtb/generate"
+
+    def text_to_image(
+        self, text, font, wrap, font_size, width, height, color, background
+    ):
+        from PIL import Image, ImageDraw, ImageFont
+        import textwrap
+
+        font = self.fonts[font]
+        font = cast(ImageFont.FreeTypeFont, ImageFont.truetype(font, font_size))
+        if wrap == 0:
+            wrap = width / font_size
+        lines = textwrap.wrap(text, width=wrap)
+        log.debug(f"Lines: {lines}")
+        line_height = bbox_dim(font.getbbox("hg"))[1]
+        img_height = height  # line_height * len(lines)
+        img_width = width  # max(font.getsize(line)[0] for line in lines)
+
+        img = Image.new("RGBA", (img_width, img_height), background)
+        draw = ImageDraw.Draw(img)
+        y_text = 0
+        # - bbox is [left, upper, right, lower]
+        for line in lines:
+            width, height = bbox_dim(font.getbbox(line))
+            draw.text((0, y_text), line, color, font=font)
+            y_text += height
+
+        # img.save(os.path.join(folder_paths.base_path, f'{str(uuid.uuid4())}.png'))
+        return (pil2tensor(img),)
+
+
+__nodes__ = [
+    QrCode,
+    UnsplashImage,
+    TextToImage
+    #  MtbExamples,
+]

nodes/graph_utils.py

@@ -1,69 +1,326 @@
+import io
+import json
+import urllib.parse
+import urllib.request
+
+import numpy as np
 import torch
-import folder_paths
-import os
+from PIL import Image
+
+from ..log import log
+from ..utils import apply_easing, get_server_info, pil2tensor
 
 
-class SaveTensors:
-    """Debug node that will probably be removed in the future"""
+def get_image(filename, subfolder, folder_type):
+    log.debug(
+        f"Getting image {filename} from foldertype {folder_type} {f'in subfolder: {subfolder}' if subfolder else ''}"
+    )
+    data = {"filename": filename, "subfolder": subfolder, "type": folder_type}
+    base_url, port = get_server_info()
 
-    def __init__(self):
-        self.output_dir = folder_paths.get_output_directory()
-        self.type = "output"
+    url_values = urllib.parse.urlencode(data)
+    url = f"http://{base_url}:{port}/view?{url_values}"
+    log.debug(f"Fetching image from {url}")
+    with urllib.request.urlopen(url) as response:
+        return io.BytesIO(response.read())
+
+
+class GetBatchFromHistory:
+    """Very experimental node to load images from the history of the server.
+
+    Queue items without output are ignored in the count."""
 
     @classmethod
     def INPUT_TYPES(cls):
         return {
             "required": {
-                "filename_prefix": ("STRING", {"default": "ComfyPickle"}),
+                "enable": ("BOOLEAN", {"default": True}),
+                "count": ("INT", {"default": 1, "min": 0}),
+                "offset": ("INT", {"default": 0, "min": -1e9, "max": 1e9}),
+                "internal_count": ("INT", {"default": 0}),
             },
             "optional": {
-                "image": ("IMAGE",),
-                "mask": ("MASK",),
-                "latent": ("LATENT",),
+                "passthrough_image": ("IMAGE",),
             },
         }
 
-    FUNCTION = "save"
-    OUTPUT_NODE = True
-    RETURN_TYPES = ()
-    CATEGORY = "utils"
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("images",)
+    CATEGORY = "mtb/animation"
+    FUNCTION = "load_from_history"
 
-    def save(
+    def load_from_history(
         self,
-        filename_prefix,
-        image: torch.Tensor = None,
-        mask: torch.Tensor = None,
-        latent: torch.Tensor = None,
+        enable=True,
+        count=0,
+        offset=0,
+        internal_count=0,  # hacky way to invalidate the node
+        passthrough_image=None,
     ):
-        (
-            full_output_folder,
-            filename,
-            counter,
-            subfolder,
-            filename_prefix,
-        ) = folder_paths.get_save_image_path(filename_prefix, self.output_dir)
+        if not enable or count == 0:
+            if passthrough_image is not None:
+                log.debug("Using passthrough image")
+                return (passthrough_image,)
+            log.debug("Load from history is disabled for this iteration")
+            return (torch.zeros(0),)
+        frames = []
 
-        if image is not None:
-            image_file = f"{filename}_image_{counter:05}.pt"
-            torch.save(image, os.path.join(full_output_folder, image_file))
-            # np.save(os.path.join(full_output_folder, image_file), image.cpu().numpy())
+        base_url, port = get_server_info()
 
-        if mask is not None:
-            mask_file = f"{filename}_mask_{counter:05}.pt"
-            torch.save(mask, os.path.join(full_output_folder, mask_file))
-            # np.save(os.path.join(full_output_folder, mask_file), mask.cpu().numpy())
+        history_url = f"http://{base_url}:{port}/history"
+        log.debug(f"Fetching history from {history_url}")
+        output = torch.zeros(0)
+        with urllib.request.urlopen(history_url) as response:
+            output = self.load_batch_frames(response, offset, count, frames)
 
-        if latent is not None:
-            # for latent we must use pickle
-            latent_file = f"{filename}_latent_{counter:05}.pt"
-            torch.save(latent, os.path.join(full_output_folder, latent_file))
-            # pickle.dump(latent, open(os.path.join(full_output_folder, latent_file), "wb"))
+        if output.size(0) == 0:
+            log.warn("No output found in history")
 
-            # np.save(os.path.join(full_output_folder, latent_file), latent[""].cpu().numpy())
+        return (output,)
 
-        return f"{filename_prefix}_{counter:05}"
+    def load_batch_frames(self, response, offset, count, frames):
+        history = json.loads(response.read())
+
+        output_images = []
+
+        for run in history.values():
+            for node_output in run["outputs"].values():
+                if "images" in node_output:
+                    for image in node_output["images"]:
+                        image_data = get_image(
+                            image["filename"], image["subfolder"], image["type"]
+                        )
+                        output_images.append(image_data)
+
+        if not output_images:
+            return torch.zeros(0)
+
+        # Directly get desired range of images
+        start_index = max(len(output_images) - offset - count, 0)
+        end_index = len(output_images) - offset
+        selected_images = output_images[start_index:end_index]
+
+        frames = [Image.open(image) for image in selected_images]
+
+        if not frames:
+            return torch.zeros(0)
+        elif len(frames) != count:
+            log.warning(f"Expected {count} images, got {len(frames)} instead")
+
+        return pil2tensor(frames)
+
+
+class AnyToString:
+    """Tries to take any input and convert it to a string"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {"input": ("*")},
+        }
+
+    RETURN_TYPES = ("STRING",)
+    FUNCTION = "do_str"
+    CATEGORY = "mtb/converters"
+
+    def do_str(self, input):
+        if isinstance(input, str):
+            return (input,)
+        elif isinstance(input, torch.Tensor):
+            return (f"Tensor of shape {input.shape} and dtype {input.dtype}",)
+        elif isinstance(input, Image.Image):
+            return (f"PIL Image of size {input.size} and mode {input.mode}",)
+        elif isinstance(input, np.ndarray):
+            return (f"Numpy array of shape {input.shape} and dtype {input.dtype}",)
+
+        elif isinstance(input, dict):
+            return (f"Dictionary of {len(input)} items, with keys {input.keys()}",)
+
+        else:
+            log.debug(f"Falling back to string conversion of {input}")
+            return (str(input),)
+
+
+class StringReplace:
+    """Basic string replacement"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "string": ("STRING", {"forceInput": True}),
+                "old": ("STRING", {"default": ""}),
+                "new": ("STRING", {"default": ""}),
+            }
+        }
+
+    FUNCTION = "replace_str"
+    RETURN_TYPES = ("STRING",)
+    CATEGORY = "mtb/string"
+
+    def replace_str(self, string: str, old: str, new: str):
+        log.debug(f"Current string: {string}")
+        log.debug(f"Find string: {old}")
+        log.debug(f"Replace string: {new}")
+
+        string = string.replace(old, new)
+
+        log.debug(f"New string: {string}")
+
+        return (string,)
+
+
+class MTB_MathExpression:
+    """Node to evaluate a simple math expression string"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "expression": ("STRING", {"default": "", "multiline": True}),
+            }
+        }
+
+    FUNCTION = "eval_expression"
+    RETURN_TYPES = ("FLOAT", "INT")
+    RETURN_NAMES = ("result (float)", "result (int)")
+    CATEGORY = "mtb/math"
+    DESCRIPTION = "evaluate a simple math expression string (!! Fallsback to eval)"
+
+    def eval_expression(self, expression, **kwargs):
+        import math
+        from ast import literal_eval
+
+        for key, value in kwargs.items():
+            print(f"Replacing placeholder <{key}> with value {value}")
+            expression = expression.replace(f"<{key}>", str(value))
+
+        result = -1
+        try:
+            result = literal_eval(expression)
+        except SyntaxError as e:
+            raise ValueError(
+                f"The expression syntax is wrong '{expression}': {e}"
+            ) from e
+
+        except ValueError:
+            try:
+                expression = expression.replace("^", "**")
+                result = eval(expression)
+            except Exception as e:
+                # Handle any other exceptions and provide a meaningful error message
+                raise ValueError(
+                    f"Error evaluating expression '{expression}': {e}"
+                ) from e
+
+        return (result, int(result))
+
+
+class FitNumber:
+    """Fit the input float using a source and target range"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "value": ("FLOAT", {"default": 0, "forceInput": True}),
+                "clamp": ("BOOLEAN", {"default": False}),
+                "source_min": ("FLOAT", {"default": 0.0, "step": 0.01}),
+                "source_max": ("FLOAT", {"default": 1.0, "step": 0.01}),
+                "target_min": ("FLOAT", {"default": 0.0, "step": 0.01}),
+                "target_max": ("FLOAT", {"default": 1.0, "step": 0.01}),
+                "easing": (
+                    [
+                        "Linear",
+                        "Sine In",
+                        "Sine Out",
+                        "Sine In/Out",
+                        "Quart In",
+                        "Quart Out",
+                        "Quart In/Out",
+                        "Cubic In",
+                        "Cubic Out",
+                        "Cubic In/Out",
+                        "Circ In",
+                        "Circ Out",
+                        "Circ In/Out",
+                        "Back In",
+                        "Back Out",
+                        "Back In/Out",
+                        "Elastic In",
+                        "Elastic Out",
+                        "Elastic In/Out",
+                        "Bounce In",
+                        "Bounce Out",
+                        "Bounce In/Out",
+                    ],
+                    {"default": "Linear"},
+                ),
+            }
+        }
+
+    FUNCTION = "set_range"
+    RETURN_TYPES = ("FLOAT",)
+    CATEGORY = "mtb/math"
+    DESCRIPTION = "Fit the input float using a source and target range"
+
+    def set_range(
+        self,
+        value: float,
+        clamp: bool,
+        source_min: float,
+        source_max: float,
+        target_min: float,
+        target_max: float,
+        easing: str,
+    ):
+        if source_min == source_max:
+            normalized_value = 0
+        else:
+            normalized_value = (value - source_min) / (source_max - source_min)
+        if clamp:
+            normalized_value = max(min(normalized_value, 1), 0)
+
+        eased_value = apply_easing(normalized_value, easing)
+
+        # - Convert the eased value to the target range
+        res = target_min + (target_max - target_min) * eased_value
+
+        return (res,)
+
+
+class ConcatImages:
+    """Add images to batch"""
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "concatenate_tensors"
+    CATEGORY = "mtb/image"
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {"reverse": ("BOOLEAN", {"default": False})},
+        }
+
+    def concatenate_tensors(self, reverse, **kwargs):
+        tensors = tuple(kwargs.values())
+        batch_sizes = [tensor.size(0) for tensor in tensors]
+
+        concatenated = torch.cat(tensors, dim=0)
+
+        # Update the batch size in the concatenated tensor
+        concatenated_size = list(concatenated.size())
+        concatenated_size[0] = sum(batch_sizes)
+        concatenated = concatenated.view(*concatenated_size)
+
+        return (concatenated,)
 
 
 __nodes__ = [
-    SaveTensors,
+    StringReplace,
+    FitNumber,
+    GetBatchFromHistory,
+    AnyToString,
+    ConcatImages,
+    MTB_MathExpression,
 ]

nodes/image_interpolation.py

@@ -0,0 +1,123 @@
+import glob
+import os
+from pathlib import Path
+from typing import List
+
+import comfy
+import comfy.model_management as model_management
+import comfy.utils
+import folder_paths
+import numpy as np
+import tensorflow as tf
+import torch
+from frame_interpolation.eval import interpolator, util
+
+from ..errors import ModelNotFound
+from ..log import log
+from ..utils import get_model_path
+
+
+class LoadFilmModel:
+    """Loads a FILM model"""
+
+    @staticmethod
+    def get_models() -> List[Path]:
+        models_paths = get_model_path("FILM").iterdir()
+
+        return [x for x in models_paths if x.suffix in [".onnx", ".pth"]]
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "film_model": (
+                    ["L1", "Style", "VGG"],
+                    {"default": "Style"},
+                ),
+            },
+        }
+
+    RETURN_TYPES = ("FILM_MODEL",)
+    FUNCTION = "load_model"
+    CATEGORY = "mtb/frame iterpolation"
+
+    def load_model(self, film_model: str):
+        model_path = get_model_path("FILM", film_model)
+        if not model_path or not model_path.exists():
+            raise ModelNotFound(f"FILM ({model_path})")
+
+        if not (model_path / "saved_model.pb").exists():
+            model_path = model_path / "saved_model"
+
+        if not model_path.exists():
+            log.error(f"Model {model_path} does not exist")
+            raise ValueError(f"Model {model_path} does not exist")
+
+        log.info(f"Loading model {model_path}")
+
+        return (interpolator.Interpolator(model_path.as_posix(), None),)
+
+
+class FilmInterpolation:
+    """Google Research FILM frame interpolation for large motion"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "images": ("IMAGE",),
+                "interpolate": ("INT", {"default": 2, "min": 1, "max": 50}),
+                "film_model": ("FILM_MODEL",),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "do_interpolation"
+    CATEGORY = "mtb/frame iterpolation"
+
+    def do_interpolation(
+        self,
+        images: torch.Tensor,
+        interpolate: int,
+        film_model: interpolator.Interpolator,
+    ):
+        n = images.size(0)
+        # check if images is an empty tensor and return it...
+        if n == 0:
+            return (images,)
+
+        # check if tensorflow GPU is available
+        available_gpus = tf.config.list_physical_devices("GPU")
+        if not len(available_gpus):
+            log.warning(
+                "Tensorflow GPU not available, falling back to CPU this will be very slow"
+            )
+        else:
+            log.debug(f"Tensorflow GPU available, using {available_gpus}")
+
+        num_frames = (n - 1) * (2 ** (interpolate) - 1)
+        log.debug(f"Will interpolate into {num_frames} frames")
+
+        in_frames = [images[i] for i in range(n)]
+        out_tensors = []
+
+        pbar = comfy.utils.ProgressBar(num_frames)
+
+        for frame in util.interpolate_recursively_from_memory(
+            in_frames, interpolate, film_model
+        ):
+            out_tensors.append(
+                torch.from_numpy(frame) if isinstance(frame, np.ndarray) else frame
+            )
+            model_management.throw_exception_if_processing_interrupted()
+            pbar.update(1)
+
+        out_tensors = torch.cat([tens.unsqueeze(0) for tens in out_tensors], dim=0)
+
+        log.debug(f"Returning {len(out_tensors)} tensors")
+        log.debug(f"Output shape {out_tensors.shape}")
+        log.debug(f"Output type {out_tensors.dtype}")
+        return (out_tensors,)
+
+
+__nodes__ = [LoadFilmModel, FilmInterpolation]

nodes/image_processing.py

@@ -1,32 +1,42 @@
-import torch
-from skimage.filters import gaussian
-from skimage.restoration import denoise_tv_chambolle
-from skimage.util import compare_images
-from skimage.color import rgb2hsv, hsv2rgb
-import numpy as np
-import torchvision.transforms.functional as F
-from PIL import Image, ImageChops
-from ..utils import tensor2pil, pil2tensor, img_np_to_tensor, img_tensor_to_np
-import cv2
-import torch
-from ..log import log
-import folder_paths
-from PIL.PngImagePlugin import PngInfo
+import itertools
 import json
+import math
 import os
 
-try:
-    from cv2.ximgproc import guidedFilter
-except ImportError:
-    log.error("guidedFilter not found, use opencv-contrib-python")
+import cv2
+import folder_paths
+import numpy as np
+import torch
+import torch.nn.functional as F
+from PIL import Image
+from PIL.PngImagePlugin import PngInfo
+from skimage.filters import gaussian
+from skimage.util import compare_images
+
+from ..log import log
+from ..utils import pil2tensor, tensor2np, tensor2pil
+
+# try:
+#     from cv2.ximgproc import guidedFilter
+# except ImportError:
+#     log.warning("cv2.ximgproc.guidedFilter not found, use opencv-contrib-python")
+
+
+def gaussian_kernel(kernel_size: int, sigma_x: float, sigma_y: float, device=None):
+    x, y = torch.meshgrid(
+        torch.linspace(-1, 1, kernel_size, device=device),
+        torch.linspace(-1, 1, kernel_size, device=device),
+        indexing="ij",
+    )
+    d_x = x * x / (2.0 * sigma_x * sigma_x)
+    d_y = y * y / (2.0 * sigma_y * sigma_y)
+    g = torch.exp(-(d_x + d_y))
+    return g / g.sum()
 
 
 class ColorCorrect:
     """Various color correction methods"""
 
-    def __init__(self):
-        pass
-
     @classmethod
     def INPUT_TYPES(cls):
         return {
@@ -66,7 +76,7 @@ class ColorCorrect:
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "correct"
-    CATEGORY = "image/postprocessing"
+    CATEGORY = "mtb/image processing"
 
     @staticmethod
     def gamma_correction_tensor(image, gamma):
@@ -88,19 +98,21 @@ class ColorCorrect:
 
     @staticmethod
     def hsv_adjustment(image: torch.Tensor, hue, saturation, value):
-        image = tensor2pil(image)
-        hsv_image = image.convert("HSV")
+        images = tensor2pil(image)
+        out = []
+        for img in images:
+            hsv_image = img.convert("HSV")
 
-        h, s, v = hsv_image.split()
+            h, s, v = hsv_image.split()
 
-        h = h.point(lambda x: (x + hue * 255) % 256)
-        s = s.point(lambda x: int(x * saturation))
-        v = v.point(lambda x: int(x * value))
+            h = h.point(lambda x: (x + hue * 255) % 256)
+            s = s.point(lambda x: int(x * saturation))
+            v = v.point(lambda x: int(x * value))
 
-        hsv_image = Image.merge("HSV", (h, s, v))
-        rgb_image = hsv_image.convert("RGB")
-
-        return pil2tensor(rgb_image)
+            hsv_image = Image.merge("HSV", (h, s, v))
+            rgb_image = hsv_image.convert("RGB")
+            out.append(rgb_image)
+        return pil2tensor(out)
 
     @staticmethod
     def hsv_adjustment_tensor_not_working(image: torch.Tensor, hue, saturation, value):
@@ -180,70 +192,9 @@ class ColorCorrect:
         return (image,)
 
 
-class HsvToRgb:
-    """Convert HSV image to RGB"""
-
-    def __init__(self):
-        pass
-
-    @classmethod
-    def INPUT_TYPES(cls):
-        return {
-            "required": {
-                "image": ("IMAGE",),
-            }
-        }
-
-    RETURN_TYPES = ("IMAGE",)
-    FUNCTION = "convert"
-    CATEGORY = "image/postprocessing"
-
-    def convert(self, image):
-        image = image.numpy()
-
-        image = image.squeeze()
-        # image = image.transpose(1,2,3,0)
-        image = hsv2rgb(image)
-        image = np.expand_dims(image, axis=0)
-
-        # image = image.transpose(3,0,1,2)
-        return (torch.from_numpy(image),)
-
-
-class RgbToHsv:
-    """Convert RGB image to HSV"""
-
-    def __init__(self):
-        pass
-
-    @classmethod
-    def INPUT_TYPES(cls):
-        return {
-            "required": {
-                "image": ("IMAGE",),
-            }
-        }
-
-    RETURN_TYPES = ("IMAGE",)
-    FUNCTION = "convert"
-    CATEGORY = "image/postprocessing"
-
-    def convert(self, image):
-        image = image.numpy()
-
-        image = np.squeeze(image)
-        image = rgb2hsv(image)
-        image = np.expand_dims(image, axis=0)
-
-        return (torch.from_numpy(image),)
-
-
-class ImageCompare:
+class ImageCompare_:
     """Compare two images and return a difference image"""
 
-    def __init__(self):
-        pass
-
     @classmethod
     def INPUT_TYPES(cls):
         return {
@@ -259,7 +210,7 @@ class ImageCompare:
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "compare"
-    CATEGORY = "image"
+    CATEGORY = "mtb/image"
 
     def compare(self, imageA: torch.Tensor, imageB: torch.Tensor, mode):
         imageA = imageA.numpy()
@@ -274,43 +225,38 @@ class ImageCompare:
         return (torch.from_numpy(image),)
 
 
-class Denoise:
-    """Denoise an image using total variation minimization."""
+import requests
 
-    def __init__(self):
-        pass
+
+class LoadImageFromUrl_:
+    """Load an image from the given URL"""
 
     @classmethod
     def INPUT_TYPES(cls):
         return {
             "required": {
-                "image": ("IMAGE",),
-                "weight": (
-                    "FLOAT",
-                    {"default": 0.5, "min": 0.0, "max": 1.0, "step": 0.01},
+                "url": (
+                    "STRING",
+                    {
+                        "default": "https://upload.wikimedia.org/wikipedia/commons/thumb/a/a7/Example.jpg/800px-Example.jpg"
+                    },
                 ),
             }
         }
 
     RETURN_TYPES = ("IMAGE",)
-    FUNCTION = "denoise"
-    CATEGORY = "image/postprocessing"
+    FUNCTION = "load"
+    CATEGORY = "mtb/IO"
 
-    def denoise(self, image: torch.Tensor, weight):
-        image = image.numpy()
-        image = image.squeeze()
-        image = denoise_tv_chambolle(image, weight=weight)
-
-        image = np.expand_dims(image, axis=0)
-        return (torch.from_numpy(image),)
+    def load(self, url):
+        # get the image from the url
+        image = Image.open(requests.get(url, stream=True).raw)
+        return (pil2tensor(image),)
 
 
-class Blur:
+class Blur_:
     """Blur an image using a Gaussian filter."""
 
-    def __init__(self):
-        pass
-
     @classmethod
     def INPUT_TYPES(cls):
         return {
@@ -329,7 +275,7 @@ class Blur:
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "blur"
-    CATEGORY = "image/postprocessing"
+    CATEGORY = "mtb/image processing"
 
     def blur(self, image: torch.Tensor, sigmaX, sigmaY):
         image = image.numpy()
@@ -339,38 +285,107 @@ class Blur:
         return (torch.from_numpy(image),)
 
 
-# https://github.com/lllyasviel/AdverseCleaner/blob/main/clean.py
-def deglaze_np_img(np_img):
-    y = np_img.copy()
-    for _ in range(64):
-        y = cv2.bilateralFilter(y, 5, 8, 8)
-    for _ in range(4):
-        y = guidedFilter(np_img, y, 4, 16)
-    return y
-
-
-class DeglazeImage:
-    """Remove adversarial noise from images"""
+class Sharpen_:
+    """Sharpens an image using a Gaussian kernel."""
 
     @classmethod
     def INPUT_TYPES(cls):
-        return {"required": {"image": ("IMAGE",)}}
-
-    CATEGORY = "image"
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "sharpen_radius": (
+                    "INT",
+                    {"default": 1, "min": 1, "max": 31, "step": 1},
+                ),
+                "sigma_x": (
+                    "FLOAT",
+                    {"default": 1.0, "min": 0.1, "max": 10.0, "step": 0.1},
+                ),
+                "sigma_y": (
+                    "FLOAT",
+                    {"default": 1.0, "min": 0.1, "max": 10.0, "step": 0.1},
+                ),
+                "alpha": (
+                    "FLOAT",
+                    {"default": 1.0, "min": 0.0, "max": 5.0, "step": 0.1},
+                ),
+            },
+        }
 
     RETURN_TYPES = ("IMAGE",)
-    FUNCTION = "deglaze_image"
+    FUNCTION = "do_sharp"
+    CATEGORY = "mtb/image processing"
 
-    def deglaze_image(self, image):
-        return (img_np_to_tensor(deglaze_np_img(img_tensor_to_np(image))),)
+    def do_sharp(
+        self,
+        image: torch.Tensor,
+        sharpen_radius: int,
+        sigma_x: float,
+        sigma_y: float,
+        alpha: float,
+    ):
+        if sharpen_radius == 0:
+            return (image,)
+
+        channels = image.shape[3]
+
+        kernel_size = 2 * sharpen_radius + 1
+        kernel = gaussian_kernel(kernel_size, sigma_x, sigma_y) * -(alpha * 10)
+
+        # Modify center of kernel to make it a sharpening kernel
+        center = kernel_size // 2
+        kernel[center, center] = kernel[center, center] - kernel.sum() + 1.0
+
+        kernel = kernel.repeat(channels, 1, 1).unsqueeze(1)
+        tensor_image = image.permute(0, 3, 1, 2)
+
+        tensor_image = F.pad(
+            tensor_image,
+            (sharpen_radius, sharpen_radius, sharpen_radius, sharpen_radius),
+            "reflect",
+        )
+        sharpened = F.conv2d(tensor_image, kernel, padding=center, groups=channels)
+
+        # Remove padding
+        sharpened = sharpened[
+            :, :, sharpen_radius:-sharpen_radius, sharpen_radius:-sharpen_radius
+        ]
+
+        sharpened = sharpened.permute(0, 2, 3, 1)
+        result = torch.clamp(sharpened, 0, 1)
+
+        return (result,)
+
+
+# https://github.com/lllyasviel/AdverseCleaner/blob/main/clean.py
+# def deglaze_np_img(np_img):
+#     y = np_img.copy()
+#     for _ in range(64):
+#         y = cv2.bilateralFilter(y, 5, 8, 8)
+#     for _ in range(4):
+#         y = guidedFilter(np_img, y, 4, 16)
+#     return y
+
+
+# class DeglazeImage:
+#     """Remove adversarial noise from images"""
+
+#     @classmethod
+#     def INPUT_TYPES(cls):
+#         return {"required": {"image": ("IMAGE",)}}
+
+#     CATEGORY = "mtb/image processing"
+
+#     RETURN_TYPES = ("IMAGE",)
+#     FUNCTION = "deglaze_image"
+
+#     def deglaze_image(self, image):
+#         return (np2tensor(deglaze_np_img(tensor2np(image))),)
 
 
 class MaskToImage:
     """Converts a mask (alpha) to an RGB image with a color and background"""
 
-    def __init__(self):
-        pass
-
     @classmethod
     def INPUT_TYPES(cls):
         return {
@@ -381,29 +396,33 @@ class MaskToImage:
             }
         }
 
-    CATEGORY = "image/mask"
+    CATEGORY = "mtb/generate"
 
     RETURN_TYPES = ("IMAGE",)
 
     FUNCTION = "render_mask"
 
     def render_mask(self, mask, color, background):
-        mask = img_tensor_to_np(mask)
-        mask = Image.fromarray(mask).convert("L")
+        masks = tensor2np(mask)
+        images = []
+        for m in masks:
+            _mask = Image.fromarray(m).convert("L")
 
-        image = Image.new("RGBA", mask.size, color=color)
-        # apply the mask
-        image = Image.composite(
-            image, Image.new("RGBA", mask.size, color=background), mask
-        )
+            log.debug(f"Converted mask to PIL Image format, size: {_mask.size}")
 
-        # image = ImageChops.multiply(image, mask)
-        # apply over background
-        # image = Image.alpha_composite(Image.new("RGBA", image.size, color=background), image)
+            image = Image.new("RGBA", _mask.size, color=color)
+            # apply the mask
+            image = Image.composite(
+                image, Image.new("RGBA", _mask.size, color=background), _mask
+            )
 
-        image = pil2tensor(image.convert("RGB"))
+            # image = ImageChops.multiply(image, mask)
+            # apply over background
+            # image = Image.alpha_composite(Image.new("RGBA", image.size, color=background), image)
 
-        return (image,)
+            images.append(image.convert("RGB"))
+
+        return (pil2tensor(images),)
 
 
 class ColoredImage:
@@ -419,68 +438,102 @@ class ColoredImage:
                 "color": ("COLOR",),
                 "width": ("INT", {"default": 512, "min": 16, "max": 8160}),
                 "height": ("INT", {"default": 512, "min": 16, "max": 8160}),
-            }
+            },
+            "optional": {
+                "foreground_image": ("IMAGE",),
+                "foreground_mask": ("MASK",),
+            },
         }
 
-    CATEGORY = "image"
+    CATEGORY = "mtb/generate"
 
     RETURN_TYPES = ("IMAGE",)
 
     FUNCTION = "render_img"
 
-    def render_img(self, color, width, height):
-        image = Image.new("RGB", (width, height), color=color)
+    def render_img(
+        self, color, width, height, foreground_image=None, foreground_mask=None
+    ):
+        image = Image.new("RGBA", (width, height), color=color)
+        output = []
+        if foreground_image is not None:
+            if foreground_mask is None:
+                fg_images = tensor2pil(foreground_image)
+                for img in fg_images:
+                    if image.size != img.size:
+                        raise ValueError(
+                            f"Dimension mismatch: image {image.size}, img {img.size}"
+                        )
 
-        image = pil2tensor(image)
+                    if img.mode != "RGBA":
+                        raise ValueError(
+                            f"Foreground image must be in 'RGBA' mode when no mask is provided, got {img.mode}"
+                        )
 
-        return (image,)
+                    output.append(Image.alpha_composite(image, img).convert("RGB"))
+
+            elif foreground_image.size[0] != foreground_mask.size[0]:
+                raise ValueError("Foreground image and mask must have same batch size")
+            else:
+                fg_images = tensor2pil(foreground_image)
+                fg_masks = tensor2pil(foreground_mask)
+                output.extend(
+                    Image.composite(
+                        fg_image.convert("RGBA"),
+                        image,
+                        fg_mask,
+                    ).convert("RGB")
+                    for fg_image, fg_mask in zip(fg_images, fg_masks)
+                )
+        elif foreground_mask is not None:
+            log.warn("Mask ignored because no foreground image is given")
+
+        output = pil2tensor(output)
+
+        return (output,)
 
 
 class ImagePremultiply:
     """Premultiply image with mask"""
 
-    def __init__(self):
-        pass
-
     @classmethod
     def INPUT_TYPES(cls):
         return {
             "required": {
                 "image": ("IMAGE",),
                 "mask": ("MASK",),
-                "invert": (["True", "False"], {"default": "False"}),
+                "invert": ("BOOLEAN", {"default": False}),
             }
         }
 
-    CATEGORY = "image"
+    CATEGORY = "mtb/image"
     RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("RGBA",)
     FUNCTION = "premultiply"
 
     def premultiply(self, image, mask, invert):
-        invert = invert == "True"
-        image = tensor2pil(image)
-        mask = tensor2pil(mask).convert("L")
+        images = tensor2pil(image)
+        masks = tensor2pil(mask) if invert else tensor2pil(1.0 - mask)
+        single = len(mask) == 1
+        masks = [x.convert("L") for x in masks]
 
-        if invert:
-            mask = ImageChops.invert(mask)
+        out = []
+        for i, img in enumerate(images):
+            cur_mask = masks[0] if single else masks[i]
 
-        image.putalpha(mask)
+            img.putalpha(cur_mask)
+            out.append(img)
 
         # if invert:
         #     image = Image.composite(image,Image.new("RGBA", image.size, color=(0,0,0,0)), mask)
         # else:
         #     image = Image.composite(Image.new("RGBA", image.size, color=(0,0,0,0)), image, mask)
 
-        return (pil2tensor(image),)
+        return (pil2tensor(out),)
 
 
 class ImageResizeFactor:
-    """
-    Extracted mostly from WAS Node Suite, with a few edits (most notably multiple image support) and less features.
-    """
-
-    def __init__(self):
-        pass
+    """Extracted mostly from WAS Node Suite, with a few edits (most notably multiple image support) and less features."""
 
     @classmethod
     def INPUT_TYPES(cls):
@@ -491,10 +544,18 @@ class ImageResizeFactor:
                     "FLOAT",
                     {"default": 2, "min": 0.01, "max": 16.0, "step": 0.01},
                 ),
-                "supersample": (["true", "false"], {"default": "true"}),
+                "supersample": ("BOOLEAN", {"default": True}),
                 "resampling": (
-                    ["lanczos", "nearest", "bilinear", "bicubic"],
-                    {"default": "lanczos"},
+                    [
+                        "nearest",
+                        "linear",
+                        "bilinear",
+                        "bicubic",
+                        "trilinear",
+                        "area",
+                        "nearest-exact",
+                    ],
+                    {"default": "nearest"},
                 ),
             },
             "optional": {
@@ -502,103 +563,68 @@ class ImageResizeFactor:
             },
         }
 
-    CATEGORY = "image"
+    CATEGORY = "mtb/image"
     RETURN_TYPES = ("IMAGE", "MASK")
     FUNCTION = "resize"
 
-    def resize_image(
-        self,
-        image: torch.Tensor,
-        factor: float = 0.5,
-        supersample=False,
-        resample="lanczos",
-        mask=None,
-    ) -> torch.Tensor:
-        batch_count = 1
-        img = tensor2pil(image)
-
-        if isinstance(img, list):
-            log.debug("Multiple images detected (list)")
-            out = []
-            for im in img:
-                im = self.resize_image(
-                    pil2tensor(im), factor, supersample, resample, mask
-                )
-                out.append(im)
-            return torch.cat(out, dim=0)
-        elif isinstance(img, torch.Tensor):
-            if len(image.shape) > 3:
-                batch_count = image.size(0)
-
-        if batch_count > 1:
-            log.debug("Multiple images detected (batch count)")
-            out = [
-                self.resize_image(image[i], factor, supersample, resample, mask)
-                for i in range(batch_count)
-            ]
-            return torch.cat(out, dim=0)
-
-        log.debug("Resizing image")
-        # Get the current width and height of the image
-        current_width, current_height = img.size
-
-        log.debug(f"Current width: {current_width}, Current height: {current_height}")
-
-        # Calculate the new width and height based on the given mode and parameters
-        new_width, new_height = int(factor * current_width), int(
-            factor * current_height
-        )
-
-        log.debug(f"New width: {new_width}, New height: {new_height}")
-
-        # Define a dictionary of resampling filters
-        resample_filters = {"nearest": 0, "bilinear": 2, "bicubic": 3, "lanczos": 1}
-
-        # Apply supersample
-        if supersample == "true":
-            super_size = (new_width * 8, new_height * 8)
-            log.debug(f"Applying supersample: {super_size}")
-            img = img.resize(
-                super_size, resample=Image.Resampling(resample_filters[resample])
-            )
-
-        # Resize the image using the given resampling filter
-        resized_image = img.resize(
-            (new_width, new_height),
-            resample=Image.Resampling(resample_filters[resample]),
-        )
-
-        return pil2tensor(resized_image)
-
     def resize(
         self,
         image: torch.Tensor,
         factor: float,
-        supersample: str,
+        supersample: bool,
         resampling: str,
         mask=None,
     ):
-        log.debug(f"Resizing image with factor {factor} and resampling {resampling}")
-        supersample = supersample == "true"
-        batch_count = image.size(0)
-        log.debug(f"Batch count: {batch_count}")
-        if batch_count == 1:
-            log.debug("Batch count is 1, returning single image")
-            return (self.resize_image(image, factor, supersample, resampling),)
+        # Check if the tensor has the correct dimension
+        if len(image.shape) not in [3, 4]:  # HxWxC or BxHxWxC
+            raise ValueError("Expected image tensor of shape (H, W, C) or (B, H, W, C)")
+
+        # Transpose to CxHxW or BxCxHxW for PyTorch
+        if len(image.shape) == 3:
+            image = image.permute(2, 0, 1).unsqueeze(0)  # CxHxW
         else:
-            log.debug("Batch count is greater than 1, returning multiple images")
-            images = [
-                self.resize_image(image[i], factor, supersample, resampling)
-                for i in range(batch_count)
-            ]
-            images = torch.cat(images, dim=0)
-            return (images,)
+            image = image.permute(0, 3, 1, 2)  # BxCxHxW
+
+        # Compute new dimensions
+        B, C, H, W = image.shape
+        new_H, new_W = int(H * factor), int(W * factor)
+
+        align_corner_filters = ("linear", "bilinear", "bicubic", "trilinear")
+        # Resize the image
+        resized_image = F.interpolate(
+            image,
+            size=(new_H, new_W),
+            mode=resampling,
+            align_corners=resampling in align_corner_filters,
+        )
+
+        # Optionally supersample
+        if supersample:
+            resized_image = F.interpolate(
+                resized_image,
+                scale_factor=2,
+                mode=resampling,
+                align_corners=resampling in align_corner_filters,
+            )
+
+        # Transpose back to the original format: BxHxWxC or HxWxC
+        if len(image.shape) == 4:
+            resized_image = resized_image.permute(0, 2, 3, 1)
+        else:
+            resized_image = resized_image.squeeze(0).permute(1, 2, 0)
+
+        # Apply mask if provided
+        if mask is not None:
+            if len(mask.shape) != len(resized_image.shape):
+                raise ValueError(
+                    "Mask tensor should have the same dimensions as the image tensor"
+                )
+            resized_image = resized_image * mask
+
+        return (resized_image,)
 
 
-import math
-
-
-class SaveImageGrid:
+class SaveImageGrid_:
     """Save all the images in the input batch as a grid of images."""
 
     def __init__(self):
@@ -611,7 +637,7 @@ class SaveImageGrid:
             "required": {
                 "images": ("IMAGE",),
                 "filename_prefix": ("STRING", {"default": "ComfyUI"}),
-                "save_intermediate": (["true", "false"], {"default": "false"}),
+                "save_intermediate": ("BOOLEAN", {"default": False}),
             },
             "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
         }
@@ -621,7 +647,7 @@ class SaveImageGrid:
 
     OUTPUT_NODE = True
 
-    CATEGORY = "image"
+    CATEGORY = "mtb/IO"
 
     def create_image_grid(self, image_list):
         total_images = len(image_list)
@@ -652,11 +678,10 @@ class SaveImageGrid:
         self,
         images,
         filename_prefix="Grid",
-        save_intermediate="false",
+        save_intermediate=False,
         prompt=None,
         extra_pnginfo=None,
     ):
-        save_intermediate = save_intermediate == "true"
         (
             full_output_folder,
             filename,
@@ -701,17 +726,65 @@ class SaveImageGrid:
         return {"ui": {"images": results}}
 
 
+class ImageTileOffset:
+    """Mimics an old photoshop technique to check for seamless textures"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "tiles": ("INT", {"default": 2}),
+            }
+        }
+
+    CATEGORY = "mtb/generate"
+
+    RETURN_TYPES = ("IMAGE",)
+
+    FUNCTION = "tile_image"
+
+    def tile_image(self, image: torch.Tensor, tiles: int = 2):
+        if tiles < 1:
+            raise ValueError("The number of tiles must be at least 1.")
+
+        batch_size, height, width, channels = image.shape
+        tile_height = height // tiles
+        tile_width = width // tiles
+
+        output_image = torch.zeros_like(image)
+
+        for i, j in itertools.product(range(tiles), range(tiles)):
+            start_h = i * tile_height
+            end_h = start_h + tile_height
+            start_w = j * tile_width
+            end_w = start_w + tile_width
+
+            tile = image[:, start_h:end_h, start_w:end_w, :]
+
+            output_start_h = (i + 1) % tiles * tile_height
+            output_start_w = (j + 1) % tiles * tile_width
+            output_end_h = output_start_h + tile_height
+            output_end_w = output_start_w + tile_width
+
+            output_image[
+                :, output_start_h:output_end_h, output_start_w:output_end_w, :
+            ] = tile
+
+        return (output_image,)
+
+
 __nodes__ = [
     ColorCorrect,
-    HsvToRgb,
-    RgbToHsv,
-    ImageCompare,
-    Denoise,
-    Blur,
-    DeglazeImage,
+    ImageCompare_,
+    ImageTileOffset,
+    Blur_,
+    # DeglazeImage,
     MaskToImage,
     ColoredImage,
     ImagePremultiply,
     ImageResizeFactor,
-    SaveImageGrid,
+    SaveImageGrid_,
+    LoadImageFromUrl_,
+    Sharpen_,
 ]

nodes/image_utils.py

@@ -0,0 +1,41 @@
+import torch
+
+from ..log import log
+
+
+class StackImages:
+    """Stack the input images horizontally or vertically"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {"required": {"vertical": ("BOOLEAN", {"default": False})}}
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "stack"
+    CATEGORY = "mtb/image utils"
+
+    def stack(self, vertical, **kwargs):
+        if not kwargs:
+            raise ValueError("At least one tensor must be provided.")
+
+        tensors = list(kwargs.values())
+        log.debug(
+            f"Stacking {len(tensors)} tensors {'vertically' if vertical else 'horizontally'}"
+        )
+        log.debug(list(kwargs.keys()))
+
+        ref_shape = tensors[0].shape
+        for tensor in tensors[1:]:
+            if tensor.shape[1:] != ref_shape[1:]:
+                raise ValueError(
+                    "All tensors must have the same dimensions except for the stacking dimension."
+                )
+
+        dim = 1 if vertical else 2
+
+        stacked_tensor = torch.cat(tensors, dim=dim)
+
+        return (stacked_tensor,)
+
+
+__nodes__ = [StackImages]

nodes/io.py

@@ -0,0 +1,253 @@
+import json
+import subprocess
+import uuid
+from pathlib import Path
+from typing import List, Optional
+
+import comfy.model_management as model_management
+import folder_paths
+import numpy as np
+import torch
+from comfy.model_management import get_torch_device
+from PIL import Image
+
+from ..log import log
+from ..utils import PIL_FILTER_MAP, audioInputDir, tensor2np
+
+try:
+    import librosa
+except ImportError:
+    log.warning("librosa not installed. I/O Audio features will not be available.")
+
+
+class LoadAudio_:
+    """Load an audio file from the input folder (supports upload)"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "audio": ("AUDIO_UPLOAD",),
+                "sample_rate": ("INT", {"default": 44100}),
+            }
+        }
+
+    RETURN_TYPES = ("AUDIO",)
+    RETURN_NAMES = ("audio",)
+    FUNCTION = "load_audio"
+    CATEGORY = "mtb/audio"
+
+    def load_audio(self, audio: str, sample_rate: int):
+        log.debug(f"Audio file: {audio}")
+        audio_file_path = audioInputDir / audio
+        log.debug(f"Loading audio file: {audio_file_path}")
+        audio_data, _ = librosa.load(audio_file_path.as_posix(), sr=sample_rate)
+        audio_tensor = torch.from_numpy(audio_data).to(get_torch_device())
+        return (audio_tensor.unsqueeze(0).float(),)
+
+
+class ExportWithFfmpeg:
+    """Export with FFmpeg (Experimental)"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "images": ("IMAGE",),
+                # "frames": ("FRAMES",),
+                "fps": ("FLOAT", {"default": 24, "min": 1}),
+                "prefix": ("STRING", {"default": "export"}),
+                "format": (["mov", "mp4", "mkv", "avi"], {"default": "mov"}),
+                "codec": (
+                    ["prores_ks", "libx264", "libx265"],
+                    {"default": "prores_ks"},
+                ),
+            },
+            "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
+        }
+
+    RETURN_TYPES = ("VIDEO",)
+    OUTPUT_NODE = True
+    FUNCTION = "export_prores"
+    CATEGORY = "mtb/IO"
+
+    def export_prores(
+        self,
+        images: torch.Tensor,
+        fps: float,
+        prefix: str,
+        format: str,
+        codec: str,
+        prompt=None,
+        extra_pnginfo=None,
+    ):
+        metadata = {}
+        if images.size(0) == 0:
+            return ("",)
+
+        if extra_pnginfo is not None:
+            metadata["extra"] = {}
+            for x in extra_pnginfo:
+                metadata["extra"][x] = json.dumps(extra_pnginfo[x])
+
+        if prompt is not None:
+            metadata["prompt"] = json.dumps(prompt)
+
+        output_dir = Path(folder_paths.get_output_directory())
+        pix_fmt = "rgb48le" if codec == "prores_ks" else "yuv420p"
+        file_ext = format
+        file_id = f"{prefix}_{uuid.uuid4()}.{file_ext}"
+
+        log.debug(f"Exporting to {output_dir / file_id}")
+
+        frames = tensor2np(images)
+        log.debug(f"Frames type {type(frames[0])}")
+        log.debug(f"Exporting {len(frames)} frames")
+
+        frames = [frame.astype(np.uint16) * 257 for frame in frames]
+
+        height, width, _ = frames[0].shape
+
+        out_path = (output_dir / file_id).as_posix()
+
+        metadata_cmd = []
+
+        if metadata:
+            for k, v in metadata.items():
+                metadata_cmd += [
+                    "-metadata:s:v",
+                    f"{k}='{v if isinstance(v,str) else json.dumps(v)}'",
+                ]
+
+        # Prepare the FFmpeg command
+        command = [
+            "ffmpeg",
+            "-y",
+            "-f",
+            "rawvideo",
+            "-vcodec",
+            "rawvideo",
+            "-s",
+            f"{width}x{height}",
+            "-pix_fmt",
+            pix_fmt,
+            "-r",
+            str(fps),
+            "-i",
+            "-",
+            "-c:v",
+            codec,
+            *metadata_cmd,
+            "-r",
+            str(fps),
+            "-y",
+            out_path,
+        ]
+
+        process = subprocess.Popen(command, stdin=subprocess.PIPE)
+
+        for frame in frames:
+            model_management.throw_exception_if_processing_interrupted()
+            process.stdin.write(frame.tobytes())
+
+        process.stdin.close()
+        process.wait()
+
+        return (out_path,)
+
+
+def prepare_animated_batch(
+    batch: torch.Tensor,
+    pingpong=False,
+    resize_by=1.0,
+    resample_filter: Optional[Image.Resampling] = None,
+    image_type=np.uint8,
+) -> List[Image.Image]:
+    images = tensor2np(batch)
+    images = [frame.astype(image_type) for frame in images]
+
+    height, width, _ = batch[0].shape
+
+    if pingpong:
+        reversed_frames = images[::-1]
+        images.extend(reversed_frames)
+    pil_images = [Image.fromarray(frame) for frame in images]
+
+    # Resize frames if necessary
+    if abs(resize_by - 1.0) > 1e-6:
+        new_width = int(width * resize_by)
+        new_height = int(height * resize_by)
+        pil_images_resized = [
+            frame.resize((new_width, new_height), resample=resample_filter)
+            for frame in pil_images
+        ]
+        pil_images = pil_images_resized
+
+    return pil_images
+
+
+# todo: deprecate for apng
+class SaveGif:
+    """Save the images from the batch as a GIF"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "fps": ("INT", {"default": 12, "min": 1, "max": 120}),
+                "resize_by": ("FLOAT", {"default": 1.0, "min": 0.1}),
+                "optimize": ("BOOLEAN", {"default": False}),
+                "pingpong": ("BOOLEAN", {"default": False}),
+            },
+            "optional": {
+                "resample_filter": (list(PIL_FILTER_MAP.keys()),),
+            },
+        }
+
+    RETURN_TYPES = ()
+    OUTPUT_NODE = True
+    CATEGORY = "mtb/IO"
+    FUNCTION = "save_gif"
+
+    def save_gif(
+        self,
+        image,
+        fps=12,
+        resize_by=1.0,
+        optimize=False,
+        pingpong=False,
+        resample_filter=None,
+    ):
+        if image.size(0) == 0:
+            return ("",)
+
+        if resample_filter is not None:
+            resample_filter = PIL_FILTER_MAP.get(resample_filter)
+
+        pil_images = prepare_animated_batch(
+            image,
+            pingpong,
+            resize_by,
+            resample_filter,
+        )
+
+        ruuid = uuid.uuid4()
+        ruuid = ruuid.hex[:10]
+        out_path = f"{folder_paths.output_directory}/{ruuid}.gif"
+
+        # Create the GIF from PIL images
+        pil_images[0].save(
+            out_path,
+            save_all=True,
+            append_images=pil_images[1:],
+            optimize=optimize,
+            duration=int(1000 / fps),
+            loop=0,
+        )
+
+        results = [{"filename": f"{ruuid}.gif", "subfolder": "", "type": "output"}]
+        return {"ui": {"gif": results}}
+
+
+__nodes__ = [SaveGif, ExportWithFfmpeg, LoadAudio_]

nodes/latent_processing.py

@@ -1,9 +1,8 @@
 import torch
 
+
 class LatentLerp:
     """Linear interpolation (blend) between two latent vectors"""
-    def __init__(self):
-        pass
 
     @classmethod
     def INPUT_TYPES(cls):
@@ -18,7 +17,7 @@ class LatentLerp:
     RETURN_TYPES = ("LATENT",)
     FUNCTION = "lerp_latent"
 
-    CATEGORY = "latent"
+    CATEGORY = "mtb/latent"
 
     def lerp_latent(self, A, B, t):
         a = A.copy()
@@ -28,6 +27,7 @@ class LatentLerp:
 
         return (a,)
 
+
 __nodes__ = [
     LatentLerp,
-]
+]

nodes/mask.py

@@ -1,57 +1,108 @@
-from rembg import remove
-from ..utils import pil2tensor, tensor2pil
+import comfy.utils
 from PIL import Image
+from rembg import remove
+
+from ..utils import pil2tensor, tensor2pil
+
 
 class ImageRemoveBackgroundRembg:
-    def __init__(self):
-        pass
+    """Removes the background from the input using Rembg."""
 
     @classmethod
     def INPUT_TYPES(cls):
         return {
             "required": {
                 "image": ("IMAGE",),
-                "alpha_matting": (["True","False"], {"default":"False"},),
-                "alpha_matting_foreground_threshold": ("INT", {"default":240, "min": 0, "max": 255},),
-                "alpha_matting_background_threshold": ("INT", {"default":10, "min": 0, "max": 255},),
-                "alpha_matting_erode_size": ("INT", {"default":10, "min": 0, "max": 255},),
-                "post_process_mask": (["True","False"], {"default":"False"},),
-                "bgcolor": ("COLOR", {"default":"black"},),
-                
+                "alpha_matting": (
+                    "BOOLEAN",
+                    {"default": False},
+                ),
+                "alpha_matting_foreground_threshold": (
+                    "INT",
+                    {"default": 240, "min": 0, "max": 255},
+                ),
+                "alpha_matting_background_threshold": (
+                    "INT",
+                    {"default": 10, "min": 0, "max": 255},
+                ),
+                "alpha_matting_erode_size": (
+                    "INT",
+                    {"default": 10, "min": 0, "max": 255},
+                ),
+                "post_process_mask": (
+                    "BOOLEAN",
+                    {"default": False},
+                ),
+                "bgcolor": (
+                    "COLOR",
+                    {"default": "#000000"},
+                ),
             },
         }
 
-    RETURN_TYPES = ("IMAGE","MASK","IMAGE",)
-    RETURN_NAMES = ("Image (rgba)","Mask","Image",)
+    RETURN_TYPES = (
+        "IMAGE",
+        "MASK",
+        "IMAGE",
+    )
+    RETURN_NAMES = (
+        "Image (rgba)",
+        "Mask",
+        "Image",
+    )
     FUNCTION = "remove_background"
-    CATEGORY = "image"
+    CATEGORY = "mtb/image"
+
     # bgcolor: Optional[Tuple[int, int, int, int]]
-    def remove_background(self, image, alpha_matting, alpha_matting_foreground_threshold, alpha_matting_background_threshold, alpha_matting_erode_size, post_process_mask, bgcolor):
-        image = remove(
-                data=tensor2pil(image),
-                alpha_matting=alpha_matting == "True",
+    def remove_background(
+        self,
+        image,
+        alpha_matting,
+        alpha_matting_foreground_threshold,
+        alpha_matting_background_threshold,
+        alpha_matting_erode_size,
+        post_process_mask,
+        bgcolor,
+    ):
+        pbar = comfy.utils.ProgressBar(image.size(0))
+        images = tensor2pil(image)
+
+        out_img = []
+        out_mask = []
+        out_img_on_bg = []
+
+        for img in images:
+            img_rm = remove(
+                data=img,
+                alpha_matting=alpha_matting,
                 alpha_matting_foreground_threshold=alpha_matting_foreground_threshold,
                 alpha_matting_background_threshold=alpha_matting_background_threshold,
                 alpha_matting_erode_size=alpha_matting_erode_size,
                 session=None,
                 only_mask=False,
-                post_process_mask=post_process_mask == "True",
-                bgcolor=None
+                post_process_mask=post_process_mask,
+                bgcolor=None,
             )
-        
-        
-        # extract the alpha to a new image
-        mask = image.getchannel(3)
-        
-        # add our bgcolor behind the image
-        image_on_bg = Image.new("RGBA", image.size, bgcolor)
-        
-        image_on_bg.paste(image, mask=mask)
-        
-        
-        return (pil2tensor(image), pil2tensor(mask), pil2tensor(image_on_bg))
+
+            # extract the alpha to a new image
+            mask = img_rm.getchannel(3)
+
+            # add our bgcolor behind the image
+            image_on_bg = Image.new("RGBA", img_rm.size, bgcolor)
+
+            image_on_bg.paste(img_rm, mask=mask)
+
+            image_on_bg = image_on_bg.convert("RGB")
+
+            out_img.append(img_rm)
+            out_mask.append(mask)
+            out_img_on_bg.append(image_on_bg)
+
+            pbar.update(1)
+
+        return (pil2tensor(out_img), pil2tensor(out_mask), pil2tensor(out_img_on_bg))
 
 
 __nodes__ = [
     ImageRemoveBackgroundRembg,
-]
+]

nodes/model.py

@@ -0,0 +1,99 @@
+import copy
+
+import torch
+
+from ..log import log
+
+
+class VaeDecode_:
+    """Wrapper for the 2 core decoders but also adding the sd seamless hack, taken from: FlyingFireCo/tiled_ksampler"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "samples": ("LATENT",),
+                "vae": ("VAE",),
+                "seamless_model": ("BOOLEAN", {"default": False}),
+                "use_tiling_decoder": ("BOOLEAN", {"default": True}),
+                "tile_size": (
+                    "INT",
+                    {"default": 512, "min": 320, "max": 4096, "step": 64},
+                ),
+            }
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "decode"
+
+    CATEGORY = "mtb/decode"
+
+    def decode(
+        self, vae, samples, seamless_model, use_tiling_decoder=True, tile_size=512
+    ):
+        if seamless_model:
+            if use_tiling_decoder:
+                log.error(
+                    "You cannot use seamless mode with tiling decoder together, skipping tiling."
+                )
+                use_tiling_decoder = False
+            for layer in [
+                layer
+                for layer in vae.first_stage_model.modules()
+                if isinstance(layer, torch.nn.Conv2d)
+            ]:
+                layer.padding_mode = "circular"
+        if use_tiling_decoder:
+            return (
+                vae.decode_tiled(
+                    samples["samples"],
+                    tile_x=tile_size // 8,
+                    tile_y=tile_size // 8,
+                ),
+            )
+        else:
+            return (vae.decode(samples["samples"]),)
+
+
+class ModelPatchSeamless:
+    """Uses the stable diffusion 'hack' to infer seamless images by setting the model layers padding mode to circular (experimental)"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "model": ("MODEL",),
+                "tiling": (
+                    "BOOLEAN",
+                    {"default": True},
+                ),  # kept for testing not sure why it should be false
+            }
+        }
+
+    RETURN_TYPES = ("MODEL", "MODEL")
+    RETURN_NAMES = (
+        "Original Model (passthrough)",
+        "Patched Model",
+    )
+    FUNCTION = "hack"
+
+    CATEGORY = "mtb/textures"
+
+    def apply_circular(self, model, enable):
+        for layer in [
+            layer for layer in model.modules() if isinstance(layer, torch.nn.Conv2d)
+        ]:
+            layer.padding_mode = "circular" if enable else "zeros"
+        return model
+
+    def hack(
+        self,
+        model,
+        tiling,
+    ):
+        hacked_model = copy.deepcopy(model)
+        self.apply_circular(hacked_model.model, tiling)
+        return (model, hacked_model)
+
+
+__nodes__ = [ModelPatchSeamless, VaeDecode_]

nodes/number.py

@@ -1,26 +1,87 @@
-class IntToNumber:
-    """Node addon for the WAS Suite. Converts a "comfy" INT to a NUMBER."""
-
-    def __init__(self):
-        pass
+class IntToBool:
+    """Basic int to bool conversion"""
 
     @classmethod
     def INPUT_TYPES(cls):
         return {
             "required": {
-                "int": ("INT", {"default": 0, "min": 0, "max": 1e9, "step": 1}),
+                "int": (
+                    "INT",
+                    {
+                        "default": 0,
+                    },
+                ),
+            }
+        }
+
+    RETURN_TYPES = ("BOOLEAN",)
+    FUNCTION = "int_to_bool"
+    CATEGORY = "mtb/number"
+
+    def int_to_bool(self, int):
+        return (bool(int),)
+
+
+class IntToNumber:
+    """Node addon for the WAS Suite. Converts a "comfy" INT to a NUMBER."""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "int": (
+                    "INT",
+                    {
+                        "default": 0,
+                        "min": -1e9,
+                        "max": 1e9,
+                        "step": 1,
+                        "forceInput": True,
+                    },
+                ),
             }
         }
 
     RETURN_TYPES = ("NUMBER",)
     FUNCTION = "int_to_number"
-    CATEGORY = "number"
+    CATEGORY = "mtb/number"
 
     def int_to_number(self, int):
+        return (int,)
+
+
+class FloatToNumber:
+    """Node addon for the WAS Suite. Converts a "comfy" FLOAT to a NUMBER."""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "float": (
+                    "FLOAT",
+                    {
+                        "default": 0,
+                        "min": -1e9,
+                        "max": 1e9,
+                        "step": 1,
+                        "forceInput": True,
+                    },
+                ),
+            }
+        }
+
+    RETURN_TYPES = ("NUMBER",)
+    FUNCTION = "float_to_number"
+    CATEGORY = "mtb/number"
+
+    def float_to_number(self, float):
+        return (float,)
 
         return (int,)
 
-__nodes__ = [
-    IntToNumber,
 
-]
+__nodes__ = [
+    FloatToNumber,
+    IntToBool,
+    IntToNumber,
+]

nodes/transform.py

@@ -0,0 +1,110 @@
+import torch
+import torchvision.transforms.functional as TF
+from ..utils import log, hex_to_rgb, tensor2pil, pil2tensor
+from math import sqrt, ceil
+from typing import cast
+from PIL import Image
+
+
+class TransformImage:
+    """Save torch tensors (image, mask or latent) to disk, useful to debug things outside comfy
+
+
+    it return a tensor representing the transformed images with the same shape as the input tensor
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "x": ("FLOAT", {"default": 0, "step": 1, "min": -4096, "max": 4096}),
+                "y": ("FLOAT", {"default": 0, "step": 1, "min": -4096, "max": 4096}),
+                "zoom": ("FLOAT", {"default": 1.0, "min": 0.001, "step": 0.01}),
+                "angle": ("FLOAT", {"default": 0, "step": 1, "min": -360, "max": 360}),
+                "shear": (
+                    "FLOAT",
+                    {"default": 0, "step": 1, "min": -4096, "max": 4096},
+                ),
+                "border_handling": (
+                    ["edge", "constant", "reflect", "symmetric"],
+                    {"default": "edge"},
+                ),
+                "constant_color": ("COLOR", {"default": "#000000"}),
+            },
+        }
+
+    FUNCTION = "transform"
+    RETURN_TYPES = ("IMAGE",)
+    CATEGORY = "mtb/transform"
+
+    def transform(
+        self,
+        image: torch.Tensor,
+        x: float,
+        y: float,
+        zoom: float,
+        angle: float,
+        shear: float,
+        border_handling="edge",
+        constant_color=None,
+    ):
+        x = int(x)
+        y = int(y)
+        angle = int(angle)
+
+        log.debug(f"Zoom: {zoom} | x: {x}, y: {y}, angle: {angle}, shear: {shear}")
+
+        if image.size(0) == 0:
+            return (torch.zeros(0),)
+        transformed_images = []
+        frames_count, frame_height, frame_width, frame_channel_count = image.size()
+
+        new_height, new_width = int(frame_height * zoom), int(frame_width * zoom)
+
+        log.debug(f"New height: {new_height}, New width: {new_width}")
+
+        # - Calculate diagonal of the original image
+        diagonal = sqrt(frame_width**2 + frame_height**2)
+        max_padding = ceil(diagonal * zoom - min(frame_width, frame_height))
+        # Calculate padding for zoom
+        pw = int(frame_width - new_width)
+        ph = int(frame_height - new_height)
+
+        pw += abs(max_padding)
+        ph += abs(max_padding)
+
+        padding = [max(0, pw + x), max(0, ph + y), max(0, pw - x), max(0, ph - y)]
+
+        constant_color = hex_to_rgb(constant_color)
+        log.debug(f"Fill Tuple: {constant_color}")
+
+        for img in tensor2pil(image):
+            img = TF.pad(
+                img,  # transformed_frame,
+                padding=padding,
+                padding_mode=border_handling,
+                fill=constant_color or 0,
+            )
+
+            img = cast(
+                Image.Image,
+                TF.affine(img, angle=angle, scale=zoom, translate=[x, y], shear=shear),
+            )
+
+            left = abs(padding[0])
+            upper = abs(padding[1])
+            right = img.width - abs(padding[2])
+            bottom = img.height - abs(padding[3])
+
+            # log.debug("crop is [:,top:bottom, left:right] for tensors")
+            log.debug("crop is [left, top, right, bottom] for PIL")
+            log.debug(f"crop is {left}, {upper}, {right}, {bottom}")
+            img = img.crop((left, upper, right, bottom))
+
+            transformed_images.append(img)
+
+        return (pil2tensor(transformed_images),)
+
+
+__nodes__ = [TransformImage]

nodes/video.py

@@ -10,96 +10,189 @@ from pathlib import Path
 import json
 
 from ..log import log
+
+
 class LoadImageSequence:
     """Load an image sequence from a folder. The current frame is used to determine which image to load.
 
     Usually used in conjunction with the `Primitive` node set to increment to load a sequence of images from a folder.
+    Use -1 to load all matching frames as a batch.
     """
+
     @classmethod
     def INPUT_TYPES(cls):
         return {
             "required": {
-                "path": ("STRING",{"default":"videos/####.png"}),
-                "current_frame": ("INT",{"default":0, "min":0, "max": 9999999},),
+                "path": ("STRING", {"default": "videos/####.png"}),
+                "current_frame": (
+                    "INT",
+                    {"default": 0, "min": -1, "max": 9999999},
+                ),
             }
         }
 
-    CATEGORY = "video"
+    CATEGORY = "mtb/IO"
     FUNCTION = "load_image"
-    RETURN_TYPES = ("IMAGE", "MASK", "INT",)
-    RETURN_NAMES = ("image", "mask", "current_frame",)
+    RETURN_TYPES = (
+        "IMAGE",
+        "MASK",
+        "INT",
+    )
+    RETURN_NAMES = (
+        "image",
+        "mask",
+        "current_frame",
+    )
 
     def load_image(self, path=None, current_frame=0):
+        load_all = current_frame == -1
+
+        if load_all:
+            log.debug(f"Loading all frames from {path}")
+            frames = resolve_all_frames(path)
+            log.debug(f"Found {len(frames)} frames")
+
+            imgs = []
+            masks = []
+
+            for frame in frames:
+                img, mask = img_from_path(frame)
+                imgs.append(img)
+                masks.append(mask)
+
+            out_img = torch.cat(imgs, dim=0)
+            out_mask = torch.cat(masks, dim=0)
+
+            return (
+                out_img,
+                out_mask,
+            )
+
         log.debug(f"Loading image: {path}, {current_frame}")
         print(f"Loading image: {path}, {current_frame}")
         resolved_path = resolve_path(path, current_frame)
         image_path = folder_paths.get_annotated_filepath(resolved_path)
-        i = Image.open(image_path)
-        i = ImageOps.exif_transpose(i)
-        image = i.convert("RGB")
-        image = np.array(image).astype(np.float32) / 255.0
-        image = torch.from_numpy(image)[None,]
-        if 'A' in i.getbands():
-            mask = np.array(i.getchannel('A')).astype(np.float32) / 255.0
-            mask = 1. - torch.from_numpy(mask)
-        else:
-            mask = torch.zeros((64, 64), dtype=torch.float32, device="cpu")
-        return (image, mask, current_frame,)
+        image, mask = img_from_path(image_path)
+        return (
+            image,
+            mask,
+            current_frame,
+        )
 
     @staticmethod
     def IS_CHANGED(path="", current_frame=0):
         print(f"Checking if changed: {path}, {current_frame}")
         resolved_path = resolve_path(path, current_frame)
         image_path = folder_paths.get_annotated_filepath(resolved_path)
-        if os.path.exists(image_path): 
+        if os.path.exists(image_path):
             m = hashlib.sha256()
-            with open(image_path, 'rb') as f:
+            with open(image_path, "rb") as f:
                 m.update(f.read())
             return m.digest().hex()
         return "NONE"
 
     # @staticmethod
     # def VALIDATE_INPUTS(path="", current_frame=0):
-        
+
     #     print(f"Validating inputs: {path}, {current_frame}")
     #     resolved_path = resolve_path(path, current_frame)
     #     if not folder_paths.exists_annotated_filepath(resolved_path):
     #         return f"Invalid image file: {resolved_path}"
     #     return True
 
+
+import glob
+
+
+def img_from_path(path):
+    img = Image.open(path)
+    img = ImageOps.exif_transpose(img)
+    image = img.convert("RGB")
+    image = np.array(image).astype(np.float32) / 255.0
+    image = torch.from_numpy(image)[None,]
+    if "A" in img.getbands():
+        mask = np.array(img.getchannel("A")).astype(np.float32) / 255.0
+        mask = 1.0 - torch.from_numpy(mask)
+    else:
+        mask = torch.zeros((64, 64), dtype=torch.float32, device="cpu")
+    return (
+        image,
+        mask,
+    )
+
+
+def resolve_all_frames(pattern):
+    folder_path, file_pattern = os.path.split(pattern)
+
+    log.debug(f"Resolving all frames in {folder_path}")
+    frames = []
+    hash_count = file_pattern.count("#")
+    frame_pattern = re.sub(r"#+", "*", file_pattern)
+
+    log.debug(f"Found pattern: {frame_pattern}")
+
+    matching_files = glob.glob(os.path.join(folder_path, frame_pattern))
+
+    log.debug(f"Found {len(matching_files)} matching files")
+
+    frame_regex = re.escape(file_pattern).replace(r"\#", r"(\d+)")
+
+    frame_number_regex = re.compile(frame_regex)
+
+    for file in matching_files:
+        match = frame_number_regex.search(file)
+        if match:
+            frame_number = match.group(1)
+            log.debug(f"Found frame number: {frame_number}")
+            # resolved_file = pattern.replace("*" * frame_number.count("#"), frame_number)
+            frames.append(file)
+
+    frames.sort()  # Sort frames alphabetically
+    return frames
+
+
 def resolve_path(path, frame):
     hashes = path.count("#")
     padded_number = str(frame).zfill(hashes)
     return re.sub("#+", padded_number, path)
 
+
 class SaveImageSequence:
     """Save an image sequence to a folder. The current frame is used to determine which image to save.
 
     This is merely a wrapper around the `save_images` function with formatting for the output folder and filename.
     """
+
     def __init__(self):
         self.output_dir = folder_paths.get_output_directory()
         self.type = "output"
-        
 
     @classmethod
     def INPUT_TYPES(cls):
-        return {"required": {
-                    "images": ("IMAGE", ),
-                    "filename_prefix": ("STRING", {"default": "Sequence"}),
-                    "current_frame": ("INT", {"default": 0, "min": 0, "max": 9999999}),
-                    },
-                "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
-                }
+        return {
+            "required": {
+                "images": ("IMAGE",),
+                "filename_prefix": ("STRING", {"default": "Sequence"}),
+                "current_frame": ("INT", {"default": 0, "min": 0, "max": 9999999}),
+            },
+            "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
+        }
 
     RETURN_TYPES = ()
     FUNCTION = "save_images"
 
     OUTPUT_NODE = True
 
-    CATEGORY = "image"
+    CATEGORY = "mtb/IO"
 
-    def save_images(self, images, filename_prefix="Sequence", current_frame=0, prompt=None, extra_pnginfo=None):
+    def save_images(
+        self,
+        images,
+        filename_prefix="Sequence",
+        current_frame=0,
+        prompt=None,
+        extra_pnginfo=None,
+    ):
         # full_output_folder, filename, counter, subfolder, filename_prefix = folder_paths.get_save_image_path(filename_prefix, self.output_dir, images[0].shape[1], images[0].shape[0])
         # results = list()
         # for image in images:
@@ -120,30 +213,39 @@ class SaveImageSequence:
         #         "type": self.type
         #     })
         #     counter += 1
-    
+
         if len(images) > 1:
             raise ValueError("Can only save one image at a time")
-        
+
         resolved_path = Path(self.output_dir) / filename_prefix
         resolved_path.mkdir(parents=True, exist_ok=True)
-        
+
         resolved_img = resolved_path / f"{filename_prefix}_{current_frame:05}.png"
-        
+
         output_image = images[0].cpu().numpy()
-        img = Image.fromarray(np.clip(output_image * 255., 0, 255).astype(np.uint8))
+        img = Image.fromarray(np.clip(output_image * 255.0, 0, 255).astype(np.uint8))
         metadata = PngInfo()
         if prompt is not None:
             metadata.add_text("prompt", json.dumps(prompt))
         if extra_pnginfo is not None:
             for x in extra_pnginfo:
                 metadata.add_text(x, json.dumps(extra_pnginfo[x]))
-                
+
         img.save(resolved_img, pnginfo=metadata, compress_level=4)
-        return { "ui": { "images": [ { "filename": resolved_img.name, "subfolder": resolved_path.name, "type": self.type } ] } }
-                
-                
-        
+        return {
+            "ui": {
+                "images": [
+                    {
+                        "filename": resolved_img.name,
+                        "subfolder": resolved_path.name,
+                        "type": self.type,
+                    }
+                ]
+            }
+        }
+
+
 __nodes__ = [
     LoadImageSequence,
     SaveImageSequence,
-]
+]

pyrightconfig.json

@@ -0,0 +1,18 @@
+{
+    "exclude": [
+        "**/node_modules",
+        "**/__pycache__",
+    ],
+    "ignore": [
+        "extern"
+    ],
+    "defineConstant": {
+        "DEBUG": true
+    },
+    "venvPath": "../../../.venv/",
+    "reportMissingImports": true,
+    "reportMissingTypeStubs": false,
+    "pythonVersion": "3.10",
+    "pythonPlatform": "All",
+    "reportOptionalMemberAccess": "none"
+}

requirements-wheels.txt

@@ -1,3 +0,0 @@
-insightface==0.7.3
-mmcv==2.0.0
-mmdet==3.0.0

requirements.txt

@@ -1,9 +1,10 @@
-onnxruntime-gpu
-imageio
 qrcode[pil]
-numpy==1.23.5
-ifnude==0.0.3
-insightface==0.7.3
-mmcv==2.0.0
-mmdet==3.0.0
-rembg==2.0.37
+onnxruntime-gpu
+requirements-parser
+# opencv-contrib
+rembg
+imageio_ffmpeg
+rich
+rich_argparse
+librosa
+torchaudio

scripts/a111_extract.py

@@ -0,0 +1,112 @@
+from pathlib import Path
+from PIL import Image
+from PIL.PngImagePlugin import PngImageFile, PngInfo
+import json
+from pprint import pprint
+import argparse
+from rich.console import Console
+from rich.progress import Progress
+from rich_argparse import RichHelpFormatter
+
+
+def parse_a111(params, verbose=False):
+    # params = [p.split(": ") for p in params.split("\n")]
+    params = params.split("\n")
+
+    prompt = params[0].strip()
+    neg = params[1].split(":")[1].strip()
+
+    settings = {}
+    try:
+        settings = {
+            s.split(":")[0].strip(): s.split(":")[1].strip()
+            for s in params[2].split(",")
+        }
+
+    except IndexError:
+        settings = {"raw": params[2].strip()}
+
+    if verbose:
+        print(f"PROMPT: {prompt}")
+        print(f"NEG: {neg}")
+        print("SETTINGS:")
+        pprint(settings, indent=4)
+
+    return {"prompt": prompt, "negative": neg, "settings": settings}
+
+
+import glob
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Crude metadata extractor from A111 pngs",
+        formatter_class=RichHelpFormatter
+    )
+    parser.add_argument("inputs", nargs="*", help="Input image files")
+    parser.add_argument("--output", help="Output JSON file")
+    parser.add_argument("-v", "--verbose", action="store_true", help="Verbose mode")
+    parser.add_argument(
+        "--glob", help="Enable glob pattern matching", metavar="PATTERN"
+    )
+
+    args = parser.parse_args()
+
+    # - checks
+    if not args.glob and not args.inputs:
+        parser.error("Either --glob flag or inputs must be provided.")
+    if args.glob:
+        glob_pattern = args.glob
+        try:
+            pattern_path = str(Path(glob_pattern).expanduser().resolve())
+
+            if not any(glob.glob(pattern_path)):
+                raise ValueError(f"No files found for glob pattern: {glob_pattern}")
+        except Exception as e:
+            console = Console()
+            console.print(
+                f"[bold red]Error: Invalid glob pattern '{glob_pattern}': {e}[/bold red]"
+            )
+
+            exit(1)
+    else:
+        glob_pattern = None
+
+    input_files = []
+
+    if glob_pattern:
+        input_files = list(glob.glob(str(Path(glob_pattern).expanduser().resolve())))
+    else:
+        input_files = [Path(p) for p in args.inputs]
+
+    console = Console()
+    console.print("Input Files:", style="bold", end=" ")
+    console.print(f"{len(input_files):03d} files", style="cyan")
+    # for input_file in args.inputs:
+    #     console.print(f"- {input_file}", style="cyan")
+    console.print("\nOutput File:", style="bold", end=" ")
+    console.print(f"{Path(args.output).resolve().absolute()}", style="cyan")
+
+    with Progress(console=console, auto_refresh=True) as progress:
+        # files = Path(pth).rglob("*.png")
+        unique_info = {}
+        last = None
+
+        task = progress.add_task("[cyan]Extracting meta...", total=len(input_files) + 1)
+        for p in input_files:
+            im = Image.open(p)
+            parsed = parse_a111(im.info["parameters"], args.verbose)
+
+            if parsed != last:
+                unique_info[Path(p).stem] = parsed
+
+            last = parsed
+            progress.update(task, advance=1)
+            progress.refresh()
+
+        unique_info = json.dumps(unique_info, indent=4)
+        with open(args.output, "w") as f:
+            f.write(unique_info)
+            progress.update(task, advance=1)
+            progress.refresh()
+
+    console.print("\nProcessing completed!", style="bold green")

scripts/comfy_meta.py

@@ -0,0 +1,213 @@
+import argparse
+import json
+from PIL import Image, PngImagePlugin
+from rich.console import Console
+from rich import print
+from rich_argparse import RichHelpFormatter
+import os
+from pathlib import Path
+
+console = Console()
+
+# BNK_CutoffSetRegions
+# BNK_CutoffRegionsToConditioning
+# BNK_CutoffBasePrompt
+
+
+# Extracts metadata from a PNG image and returns it as a dictionary
+def extract_metadata(image_path):
+    image = Image.open(image_path)
+    prompt = image.info.get("prompt", "")
+    workflow = image.info.get("workflow", "")
+
+    if workflow:
+        workflow = json.loads(workflow)
+
+    if prompt:
+        prompt = json.loads(prompt)
+
+    console.print(f"Metadata extracted from [cyan]{image_path}[/cyan].")
+
+    return {
+        "prompt": prompt,
+        "workflow": workflow,
+    }
+
+
+# Embeds metadata into a PNG image
+def embed_metadata(image_path, metadata):
+    image = Image.open(image_path)
+    o_metadata = image.info
+
+    pnginfo = PngImagePlugin.PngInfo()
+    if prompt := metadata.get("prompt"):
+        pnginfo.add_text("prompt", json.dumps(prompt))
+    elif "prompt" in o_metadata:
+        pnginfo.add_text("prompt", o_metadata["prompt"])
+
+    if workflow := metadata.get("workflow"):
+        pnginfo.add_text("workflow", json.dumps(workflow))
+    elif "workflow" in o_metadata:
+        pnginfo.add_text("workflow", o_metadata["workflow"])
+
+    imgp = Path(image_path)
+    output = imgp.with_stem(f"{imgp.stem}_comfy_embed")
+    index = 1
+    while output.exists():
+        output = imgp.with_stem(f"{imgp.stem}_{index}_comfy_embed").with_suffix(".png")
+        index += 1
+
+    image.save(output, pnginfo=pnginfo)
+    console.print(f"Metadata embedded into [cyan]{output}[/cyan].")
+
+
+# CLI subcommand: extract
+def extract(args):
+    input_files = []
+    for input_path in args.input:
+        if os.path.isdir(input_path):
+            folder_path = input_path
+            input_files.extend(
+                [
+                    os.path.join(folder_path, file_name)
+                    for file_name in os.listdir(folder_path)
+                    if file_name.lower().endswith((".png", ".jpg", ".jpeg"))
+                ]
+            )
+        else:
+            input_files.append(input_path)
+
+    if len(input_files) == 1:
+        metadata = extract_metadata(input_files[0])
+        if args.print_output:
+            print(json.dumps(metadata, indent=4))
+        else:
+            if not args.output:
+                output = Path(input_files[0]).with_suffix(".json")
+                index = 1
+                while output.exists():
+                    output = (
+                        Path(input_files[0])
+                        .with_stem(f"{Path(input_files[0]).stem}_{index}")
+                        .with_suffix(".json")
+                    )
+                    index += 1
+            else:
+                output = args.output
+            with open(output, "w") as file:
+                json.dump(metadata, file, indent=4)
+            console.print(f"Metadata extracted and saved to [cyan]{output}[/cyan].")
+    else:
+        metadata_dict = {}
+        for input_file in input_files:
+            metadata = extract_metadata(input_file)
+            filename = os.path.basename(input_file)
+            output = (
+                Path(args.output) / f"{filename}.json"
+                if args.output
+                else Path(input_file).with_suffix(".json")
+            )
+            index = 1
+            while output.exists():
+                output = Path(args.output).parent / f"{filename}_{index}.json"
+                index += 1
+            with open(output, "w") as file:
+                json.dump(metadata, file, indent=4)
+            metadata_dict[filename] = metadata
+        if args.output:
+            with open(args.output, "w") as file:
+                json.dump(metadata_dict, file, indent=4)
+            console.print(
+                f"Metadata extracted and saved to [cyan]{args.output}[/cyan]."
+            )
+        else:
+            console.print("Multiple metadata files created.")
+
+
+# CLI subcommand: embed
+def embed(args):
+    input_files = []
+    for input_path in args.input:
+        if os.path.isdir(input_path):
+            folder_path = input_path
+            input_files.extend(
+                [
+                    os.path.join(folder_path, file_name)
+                    for file_name in os.listdir(folder_path)
+                    if file_name.lower().endswith(".json")
+                ]
+            )
+        else:
+            input_files.append(input_path)
+
+    for input_file in input_files:
+        with open(input_file) as file:
+            metadata = json.load(file)
+        image_path = input_file.replace(".json", ".png")
+        if args.output:
+            output_dir = args.output
+            if os.path.isdir(output_dir):
+                output_path = os.path.join(output_dir, os.path.basename(image_path))
+                index = 1
+                while os.path.exists(output_path):
+                    output_path = os.path.join(
+                        output_dir,
+                        f"{os.path.basename(image_path)}_{index}.png",
+                    )
+                    index += 1
+            else:
+                output_path = output_dir
+        else:
+            output_path = image_path.replace(".png", "_comfy_embed.png")
+
+        embed_metadata(image_path, metadata)
+        # os.rename(image_path, output_path)
+        console.print(f"Metadata embedded into [cyan]{output_path}[/cyan].")
+
+
+if __name__ == "__main__":
+    # Create the main CLI parser
+    parser = argparse.ArgumentParser(
+        prog="image-metadata-cli", formatter_class=RichHelpFormatter
+    )
+    subparsers = parser.add_subparsers(title="subcommands")
+
+    # Parser for the "extract" subcommand
+    extract_parser = subparsers.add_parser(
+        "extract",
+        help="Extract metadata from PNG image(s) or folder",
+        formatter_class=RichHelpFormatter,
+    )
+    extract_parser.add_argument(
+        "input", nargs="+", help="Input PNG image file(s) or folder path"
+    )
+    extract_parser.add_argument(
+        "--print",
+        dest="print_output",
+        action="store_true",
+        help="Print the output to stdout",
+    )
+    extract_parser.add_argument("--output", help="Output JSON file(s) or directory")
+    extract_parser.set_defaults(func=extract)
+
+    # Parser for the "embed" subcommand
+    embed_parser = subparsers.add_parser(
+        "embed",
+        help="Embed metadata into PNG image(s) or folder",
+        formatter_class=RichHelpFormatter,
+    )
+    embed_parser.add_argument(
+        "input", nargs="+", help="Input JSON file(s) or folder path"
+    )
+    embed_parser.add_argument("--output", help="Output PNG image file(s) or directory")
+    embed_parser.set_defaults(func=embed)
+
+    # Parse the command-line arguments and execute the appropriate subcommand
+    args = parser.parse_args()
+    if hasattr(args, "func"):
+        try:
+            args.func(args)
+        except ValueError as e:
+            console.print(f"[bold red]Error:[/bold red] {str(e)}")
+    else:
+        parser.print_help()

scripts/download_models.py

@@ -2,6 +2,8 @@ import os
 import requests
 from rich.console import Console
 from tqdm import tqdm
+import subprocess
+import sys
 
 try:
     import folder_paths
@@ -26,6 +28,25 @@ models_to_download = {
         ],
         "destination": "insightface",
     },
+    "GFPGAN (face enhancement)": {
+        "size": 332,
+        "download_url": [
+            "https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth",
+            "https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.4.pth"
+            # TODO: provide a way to selectively download models from "packs"
+            # https://github.com/TencentARC/GFPGAN/releases/download/v0.1.0/GFPGANv1.pth
+            # https://github.com/TencentARC/GFPGAN/releases/download/v0.2.0/GFPGANCleanv1-NoCE-C2.pth
+            # https://github.com/TencentARC/GFPGAN/releases/download/v1.3.4/RestoreFormer.pth
+        ],
+        "destination": "face_restore",
+    },
+    "FILM: Frame Interpolation for Large Motion": {
+        "size": 402,
+        "download_url": [
+            "https://drive.google.com/drive/folders/131_--QrieM4aQbbLWrUtbO2cGbX8-war"
+        ],
+        "destination": "FILM",
+    },
 }
 
 console = Console()
@@ -41,6 +62,35 @@ def download_model(download_url, destination):
         return
 
     filename = os.path.basename(urlparse(download_url).path)
+    response = None
+    if "drive.google.com" in download_url:
+        try:
+            import gdown
+        except ImportError:
+            print("Installing gdown")
+            subprocess.check_call(
+                [
+                    sys.executable,
+                    "-m",
+                    "pip",
+                    "install",
+                    "git+https://github.com/melMass/gdown@main",
+                ]
+            )
+            import gdown
+
+        if "/folders/" in download_url:
+            # download folder
+            try:
+                gdown.download_folder(download_url, output=destination, resume=True)
+            except TypeError:
+                gdown.download_folder(download_url, output=destination)
+
+            return
+        # download from google drive
+        gdown.download(download_url, destination, quiet=False, resume=True)
+        return
+
     response = requests.get(download_url, stream=True)
     total_size = int(response.headers.get("content-length", 0))
 
@@ -93,7 +143,7 @@ def handle_interrupt():
     console.print("Interrupted by user.", style="bold red")
 
 
-def main(models_to_download):
+def main(models_to_download, skip_input=False):
     try:
         models_to_download_selected = {}
 
@@ -129,13 +179,16 @@ def main(models_to_download):
             console.print("No new models to download.")
             return
 
-        models_to_download_selected = ask_user_for_downloads(
-            models_to_download_selected
+        models_to_download_selected = (
+            ask_user_for_downloads(models_to_download_selected)
+            if not skip_input
+            else models_to_download_selected
         )
 
         for model_name, model_details in models_to_download_selected.items():
             download_url = model_details["download_url"]
             destination = model_details["destination"]
+            console.print(f"Downloading {model_name}...")
             download_model(download_url, destination)
 
     except KeyboardInterrupt:
@@ -143,4 +196,10 @@ def main(models_to_download):
 
 
 if __name__ == "__main__":
-    main(models_to_download)
+    import argparse
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("-y", "--yes", action="store_true", help="skip user input")
+
+    args = parser.parse_args()
+    main(models_to_download, args.yes)

scripts/interpolate_frames.py

@@ -0,0 +1,142 @@
+import glob
+from pathlib import Path
+import uuid
+import sys
+from typing import List
+
+sys.path.append((Path(__file__).parent / "extern").as_posix())
+
+
+import argparse
+from rich_argparse import RichHelpFormatter
+from rich.console import Console
+from rich.progress import Progress
+
+import numpy as np
+import subprocess
+
+
+def write_prores_444_video(output_file, frames: List[np.ndarray], fps):
+    # Convert float images to the range of 0-65535 (12-bit color depth)
+    frames = [(frame * 65535).clip(0, 65535).astype(np.uint16) for frame in frames]
+
+    height, width, _ = frames[0].shape
+
+    # Prepare the FFmpeg command
+    command = [
+        "ffmpeg",
+        "-y",  # Overwrite output file if it already exists
+        "-f",
+        "rawvideo",
+        "-vcodec",
+        "rawvideo",
+        "-s",
+        f"{width}x{height}",
+        "-pix_fmt",
+        "rgb48le",
+        "-r",
+        str(fps),
+        "-i",
+        "-",
+        "-c:v",
+        "prores_ks",
+        "-profile:v",
+        "4",
+        "-pix_fmt",
+        "yuva444p10le",
+        "-r",
+        str(fps),
+        "-y",  # Overwrite output file if it already exists
+        output_file,
+    ]
+
+    process = subprocess.Popen(command, stdin=subprocess.PIPE)
+
+    for frame in frames:
+        process.stdin.write(frame.tobytes())
+
+    process.stdin.close()
+    process.wait()
+
+
+if __name__ == "__main__":
+    default_output = f"./output_{uuid.uuid4()}.mov"
+    parser = argparse.ArgumentParser(
+        description="FILM frame interpolation", formatter_class=RichHelpFormatter
+    )
+    parser.add_argument("inputs", nargs="*", help="Input image files")
+    parser.add_argument("--output", help="Output JSON file", default=default_output)
+    parser.add_argument("-v", "--verbose", action="store_true", help="Verbose mode")
+    parser.add_argument(
+        "--glob", help="Enable glob pattern matching", metavar="PATTERN"
+    )
+    parser.add_argument(
+        "--interpolate", type=int, default=4, help="Time for interpolated frames"
+    )
+    parser.add_argument("--fps", type=int, default=30, help="Out FPS")
+    align = 64
+    block_width = 2
+    block_height = 2
+
+    args = parser.parse_args()
+
+    # - checks
+    if not args.glob and not args.inputs:
+        parser.error("Either --glob flag or inputs must be provided.")
+    if args.glob:
+        glob_pattern = args.glob
+        try:
+            pattern_path = str(Path(glob_pattern).expanduser().resolve())
+
+            if not any(glob.glob(pattern_path)):
+                raise ValueError(f"No files found for glob pattern: {glob_pattern}")
+        except Exception as e:
+            console = Console()
+            console.print(
+                f"[bold red]Error: Invalid glob pattern '{glob_pattern}': {e}[/bold red]"
+            )
+
+            exit(1)
+    else:
+        glob_pattern = None
+
+    input_files: List[Path] = []
+
+    if glob_pattern:
+        input_files = [
+            Path(p)
+            for p in list(glob.glob(str(Path(glob_pattern).expanduser().resolve())))
+        ]
+    else:
+        input_files = [Path(p) for p in args.inputs]
+
+    console = Console()
+    console.print("Input Files:", style="bold", end=" ")
+    console.print(f"{len(input_files):03d} files", style="cyan")
+    # for input_file in args.inputs:
+    #     console.print(f"- {input_file}", style="cyan")
+    console.print("\nOutput File:", style="bold", end=" ")
+    console.print(f"{Path(args.output).resolve().absolute()}", style="cyan")
+
+    with Progress(console=console, auto_refresh=True) as progress:
+        from frame_interpolation.eval import util
+        from frame_interpolation.eval import util, interpolator
+
+        # files = Path(pth).rglob("*.png")
+
+        model = interpolator.Interpolator(
+            "G:/MODELS/FILM/pretrained_models/film_net/Style", None
+        )  # [2,2]
+
+        task = progress.add_task("[cyan]Interpolating frames...", total=1)
+
+        frames = list(
+            util.interpolate_recursively_from_files(
+                [x.as_posix() for x in input_files], args.interpolate, model
+            )
+        )
+
+        # mediapy.write_video(args.output, frames, fps=args.fps)
+        write_prores_444_video(args.output, frames, fps=args.fps)
+        progress.update(task, advance=1)
+        progress.refresh()

utils.py

@@ -1,60 +1,753 @@
-from PIL import Image
-import numpy as np
-import torch
-from pathlib import Path
+import contextlib
+import functools
+import math
+import os
+import shlex
+import shutil
+import socket
+import subprocess
 import sys
+import uuid
+from pathlib import Path
+from typing import List, Optional, Union
+
+import folder_paths
+import numpy as np
+import requests
+import torch
+from PIL import Image
+
+from .install import pip_map
+
+try:
+    from .log import log
+except ImportError:
+    try:
+        from log import log
+
+        log.warn("Imported log without relative path")
+    except ImportError:
+        import logging
+
+        log = logging.getLogger("comfy mtb utils")
+        log.warn("[comfy mtb] You probably called the file outside a module.")
+
+
+# region SANITY_CHECK Utilities
+
+
+def make_report():
+    pass
+
+
+# endregion
+
+
+# region SERVER Utilities
+class IPChecker:
+    def __init__(self):
+        self.ips = list(self.get_local_ips())
+        log.debug(f"Found {len(self.ips)} local ips")
+        self.checked_ips = set()
+
+    def get_working_ip(self, test_url_template):
+        for ip in self.ips:
+            if ip not in self.checked_ips:
+                self.checked_ips.add(ip)
+                test_url = test_url_template.format(ip)
+                if self._test_url(test_url):
+                    return ip
+        return None
+
+    @staticmethod
+    def get_local_ips(prefix="192.168."):
+        hostname = socket.gethostname()
+        log.debug(f"Getting local ips for {hostname}")
+        for info in socket.getaddrinfo(hostname, None):
+            # Filter out IPv6 addresses if you only want IPv4
+            log.debug(info)
+            # if info[1] == socket.SOCK_STREAM and
+            if info[0] == socket.AF_INET and info[4][0].startswith(prefix):
+                yield info[4][0]
+
+    def _test_url(self, url):
+        try:
+            response = requests.get(url)
+            return response.status_code == 200
+        except Exception:
+            return False
+
+
+@functools.lru_cache(maxsize=1)
+def get_server_info():
+    from comfy.cli_args import args
+
+    ip_checker = IPChecker()
+    base_url = args.listen
+    if base_url == "0.0.0.0":
+        log.debug("Server set to 0.0.0.0, we will try to resolve the host IP")
+        base_url = ip_checker.get_working_ip(f"http://{{}}:{args.port}/history")
+        log.debug(f"Setting ip to {base_url}")
+    return (base_url, args.port)
+
+
+# endregion
+
+
+# region MISC Utilities
+def backup_file(
+    fp: Path,
+    target: Optional[Path] = None,
+    backup_dir: str = ".bak",
+    suffix: Optional[str] = None,
+    prefix: Optional[str] = None,
+):
+    if not fp.exists():
+        raise FileNotFoundError(f"No file found at {fp}")
+
+    backup_directory = target or fp.parent / backup_dir
+    backup_directory.mkdir(parents=True, exist_ok=True)
+
+    stem = fp.stem
+
+    if suffix or prefix:
+        new_stem = f"{prefix or ''}{stem}{suffix or ''}"
+    else:
+        new_stem = f"{stem}_{uuid.uuid4()}"
+
+    backup_file_path = backup_directory / f"{new_stem}{fp.suffix}"
+
+    # Perform the backup
+    shutil.copy(fp, backup_file_path)
+    log.debug(f"File backed up to {backup_file_path}")
+
+
+def hex_to_rgb(hex_color):
+    try:
+        hex_color = hex_color.lstrip("#")
+        return tuple(int(hex_color[i : i + 2], 16) for i in (0, 2, 4))
+    except ValueError:
+        log.error(f"Invalid hex color: {hex_color}")
+        return (0, 0, 0)
+
 
 def add_path(path, prepend=False):
-    
     if isinstance(path, list):
         for p in path:
             add_path(p, prepend)
         return
-    
+
     if isinstance(path, Path):
         path = path.resolve().as_posix()
-    
+
     if path not in sys.path:
         if prepend:
             sys.path.insert(0, path)
         else:
             sys.path.append(path)
-    
-    
-# Get the absolute path of the parent directory of the current script
-here = Path(__file__).parent.resolve()
 
-# Construct the absolute path to the ComfyUI directory
-comfy_dir = here.parent.parent
 
-# Construct the path to the font file
+def run_command(cmd, ignored_lines_start=None):
+    if ignored_lines_start is None:
+        ignored_lines_start = []
+
+    if isinstance(cmd, str):
+        shell_cmd = cmd
+    elif isinstance(cmd, list):
+        shell_cmd = " ".join(
+            arg.as_posix() if isinstance(arg, Path) else shlex.quote(str(arg))
+            for arg in cmd
+        )
+    else:
+        raise ValueError(
+            "Invalid 'cmd' argument. It must be a string or a list of arguments."
+        )
+
+    try:
+        _run_command(shell_cmd, ignored_lines_start)
+    except subprocess.CalledProcessError as e:
+        print(f"Command failed with return code: {e.returncode}", file=sys.stderr)
+        print(e.stderr.strip(), file=sys.stderr)
+
+    except KeyboardInterrupt:
+        print("Command execution interrupted.")
+
+
+def _run_command(shell_cmd, ignored_lines_start):
+    log.debug(f"Running {shell_cmd}")
+
+    result = subprocess.run(
+        shell_cmd,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
+        text=True,
+        shell=True,
+        check=True,
+    )
+
+    stdout_lines = result.stdout.strip().split("\n")
+    stderr_lines = result.stderr.strip().split("\n")
+
+    # Print stdout, skipping ignored lines
+    for line in stdout_lines:
+        if not any(line.startswith(ign) for ign in ignored_lines_start):
+            print(line)
+
+    # Print stderr
+    for line in stderr_lines:
+        print(line, file=sys.stderr)
+
+    print("Command executed successfully!")
+
+
+# todo use the requirements library
+reqs_map = {value: key for key, value in pip_map.items()}
+
+import importlib
+
+
+def import_install(package_name):
+    package_spec = reqs_map.get(package_name, package_name)
+
+    try:
+        importlib.import_module(package_name)
+
+    except Exception:  # (ImportError, ModuleNotFoundError):
+        run_command(
+            [Path(sys.executable).as_posix(), "-m", "pip", "install", package_spec]
+        )
+        importlib.import_module(package_name)
+
+
+# endregion
+
+
+# region GLOBAL VARIABLES
+# - detect mode
+comfy_mode = None
+if os.environ.get("COLAB_GPU"):
+    comfy_mode = "colab"
+elif "python_embeded" in sys.executable:
+    comfy_mode = "embeded"
+elif ".venv" in sys.executable:
+    comfy_mode = "venv"
+
+# - Get the absolute path of the parent directory of the current script
+here = Path(__file__).parent.absolute()
+
+# - Construct the absolute path to the ComfyUI directory
+comfy_dir = Path(folder_paths.base_path)
+models_dir = Path(folder_paths.models_dir)
+styles_dir = comfy_dir / "styles"
+audioInputDir = comfy_dir / "input" / "audio"
+
+# - Construct the path to the font file
 font_path = here / "font.ttf"
 
-# Add extern folder to path
-add_path(here / "extern")
-add_path(here / "extern" / "SadTalker")
+# - Add extern folder to path
+extern_root = here / "extern"
+add_path(extern_root)
+for pth in extern_root.iterdir():
+    if pth.is_dir():
+        add_path(pth)
 
-# Add the ComfyUI directory and custom nodes path to the sys.path list
+# - Add the ComfyUI directory and custom nodes path to the sys.path list
 add_path(comfy_dir)
 add_path((comfy_dir / "custom_nodes"))
 
-# Tensor to PIL (grabbed from WAS Suite)
-def tensor2pil(image: torch.Tensor) -> Image.Image:
-    return Image.fromarray(
-        np.clip(255.0 * image.cpu().numpy().squeeze(), 0, 255).astype(np.uint8)
-    )
+PIL_FILTER_MAP = {
+    "nearest": Image.Resampling.NEAREST,
+    "box": Image.Resampling.BOX,
+    "bilinear": Image.Resampling.BILINEAR,
+    "hamming": Image.Resampling.HAMMING,
+    "bicubic": Image.Resampling.BICUBIC,
+    "lanczos": Image.Resampling.LANCZOS,
+}
+# endregion
 
 
-# Convert PIL to Tensor (grabbed from WAS Suite)
-def pil2tensor(image: Image.Image) -> torch.Tensor:
+# region TENSOR Utilities
+def tensor2pil(image: torch.Tensor) -> List[Image.Image]:
+    batch_count = image.size(0) if len(image.shape) > 3 else 1
+    if batch_count > 1:
+        out = []
+        for i in range(batch_count):
+            out.extend(tensor2pil(image[i]))
+        return out
+
+    return [
+        Image.fromarray(
+            np.clip(255.0 * image.cpu().numpy().squeeze(), 0, 255).astype(np.uint8)
+        )
+    ]
+
+
+def pil2tensor(image: Union[Image.Image, List[Image.Image]]) -> torch.Tensor:
+    if isinstance(image, list):
+        return torch.cat([pil2tensor(img) for img in image], dim=0)
+
     return torch.from_numpy(np.array(image).astype(np.float32) / 255.0).unsqueeze(0)
 
-def img_np_to_tensor(img_np):
-    return torch.from_numpy(img_np / 255.0)[None,]
 
-def img_tensor_to_np(img_tensor):
-    img_tensor = img_tensor.clone()
-    img_tensor = img_tensor * 255.0
-    return img_tensor.squeeze(0).numpy().astype(np.float32)
+def np2tensor(img_np: Union[np.ndarray, List[np.ndarray]]) -> torch.Tensor:
+    if isinstance(img_np, list):
+        return torch.cat([np2tensor(img) for img in img_np], dim=0)
+
+    return torch.from_numpy(img_np.astype(np.float32) / 255.0).unsqueeze(0)
 
 
+def tensor2np(tensor: torch.Tensor) -> List[np.ndarray]:
+    batch_count = tensor.size(0) if len(tensor.shape) > 3 else 1
+    if batch_count > 1:
+        out = []
+        for i in range(batch_count):
+            out.extend(tensor2np(tensor[i]))
+        return out
+
+    return [np.clip(255.0 * tensor.cpu().numpy().squeeze(), 0, 255).astype(np.uint8)]
+
+
+def pad(img, left, right, top, bottom):
+    pad_width = np.array(((0, 0), (top, bottom), (left, right)))
+    print(f"pad_width: {pad_width}, shape: {pad_width.shape}")  # Debugging line
+    return np.pad(img, pad_width, mode="wrap")
+
+
+def tiles_infer(tiles, ort_session, progress_callback=None):
+    """Infer each tile with the given model. progress_callback will be called with
+    arguments : current tile idx and total tiles amount (used to show progress on
+    cursor in Blender)."""
+
+    out_channels = 3  # normal map RGB channels
+    tiles_nb = tiles.shape[0]
+    pred_tiles = np.empty((tiles_nb, out_channels, tiles.shape[2], tiles.shape[3]))
+
+    for i in range(tiles_nb):
+        if progress_callback != None:
+            progress_callback(i + 1, tiles_nb)
+        pred_tiles[i] = ort_session.run(
+            None, {"input": tiles[i : i + 1].astype(np.float32)}
+        )[0]
+
+    return pred_tiles
+
+
+def generate_mask(tile_size, stride_size):
+    """Generates a pyramidal-like mask. Used for mixing overlapping predicted tiles."""
+
+    tile_h, tile_w = tile_size
+    stride_h, stride_w = stride_size
+    ramp_h = tile_h - stride_h
+    ramp_w = tile_w - stride_w
+
+    mask = np.ones((tile_h, tile_w))
+
+    # ramps in width direction
+    mask[ramp_h:-ramp_h, :ramp_w] = np.linspace(0, 1, num=ramp_w)
+    mask[ramp_h:-ramp_h, -ramp_w:] = np.linspace(1, 0, num=ramp_w)
+    # ramps in height direction
+    mask[:ramp_h, ramp_w:-ramp_w] = np.transpose(
+        np.linspace(0, 1, num=ramp_h)[None], (1, 0)
+    )
+    mask[-ramp_h:, ramp_w:-ramp_w] = np.transpose(
+        np.linspace(1, 0, num=ramp_h)[None], (1, 0)
+    )
+
+    # Assume tiles are squared
+    assert ramp_h == ramp_w
+    # top left corner
+    corner = np.rot90(corner_mask(ramp_h), 2)
+    mask[:ramp_h, :ramp_w] = corner
+    # top right corner
+    corner = np.flip(corner, 1)
+    mask[:ramp_h, -ramp_w:] = corner
+    # bottom right corner
+    corner = np.flip(corner, 0)
+    mask[-ramp_h:, -ramp_w:] = corner
+    # bottom right corner
+    corner = np.flip(corner, 1)
+    mask[-ramp_h:, :ramp_w] = corner
+
+    return mask
+
+
+def corner_mask(side_length):
+    """Generates the corner part of the pyramidal-like mask.
+    Currently, only for square shapes."""
+
+    corner = np.zeros([side_length, side_length])
+
+    for h in range(0, side_length):
+        for w in range(0, side_length):
+            if h >= w:
+                sh = h / (side_length - 1)
+                corner[h, w] = 1 - sh
+            if h <= w:
+                sw = w / (side_length - 1)
+                corner[h, w] = 1 - sw
+
+    return corner - 0.25 * scaling_mask(side_length)
+
+
+def scaling_mask(side_length):
+    scaling = np.zeros([side_length, side_length])
+
+    for h in range(0, side_length):
+        for w in range(0, side_length):
+            sh = h / (side_length - 1)
+            sw = w / (side_length - 1)
+            if h >= w and h <= side_length - w:
+                scaling[h, w] = sw
+            if h <= w and h <= side_length - w:
+                scaling[h, w] = sh
+            if h >= w and h >= side_length - w:
+                scaling[h, w] = 1 - sh
+            if h <= w and h >= side_length - w:
+                scaling[h, w] = 1 - sw
+
+    return 2 * scaling
+
+
+def tiles_merge(tiles, stride_size, img_size, paddings):
+    """Merges the list of tiles into one image. img_size is the original size, before
+    padding."""
+
+    _, tile_h, tile_w = tiles[0].shape
+    pad_left, pad_right, pad_top, pad_bottom = paddings
+    height = img_size[1] + pad_top + pad_bottom
+    width = img_size[2] + pad_left + pad_right
+    stride_h, stride_w = stride_size
+
+    # stride must be even
+    assert (stride_h % 2 == 0) and (stride_w % 2 == 0)
+    # stride must be greater or equal than half tile
+    assert (stride_h >= tile_h / 2) and (stride_w >= tile_w / 2)
+    # stride must be smaller or equal tile size
+    assert (stride_h <= tile_h) and (stride_w <= tile_w)
+
+    merged = np.zeros((img_size[0], height, width))
+    mask = generate_mask((tile_h, tile_w), stride_size)
+
+    h_range = ((height - tile_h) // stride_h) + 1
+    w_range = ((width - tile_w) // stride_w) + 1
+
+    idx = 0
+    for h in range(0, h_range):
+        for w in range(0, w_range):
+            h_from, h_to = h * stride_h, h * stride_h + tile_h
+            w_from, w_to = w * stride_w, w * stride_w + tile_w
+            merged[:, h_from:h_to, w_from:w_to] += tiles[idx] * mask
+            idx += 1
+
+    return merged[:, pad_top:-pad_bottom, pad_left:-pad_right]
+
+
+def tiles_split(img, tile_size, stride_size):
+    """Returns list of tiles from the given image and the padding used to fit the tiles
+    in it. Input image must have dimension C,H,W."""
+    log.debug(f"Splitting img: tile {tile_size}, stride {stride_size} ")
+    tile_h, tile_w = tile_size
+    stride_h, stride_w = stride_size
+    img_h, img_w = img.shape[0], img.shape[1]
+
+    # stride must be even
+    assert (stride_h % 2 == 0) and (stride_w % 2 == 0)
+    # stride must be greater or equal than half tile
+    assert (stride_h >= tile_h / 2) and (stride_w >= tile_w / 2)
+    # stride must be smaller or equal tile size
+    assert (stride_h <= tile_h) and (stride_w <= tile_w)
+
+    # find total height & width padding sizes
+    pad_h, pad_w = 0, 0
+    remainer_h = (img_h - tile_h) % stride_h
+    remainer_w = (img_w - tile_w) % stride_w
+    if remainer_h != 0:
+        pad_h = stride_h - remainer_h
+    if remainer_w != 0:
+        pad_w = stride_w - remainer_w
+
+    # if tile bigger than image, pad image to tile size
+    if tile_h > img_h:
+        pad_h = tile_h - img_h
+    if tile_w > img_w:
+        pad_w = tile_w - img_w
+
+    # pad image, add extra stride to padding to avoid pyramid
+    # weighting leaking onto the valid part of the picture
+    pad_left = pad_w // 2 + stride_w
+    pad_right = pad_left if pad_w % 2 == 0 else pad_left + 1
+    pad_top = pad_h // 2 + stride_h
+    pad_bottom = pad_top if pad_h % 2 == 0 else pad_top + 1
+    img = pad(img, pad_left, pad_right, pad_top, pad_bottom)
+    img_h, img_w = img.shape[1], img.shape[2]
+
+    # extract tiles
+    h_range = ((img_h - tile_h) // stride_h) + 1
+    w_range = ((img_w - tile_w) // stride_w) + 1
+    tiles = np.empty([h_range * w_range, img.shape[0], tile_h, tile_w])
+    idx = 0
+    for h in range(0, h_range):
+        for w in range(0, w_range):
+            h_from, h_to = h * stride_h, h * stride_h + tile_h
+            w_from, w_to = w * stride_w, w * stride_w + tile_w
+            tiles[idx] = img[:, h_from:h_to, w_from:w_to]
+            idx += 1
+
+    return tiles, (pad_left, pad_right, pad_top, pad_bottom)
+
+
+# endregion
+
+
+# region MODEL Utilities
+def download_antelopev2():
+    antelopev2_url = "https://drive.google.com/uc?id=18wEUfMNohBJ4K3Ly5wpTejPfDzp-8fI8"
+
+    try:
+        import gdown
+
+        log.debug("Loading antelopev2 model")
+
+        dest = get_model_path("insightface")
+        archive = dest / "antelopev2.zip"
+        final_path = dest / "models" / "antelopev2"
+        if not final_path.exists():
+            log.info(f"antelopev2 not found, downloading to {dest}")
+            gdown.download(
+                antelopev2_url,
+                archive.as_posix(),
+                resume=True,
+            )
+
+            log.info(f"Unzipping antelopev2 to {final_path}")
+
+            if archive.exists():
+                # we unzip it
+                import zipfile
+
+                with zipfile.ZipFile(archive.as_posix(), "r") as zip_ref:
+                    zip_ref.extractall(final_path.parent.as_posix())
+
+    except Exception as e:
+        log.error(
+            f"Could not load or download antelopev2 model, download it manually from {antelopev2_url}"
+        )
+        raise e
+
+
+def get_model_path(fam, model=None):
+    log.debug(f"Requesting {fam} with model {model}")
+    res = None
+    if model:
+        res = folder_paths.get_full_path(fam, model)
+    else:
+        # this one can raise errors...
+        with contextlib.suppress(KeyError):
+            res = folder_paths.get_folder_paths(fam)
+
+    if res:
+        if isinstance(res, list):
+            if len(res) > 1:
+                log.warning(
+                    f"Found multiple match, we will pick the first {res[0]}\n{res}"
+                )
+            res = res[0]
+        res = Path(res)
+        log.debug(f"Resolved model path from folder_paths: {res}")
+    else:
+        res = models_dir / fam
+        if model:
+            res /= model
+
+    return res
+
+
+# endregion
+
+
+# region UV Utilities
+
+
+def create_uv_map_tensor(width=512, height=512):
+    u = torch.linspace(0.0, 1.0, steps=width)
+    v = torch.linspace(0.0, 1.0, steps=height)
+
+    U, V = torch.meshgrid(u, v)
+
+    uv_map = torch.zeros(height, width, 3, dtype=torch.float32)
+    uv_map[:, :, 0] = U.t()
+    uv_map[:, :, 1] = V.t()
+
+    return uv_map.unsqueeze(0)
+
+
+# endregion
+
+
+# region ANIMATION Utilities
+def apply_easing(value, easing_type):
+    if easing_type == "Linear":
+        return value
+
+    # Back easing functions
+    def easeInBack(t):
+        s = 1.70158
+        return t * t * ((s + 1) * t - s)
+
+    def easeOutBack(t):
+        s = 1.70158
+        return ((t - 1) * t * ((s + 1) * t + s)) + 1
+
+    def easeInOutBack(t):
+        s = 1.70158 * 1.525
+        if t < 0.5:
+            return (t * t * (t * (s + 1) - s)) * 2
+        return ((t - 2) * t * ((s + 1) * t + s) + 2) * 2
+
+    # Elastic easing functions
+    def easeInElastic(t):
+        if t == 0:
+            return 0
+        if t == 1:
+            return 1
+        p = 0.3
+        s = p / 4
+        return -(math.pow(2, 10 * (t - 1)) * math.sin((t - 1 - s) * (2 * math.pi) / p))
+
+    def easeOutElastic(t):
+        if t == 0:
+            return 0
+        if t == 1:
+            return 1
+        p = 0.3
+        s = p / 4
+        return math.pow(2, -10 * t) * math.sin((t - s) * (2 * math.pi) / p) + 1
+
+    def easeInOutElastic(t):
+        if t == 0:
+            return 0
+        if t == 1:
+            return 1
+        p = 0.3 * 1.5
+        s = p / 4
+        t = t * 2
+        if t < 1:
+            return -0.5 * (
+                math.pow(2, 10 * (t - 1)) * math.sin((t - 1 - s) * (2 * math.pi) / p)
+            )
+        return (
+            0.5 * math.pow(2, -10 * (t - 1)) * math.sin((t - 1 - s) * (2 * math.pi) / p)
+            + 1
+        )
+
+    # Bounce easing functions
+    def easeInBounce(t):
+        return 1 - easeOutBounce(1 - t)
+
+    def easeOutBounce(t):
+        if t < (1 / 2.75):
+            return 7.5625 * t * t
+        elif t < (2 / 2.75):
+            t -= 1.5 / 2.75
+            return 7.5625 * t * t + 0.75
+        elif t < (2.5 / 2.75):
+            t -= 2.25 / 2.75
+            return 7.5625 * t * t + 0.9375
+        else:
+            t -= 2.625 / 2.75
+            return 7.5625 * t * t + 0.984375
+
+    def easeInOutBounce(t):
+        if t < 0.5:
+            return easeInBounce(t * 2) * 0.5
+        return easeOutBounce(t * 2 - 1) * 0.5 + 0.5
+
+    # Quart easing functions
+    def easeInQuart(t):
+        return t * t * t * t
+
+    def easeOutQuart(t):
+        t -= 1
+        return -(t**2 * t * t - 1)
+
+    def easeInOutQuart(t):
+        t *= 2
+        if t < 1:
+            return 0.5 * t * t * t * t
+        t -= 2
+        return -0.5 * (t**2 * t * t - 2)
+
+    # Cubic easing functions
+    def easeInCubic(t):
+        return t * t * t
+
+    def easeOutCubic(t):
+        t -= 1
+        return t**2 * t + 1
+
+    def easeInOutCubic(t):
+        t *= 2
+        if t < 1:
+            return 0.5 * t * t * t
+        t -= 2
+        return 0.5 * (t**2 * t + 2)
+
+    # Circ easing functions
+    def easeInCirc(t):
+        return -(math.sqrt(1 - t * t) - 1)
+
+    def easeOutCirc(t):
+        t -= 1
+        return math.sqrt(1 - t**2)
+
+    def easeInOutCirc(t):
+        t *= 2
+        if t < 1:
+            return -0.5 * (math.sqrt(1 - t**2) - 1)
+        t -= 2
+        return 0.5 * (math.sqrt(1 - t**2) + 1)
+
+    # Sine easing functions
+    def easeInSine(t):
+        return -math.cos(t * (math.pi / 2)) + 1
+
+    def easeOutSine(t):
+        return math.sin(t * (math.pi / 2))
+
+    def easeInOutSine(t):
+        return -0.5 * (math.cos(math.pi * t) - 1)
+
+    easing_functions = {
+        "Sine In": easeInSine,
+        "Sine Out": easeOutSine,
+        "Sine In/Out": easeInOutSine,
+        "Quart In": easeInQuart,
+        "Quart Out": easeOutQuart,
+        "Quart In/Out": easeInOutQuart,
+        "Cubic In": easeInCubic,
+        "Cubic Out": easeOutCubic,
+        "Cubic In/Out": easeInOutCubic,
+        "Circ In": easeInCirc,
+        "Circ Out": easeOutCirc,
+        "Circ In/Out": easeInOutCirc,
+        "Back In": easeInBack,
+        "Back Out": easeOutBack,
+        "Back In/Out": easeInOutBack,
+        "Elastic In": easeInElastic,
+        "Elastic Out": easeOutElastic,
+        "Elastic In/Out": easeInOutElastic,
+        "Bounce In": easeInBounce,
+        "Bounce Out": easeOutBounce,
+        "Bounce In/Out": easeInOutBounce,
+    }
+
+    function_ease = easing_functions.get(easing_type)
+    if function_ease:
+        return function_ease(value)
+
+    log.error(f"Unknown easing type: {easing_type}")
+    log.error(f"Available easing types: {list(easing_functions.keys())}")
+    raise ValueError(f"Unknown easing type: {easing_type}")
+
+
+# endregion

web/README.md

@@ -0,0 +1,32 @@
+## Core
+These 3 scripts cannot be used independently and must all be present to work, they are mostly enhancing the frontend of python nodes
+- `comfy_shared`: library of methods used in `mtb_widgets` and `debug`
+
+**mtb_widgets** define ui callbacks, and various widgets like the `COLOR` type:  
+<img src="https://github.com/melMass/comfy_mtb/assets/7041726/5dbcb714-e1e2-4be7-b0e2-68a6c38c83de" width=400/>
+
+or the `BOOL` type:  
+<img src="https://github.com/melMass/comfy_mtb/assets/7041726/7601366d-601c-4f4d-b735-1a4b076770b0" width=400/>
+
+There is also `Debug` which is a node that should be able to display any data input, it handle a few cases and fallback to the string representation of the
+data otherwise:
+![debug](https://github.com/melMass/comfy_mtb/assets/7041726/1f4393e4-1c3d-4807-9501-fe8888bfae25)
+
+
+
+## Standalone
+These scripts can be taken and placed independently of `comfy_mtb` or any other files, mimicking what pythongosss did for their 
+
+- **imageFeed**: a fork of @pythongosssss ' s [image feed](https://github.com/pythongosssss/ComfyUI-Custom-Scripts/tree/main/js), it adds support for: a lightbox to see images bigger, a way to load the current session history (in case of a web page reload), and different icons, most of the work come from the original script. 
+
+
+> **NOTE**
+> 
+> The original imagefeed got updated since and offer more options, ideally I would clean my lightbox thing and PR it to pythongoss later but in the meantime the script will detect if you already use the original one and not load this fork
+
+
+- ![imagefeed2-hd](https://github.com/melMass/comfy_mtb/assets/7041726/8539f46f-78e1-459a-a11c-fddd44e63ca9)
+
+
+- **notify**: a basic toast notification system that I use in some places accross mtb, it can be used by simply calling `window.MTB.notify("Hello world!")`  
+![extract](https://github.com/melMass/comfy_mtb/assets/7041726/450c67fc-a7e9-4bea-ae49-b610d693098d)

web/color_widget.js

@@ -1,195 +0,0 @@
-// Define the Color Picker widget class
-import parseCss from '/extensions/mtb/extern/parse-css.js'
-import { app } from "/scripts/app.js";
-import { ComfyWidgets } from "/scripts/widgets.js";
-
-export function CUSTOM_INT(node, inputName, val, func, config = {}) {
-	return {
-		widget: node.addWidget(
-			"number",
-			inputName,
-			val,
-			func, 
-			Object.assign({}, { min: 0, max: 4096, step: 640, precision: 0 }, config)
-		),
-	};
-}
-const dumb_call = (v,d,node) => {
-    console.log("dumb_call", {v,d,node});
-}
-function isColorBright (rgb, threshold=240) {
-    const brightess = getBrightness(rgb)
-  
-    return brightess > threshold
-  }
-  
-  function getBrightness (rgbObj) {
-    return Math.round(((parseInt(rgbObj[0]) * 299) + (parseInt(rgbObj[1]) * 587) + (parseInt(rgbObj[2]) * 114)) /1000)
-  }
-  
-
-/**
- * @returns {import("/types/litegraph").IWidget} widget
- */
-const custom = (key,val) => {
-    /** @type {import("/types/litegraph").IWidget} */
-    const widget = {}
-    // widget.y = 0;
-    widget.name = key;
-    widget.type = "COLOR";
-    widget.options = { default: "#ff0000" };
-    widget.value = val || "#ff0000";
-    widget.draw = function (ctx,
-        node,
-        widgetWidth,
-        widgetY,
-        height) {
-        const border = 3;
-        
-        // draw a rect with a border and a fill color
-        ctx.fillStyle = "#000";
-        ctx.fillRect(0, widgetY, widgetWidth, height);
-        ctx.fillStyle = this.value;
-        ctx.fillRect(border, widgetY + border, widgetWidth - border * 2, height - border * 2);
-        // write the input name
-        // choose the fill based on the luminoisty of this.value color
-        const color = parseCss(this.value.default || this.value)
-        if (!color) {
-            return
-        }
-        ctx.fillStyle = isColorBright(color.values, 125) ? "#000" : "#fff";
-
-
-        ctx.font = "14px Arial";
-        ctx.textAlign = "center";
-        ctx.fillText(this.name, widgetWidth * 0.5, widgetY + 14);
-
-
-
-        // ctx.strokeStyle = "#fff";
-        // ctx.strokeRect(border, widgetY + border, widgetWidth - border * 2, height - border * 2);
-        
-        
-        // ctx.fillStyle = "#000";
-        // ctx.fillRect(widgetWidth/2 - border / 2 , widgetY + border / 2 , widgetWidth/2 + border / 2, height + border / 2);
-        // ctx.fillStyle = this.value;
-        // ctx.fillRect(widgetWidth/2, widgetY, widgetWidth/2, height);
-
-    }
-    widget.mouse = function (e, pos, node) {
-        if (e.type === "pointerdown") {
-        console.log({e,pos,node})
-        // get widgets of type type : "COLOR"
-        const widgets = node.widgets.filter(w => w.type === "COLOR");
-
-        for (const w of widgets) {
-            // color picker
-            const rect = [w.last_y, w.last_y + 32];
-            console.log({rect,pos})
-            if (pos[1] > rect[0] && pos[1] < rect[1]) {
-                console.log("color picker", node)
-                const picker = document.createElement("input");
-                picker.type = "color";
-                picker.value = this.value;
-                // picker.style.position = "absolute";
-                // picker.style.left = ( pos[0]) + "px";
-                // picker.style.top = (  pos[1]) + "px";
-                
-                // place at screen center
-                // picker.style.position = "absolute";
-                // picker.style.left = (window.innerWidth / 2) + "px";
-                // picker.style.top = (window.innerHeight / 2) + "px";
-                // picker.style.transform = "translate(-50%, -50%)";
-                // picker.style.zIndex = 1000;
-                
-
-
-                document.body.appendChild(picker);
-                
-                picker.addEventListener("change", () => {
-                    this.value = picker.value;
-                    node.graph._version++;
-                    node.setDirtyCanvas(true, true);
-                    document.body.removeChild(picker);
-                });
-                
-                // simulate click with screen center
-                const pointer_event = new MouseEvent('click', {
-                    bubbles: false,
-                    // cancelable: true,
-                    pointerType: "mouse",
-                    clientX: window.innerWidth / 2,
-                    clientY: window.innerHeight / 2,
-                    x: window.innerWidth / 2,
-                    y: window.innerHeight / 2,
-                    offsetX: window.innerWidth / 2,
-                    offsetY: window.innerHeight / 2,
-                    screenX: window.innerWidth / 2,
-                    screenY: window.innerHeight / 2,
-
-
-                });
-                console.log(e)
-                picker.dispatchEvent(pointer_event);
-            
-            }}}}
-    widget.computeSize = function (width) {
-        return [width, 32];
-    }
-    return widget;
-}
-
-app.registerExtension({
-    name: "mtb.ColorPicker",
-    init: () => {
-        ComfyWidgets.COLOR = function () {
-            return {
-                widget:custom("color", "#ff0000")
-            };
-        };
-    },
-    async beforeRegisterNodeDef(nodeType, nodeData, app) {
-        
-        //console.log("mtb.ColorPicker", { nodeType, nodeData, app });
-        const rinputs = nodeData.input?.required; // object with key/value pairs, "0" is the type
-        // console.log(nodeData.name, { nodeType, nodeData, app });
-
-        if (!rinputs) return;
-
-       
-        let has_color = false;
-        for (const [key, input] of Object.entries(rinputs)) {
-            if (input[0] === "COLOR") {
-                has_color = true;                
-                // input[1] = { default: "#ff0000" };
-            
-            }}
-
-        if (!has_color) return;
-        
-        const onNodeCreated = nodeType.prototype.onNodeCreated;
-        nodeType.prototype.onNodeCreated = function () {
-            const r = onNodeCreated ? onNodeCreated.apply(this, arguments) : undefined;
-            this.serialize_widgets = true;
-        // if (rinputs[0] === "COLOR") {
-        // console.log(nodeData.name, { nodeType, nodeData, app });
-
-        // loop through the inputs to find the color inputs
-         for (const [key, input] of Object.entries(rinputs)) {
-            if (input[0] === "COLOR") {
-                this.addCustomWidget(custom(key,input[1]))
-            }
-        // }
-         }
-
-         this.onRemoved = function () {
-            // When removing this node we need to remove the input from the DOM
-            for (let y in this.widgets) {
-                if (this.widgets[y].canvas) {
-                    this.widgets[y].canvas.remove();
-                }
-            }
-        };
-        }
-    }
-});

web/comfy_shared.js

@@ -0,0 +1,365 @@
+/**
+ * File: comfy_shared.js
+ * Project: comfy_mtb
+ * Author: Mel Massadian
+ *
+ * Copyright (c) 2023 Mel Massadian
+ *
+ */
+
+import { app } from '../../scripts/app.js'
+
+export const log = (...args) => {
+  if (window.MTB?.DEBUG) {
+    console.debug(...args)
+  }
+}
+
+//- WIDGET UTILS
+export const CONVERTED_TYPE = 'converted-widget'
+
+export const hasWidgets = (node) => {
+  if (!node.widgets || !node.widgets?.[Symbol.iterator]) {
+    return false
+  }
+  return true
+}
+
+export const cleanupNode = (node) => {
+  if (!hasWidgets(node)) {
+    return
+  }
+
+  for (const w of node.widgets) {
+    if (w.canvas) {
+      w.canvas.remove()
+    }
+    if (w.inputEl) {
+      w.inputEl.remove()
+    }
+    // calls the widget remove callback
+    w.onRemoved?.()
+  }
+}
+
+export function offsetDOMWidget(
+  widget,
+  ctx,
+  node,
+  widgetWidth,
+  widgetY,
+  height
+) {
+  const margin = 10
+  const elRect = ctx.canvas.getBoundingClientRect()
+  const transform = new DOMMatrix()
+    .scaleSelf(
+      elRect.width / ctx.canvas.width,
+      elRect.height / ctx.canvas.height
+    )
+    .multiplySelf(ctx.getTransform())
+    .translateSelf(margin, margin + widgetY)
+
+  const scale = new DOMMatrix().scaleSelf(transform.a, transform.d)
+  Object.assign(widget.inputEl.style, {
+    transformOrigin: '0 0',
+    transform: scale,
+    left: `${transform.a + transform.e}px`,
+    top: `${transform.d + transform.f}px`,
+    width: `${widgetWidth - margin * 2}px`,
+    // height: `${(widget.parent?.inputHeight || 32) - (margin * 2)}px`,
+    height: `${(height || widget.parent?.inputHeight || 32) - margin * 2}px`,
+
+    position: 'absolute',
+    background: !node.color ? '' : node.color,
+    color: !node.color ? '' : 'white',
+    zIndex: 5, //app.graph._nodes.indexOf(node),
+  })
+}
+
+/**
+ * Extracts the type and link type from a widget config object.
+ * @param {*} config
+ * @returns
+ */
+export function getWidgetType(config) {
+  // Special handling for COMBO so we restrict links based on the entries
+  let type = config?.[0]
+  let linkType = type
+  if (type instanceof Array) {
+    type = 'COMBO'
+    linkType = linkType.join(',')
+  }
+  return { type, linkType }
+}
+export const setupDynamicConnections = (nodeType, prefix, inputType) => {
+  const onNodeCreated = nodeType.prototype.onNodeCreated
+  nodeType.prototype.onNodeCreated = function () {
+    const r = onNodeCreated ? onNodeCreated.apply(this, arguments) : undefined
+    this.addInput(`${prefix}_1`, inputType)
+    return r
+  }
+
+  const onConnectionsChange = nodeType.prototype.onConnectionsChange
+  nodeType.prototype.onConnectionsChange = function (
+    type,
+    index,
+    connected,
+    link_info
+  ) {
+    const r = onConnectionsChange
+      ? onConnectionsChange.apply(this, arguments)
+      : undefined
+    dynamic_connection(this, index, connected, `${prefix}_`, inputType)
+  }
+}
+export const dynamic_connection = (
+  node,
+  index,
+  connected,
+  connectionPrefix = 'input_',
+  connectionType = 'PSDLAYER',
+  nameArray = []
+) => {
+  // remove all non connected inputs
+  if (!connected && node.inputs.length > 1) {
+    log(`Removing input ${index} (${node.inputs[index].name})`)
+    if (node.widgets) {
+      const w = node.widgets.find((w) => w.name === node.inputs[index].name)
+      if (w) {
+        w.onRemoved?.()
+        node.widgets.length = node.widgets.length - 1
+      }
+    }
+    node.removeInput(index)
+
+    // make inputs sequential again
+    for (let i = 0; i < node.inputs.length; i++) {
+      const name =
+        i < nameArray.length ? nameArray[i] : `${connectionPrefix}${i + 1}`
+      node.inputs[i].label = name
+      node.inputs[i].name = name
+    }
+  }
+
+  // add an extra input
+  if (node.inputs[node.inputs.length - 1].link != undefined) {
+    const nextIndex = node.inputs.length
+    const name =
+      nextIndex < nameArray.length
+        ? nameArray[nextIndex]
+        : `${connectionPrefix}${nextIndex + 1}`
+
+    log(`Adding input ${nextIndex + 1} (${name})`)
+
+    node.addInput(name, connectionType)
+  }
+}
+
+/**
+ * Appends a callback to the extra menu options of a given node type.
+ * @param {*} nodeType
+ * @param {*} cb
+ */
+export function addMenuHandler(nodeType, cb) {
+  const getOpts = nodeType.prototype.getExtraMenuOptions
+  nodeType.prototype.getExtraMenuOptions = function () {
+    const r = getOpts.apply(this, arguments)
+    cb.apply(this, arguments)
+    return r
+  }
+}
+
+export function hideWidget(node, widget, suffix = '') {
+  widget.origType = widget.type
+  widget.hidden = true
+  widget.origComputeSize = widget.computeSize
+  widget.origSerializeValue = widget.serializeValue
+  widget.computeSize = () => [0, -4] // -4 is due to the gap litegraph adds between widgets automatically
+  widget.type = CONVERTED_TYPE + suffix
+  widget.serializeValue = () => {
+    // Prevent serializing the widget if we have no input linked
+    const { link } = node.inputs.find((i) => i.widget?.name === widget.name)
+    if (link == null) {
+      return undefined
+    }
+    return widget.origSerializeValue
+      ? widget.origSerializeValue()
+      : widget.value
+  }
+
+  // Hide any linked widgets, e.g. seed+seedControl
+  if (widget.linkedWidgets) {
+    for (const w of widget.linkedWidgets) {
+      hideWidget(node, w, ':' + widget.name)
+    }
+  }
+}
+
+export function showWidget(widget) {
+  widget.type = widget.origType
+  widget.computeSize = widget.origComputeSize
+  widget.serializeValue = widget.origSerializeValue
+
+  delete widget.origType
+  delete widget.origComputeSize
+  delete widget.origSerializeValue
+
+  // Hide any linked widgets, e.g. seed+seedControl
+  if (widget.linkedWidgets) {
+    for (const w of widget.linkedWidgets) {
+      showWidget(w)
+    }
+  }
+}
+
+export function convertToWidget(node, widget) {
+  showWidget(widget)
+  const sz = node.size
+  node.removeInput(node.inputs.findIndex((i) => i.widget?.name === widget.name))
+
+  for (const widget of node.widgets) {
+    widget.last_y -= LiteGraph.NODE_SLOT_HEIGHT
+  }
+
+  // Restore original size but grow if needed
+  node.setSize([Math.max(sz[0], node.size[0]), Math.max(sz[1], node.size[1])])
+}
+
+export function convertToInput(node, widget, config) {
+  hideWidget(node, widget)
+
+  const { linkType } = getWidgetType(config)
+
+  // Add input and store widget config for creating on primitive node
+  const sz = node.size
+  node.addInput(widget.name, linkType, {
+    widget: { name: widget.name, config },
+  })
+
+  for (const widget of node.widgets) {
+    widget.last_y += LiteGraph.NODE_SLOT_HEIGHT
+  }
+
+  // Restore original size but grow if needed
+  node.setSize([Math.max(sz[0], node.size[0]), Math.max(sz[1], node.size[1])])
+}
+
+export function hideWidgetForGood(node, widget, suffix = '') {
+  widget.origType = widget.type
+  widget.origComputeSize = widget.computeSize
+  widget.origSerializeValue = widget.serializeValue
+  widget.computeSize = () => [0, -4] // -4 is due to the gap litegraph adds between widgets automatically
+  widget.type = CONVERTED_TYPE + suffix
+  // widget.serializeValue = () => {
+  //     // Prevent serializing the widget if we have no input linked
+  //     const w = node.inputs?.find((i) => i.widget?.name === widget.name);
+  //     if (w?.link == null) {
+  //         return undefined;
+  //     }
+  //     return widget.origSerializeValue ? widget.origSerializeValue() : widget.value;
+  // };
+
+  // Hide any linked widgets, e.g. seed+seedControl
+  if (widget.linkedWidgets) {
+    for (const w of widget.linkedWidgets) {
+      hideWidgetForGood(node, w, ':' + widget.name)
+    }
+  }
+}
+
+export function fixWidgets(node) {
+  if (node.inputs) {
+    for (const input of node.inputs) {
+      log(input)
+      if (input.widget || node.widgets) {
+        // if (newTypes.includes(input.type)) {
+        const matching_widget = node.widgets.find((w) => w.name === input.name)
+        if (matching_widget) {
+          // if (matching_widget.hidden) {
+          //     log(`Already hidden skipping ${matching_widget.name}`)
+          //     continue
+          // }
+          const w = node.widgets.find((w) => w.name === matching_widget.name)
+          if (w && w.type != CONVERTED_TYPE) {
+            log(w)
+            log(`hidding ${w.name}(${w.type}) from ${node.type}`)
+            log(node)
+            hideWidget(node, w)
+          } else {
+            log(`converting to widget ${w}`)
+
+            convertToWidget(node, input)
+          }
+        }
+      }
+    }
+  }
+}
+export function inner_value_change(widget, value, event = undefined) {
+  if (widget.type == 'number' || widget.type == 'BBOX') {
+    value = Number(value)
+  } else if (widget.type == 'BOOL') {
+    value = Boolean(value)
+  }
+  widget.value = value
+  if (
+    widget.options &&
+    widget.options.property &&
+    node.properties[widget.options.property] !== undefined
+  ) {
+    node.setProperty(widget.options.property, value)
+  }
+  if (widget.callback) {
+    widget.callback(widget.value, app.canvas, node, pos, event)
+  }
+}
+
+//- COLOR UTILS
+export function isColorBright(rgb, threshold = 240) {
+  const brightess = getBrightness(rgb)
+  return brightess > threshold
+}
+
+function getBrightness(rgbObj) {
+  return Math.round(
+    (parseInt(rgbObj[0]) * 299 +
+      parseInt(rgbObj[1]) * 587 +
+      parseInt(rgbObj[2]) * 114) /
+      1000
+  )
+}
+
+//- HTML / CSS UTILS
+export function defineClass(className, classStyles) {
+  const styleSheets = document.styleSheets
+
+  // Helper function to check if the class exists in a style sheet
+  function classExistsInStyleSheet(styleSheet) {
+    const rules = styleSheet.rules || styleSheet.cssRules
+    for (const rule of rules) {
+      if (rule.selectorText === `.${className}`) {
+        return true
+      }
+    }
+    return false
+  }
+
+  // Check if the class is already defined in any of the style sheets
+  let classExists = false
+  for (const styleSheet of styleSheets) {
+    if (classExistsInStyleSheet(styleSheet)) {
+      classExists = true
+      break
+    }
+  }
+
+  // If the class doesn't exist, add the new class definition to the first style sheet
+  if (!classExists) {
+    if (styleSheets[0].insertRule) {
+      styleSheets[0].insertRule(`.${className} { ${classStyles} }`, 0)
+    } else if (styleSheets[0].addRule) {
+      styleSheets[0].addRule(`.${className}`, classStyles, 0)
+    }
+  }
+}

web/debug.js

@@ -0,0 +1,122 @@
+/**
+ * File: debug.js
+ * Project: comfy_mtb
+ * Author: Mel Massadian
+ *
+ * Copyright (c) 2023 Mel Massadian
+ *
+ */
+
+import { app } from '../../scripts/app.js'
+
+import * as shared from './comfy_shared.js'
+import { log } from './comfy_shared.js'
+import { MtbWidgets } from './mtb_widgets.js'
+
+// TODO: respect inputs order...
+
+function escapeHtml(unsafe) {
+  return unsafe
+    .replace(/&/g, '&')
+    .replace(/</g, '&lt;')
+    .replace(/>/g, '&gt;')
+    .replace(/"/g, '&quot;')
+    .replace(/'/g, '&#039;')
+}
+app.registerExtension({
+  name: 'mtb.Debug',
+  async beforeRegisterNodeDef(nodeType, nodeData, app) {
+    if (nodeData.name === 'Debug (mtb)') {
+      const onNodeCreated = nodeType.prototype.onNodeCreated
+      nodeType.prototype.onNodeCreated = function () {
+        const r = onNodeCreated
+          ? onNodeCreated.apply(this, arguments)
+          : undefined
+        this.addInput(`anything_1`, '*')
+        return r
+      }
+
+      const onConnectionsChange = nodeType.prototype.onConnectionsChange
+      nodeType.prototype.onConnectionsChange = function (
+        type,
+        index,
+        connected,
+        link_info
+      ) {
+        const r = onConnectionsChange
+          ? onConnectionsChange.apply(this, arguments)
+          : undefined
+        // TODO: remove all widgets on disconnect once computed
+        shared.dynamic_connection(this, index, connected, 'anything_', '*')
+
+        //- infer type
+        if (link_info) {
+          const fromNode = this.graph._nodes.find(
+            (otherNode) => otherNode.id == link_info.origin_id
+          )
+          const type = fromNode.outputs[link_info.origin_slot].type
+          this.inputs[index].type = type
+          // this.inputs[index].label = type.toLowerCase()
+        }
+        //- restore dynamic input
+        if (!connected) {
+          this.inputs[index].type = '*'
+          this.inputs[index].label = `anything_${index + 1}`
+        }
+      }
+
+      const onExecuted = nodeType.prototype.onExecuted
+      nodeType.prototype.onExecuted = function (message) {
+        onExecuted?.apply(this, arguments)
+
+        const prefix = 'anything_'
+
+        if (this.widgets) {
+          // const pos = this.widgets.findIndex((w) => w.name === "anything_1");
+          // if (pos !== -1) {
+          for (let i = 0; i < this.widgets.length; i++) {
+            if (this.widgets[i].name !== 'output_to_console') {
+              this.widgets[i].onRemoved?.()
+            }
+          }
+          this.widgets.length = 1
+        }
+        let widgetI = 1
+
+        if (message.text) {
+          for (const txt of message.text) {
+            const w = this.addCustomWidget(
+              MtbWidgets.DEBUG_STRING(`${prefix}_${widgetI}`, escapeHtml(txt))
+            )
+            w.parent = this
+            widgetI++
+          }
+        }
+        if (message.b64_images) {
+          for (const img of message.b64_images) {
+            const w = this.addCustomWidget(
+              MtbWidgets.DEBUG_IMG(`${prefix}_${widgetI}`, img)
+            )
+            w.parent = this
+            widgetI++
+          }
+          // this.onResize?.(this.size);
+          // this.resize?.(this.size)
+        }
+
+        this.setSize(this.computeSize())
+
+        this.onRemoved = function () {
+          // When removing this node we need to remove the input from the DOM
+          for (let y in this.widgets) {
+            if (this.widgets[y].canvas) {
+              this.widgets[y].canvas.remove()
+            }
+            shared.cleanupNode(this)
+            this.widgets[y].onRemoved?.()
+          }
+        }
+      }
+    }
+  },
+})

web/imageFeed.js

@@ -0,0 +1,333 @@
+/**
+ * File: imageFeed.js
+ * Project: comfy_mtb
+ * Author: Mel Massadian
+ *
+ * Copyright (c) 2023 Mel Massadian
+ *
+ */
+
+// forked from pysssss's imageFeed.js
+
+import { api } from '../../scripts/api.js'
+import { app } from '../../scripts/app.js'
+
+const styles = {
+  lighbox: {
+    position: 'fixed',
+    top: 0,
+    left: 0,
+    width: '100vw',
+    height: '100vh',
+    background: 'rgba(0,0,0,0.5)',
+    display: 'none',
+    justifyContent: 'center',
+    alignItems: 'center',
+    zIndex: 999,
+  },
+  lightboxBtn: (extra) => ({
+    position: 'absolute',
+    top: '50%',
+    background: 'none',
+    border: 'none',
+    color: '#fff',
+    zIndex: 1000,
+    fontSize: '30px',
+    cursor: 'pointer',
+    pointerEvents: 'auto',
+    ...extra,
+  }),
+  img_list: {
+    minHeight: '30px',
+    maxHeight: '300px',
+    width: '100vw',
+    position: 'absolute',
+    bottom: 0,
+    zIndex: 10,
+    background: '#333',
+    overflow: 'auto',
+  },
+}
+
+let currentImageIndex = 0
+const imageUrls = []
+
+let image_menu = null
+let activated = true
+
+app.registerExtension({
+  name: 'mtb.ImageFeed',
+  init: async () => {
+    const pythongossFeed = app.extensions.find(
+      (e) => e.name == 'pysssss.ImageFeed'
+    )
+    if (pythongossFeed) {
+      console.warn(
+        "[mtb] - Aborting the loading of mtb's imageFeed in favor of pysssss.ImageFeed"
+      )
+      activated = false // just in case other methods are added later on
+      return
+    }
+    // - HTML & CSS
+    //- lightbox
+    const lightboxContainer = document.createElement('div')
+    Object.assign(lightboxContainer.style, styles.lighbox)
+
+    const lightboxImage = document.createElement('img')
+    Object.assign(lightboxImage.style, {
+      maxHeight: '100%',
+      maxWidth: '100%',
+      borderRadius: '5px',
+    })
+
+    // previous and next buttons
+    const lightboxPrevBtn = document.createElement('button')
+    const lightboxNextBtn = document.createElement('button')
+
+    lightboxPrevBtn.textContent = '❮'
+    lightboxNextBtn.textContent = '❯'
+
+    Object.assign(lightboxPrevBtn.style, styles.lightboxBtn({ left: '0%' }))
+    Object.assign(lightboxNextBtn.style, styles.lightboxBtn({ right: '0%' }))
+
+    // close button
+    const lightboxCloseBtn = document.createElement('button')
+    Object.assign(
+      lightboxCloseBtn.style,
+      styles.lightboxBtn({ right: '0', top: '0' })
+    )
+    lightboxCloseBtn.textContent = '❌'
+
+    const lightboxButtons = document.createElement('div')
+    Object.assign(lightboxButtons.style, {
+      position: 'absolute',
+      top: '0%',
+      right: '0%',
+      // transform: "translate(50%, -50%)",
+      height: '100%',
+      width: '100%',
+      background: 'none',
+      border: 'none',
+      color: '#fff',
+      fontSize: '30px',
+      cursor: 'pointer',
+      pointerEvents: 'none',
+    })
+
+    lightboxButtons.append(lightboxPrevBtn, lightboxNextBtn, lightboxCloseBtn)
+    lightboxContainer.append(lightboxButtons, lightboxImage)
+
+    //- image list
+    const imageListContainer = document.createElement('div')
+    Object.assign(imageListContainer.style, styles.img_list)
+
+    const createImgListBtn = (text, style) => {
+      const btn = document.createElement('button')
+      btn.type = 'button'
+      btn.textContent = text
+      Object.assign(btn.style, {
+        ...style,
+        border: 'none',
+        color: '#fff',
+        background: 'none',
+        height: '20px',
+        cursor: 'pointer',
+        position: 'absolute',
+        top: '5px',
+        fontSize: '12px',
+        lineHeight: '12px',
+      })
+      imageListContainer.append(btn)
+      return btn
+    }
+    const showBtn = document.createElement('button')
+    const closeBtn = createImgListBtn('❌', {
+      width: '20px',
+      textIndent: '-4px',
+      right: '5px',
+    })
+    const loadButton = createImgListBtn('Load Session History', {
+      right: '90px',
+    })
+    const clearButton = createImgListBtn('Clear', {
+      right: '30px',
+    })
+
+    //- tools popup button
+    showBtn.classList.add('comfy-settings-btn')
+    Object.assign(showBtn.style, {
+      right: '16px',
+      cursor: 'pointer',
+      display: 'none',
+    })
+
+    //- append to DOM
+    document.body.append(imageListContainer)
+
+    showBtn.textContent = '🖼️'
+    showBtn.onclick = () => {
+      imageListContainer.style.display = 'block'
+      showBtn.style.display = 'none'
+    }
+    document.querySelector('.comfy-settings-btn').after(showBtn)
+    document.querySelector('.comfy-settings-btn').after(lightboxContainer)
+
+    // for (const { output } of history) {
+    // 	if (output?.images) {
+    // 		for (const src of output.images) {
+    // 			const img = document.createElement("img");
+    // 			const but = document.createElement("button");
+
+    //- callbacks
+    closeBtn.onclick = () => {
+      imageListContainer.style.display = 'none'
+      showBtn.style.display = 'unset'
+    }
+
+    clearButton.onclick = () => {
+      imageListContainer.replaceChildren(closeBtn, clearButton, loadButton)
+    }
+
+    lightboxNextBtn.onclick = () => {
+      currentImageIndex = (currentImageIndex + 1) % imageUrls.length
+      const imageUrl = imageUrls[currentImageIndex]
+      lightboxImage.src = imageUrl
+    }
+
+    // Modify the lightboxPrevBtn onclick callback
+    lightboxPrevBtn.onclick = () => {
+      currentImageIndex =
+        (currentImageIndex - 1 + imageUrls.length) % imageUrls.length
+      const imageUrl = imageUrls[currentImageIndex]
+      lightboxImage.src = imageUrl
+    }
+
+    lightboxCloseBtn.onclick = () => {
+      lightboxContainer.style.display = 'none'
+    }
+    lightboxImage.onclick = lightboxNextBtn.onclick
+    /**
+     * This is the function that creates the image buttons for the image list
+     * They are wrapped in a button so that they can be clicked and open
+     * the image in the lightbox.
+     * @param {*} src
+     */
+    const createImageBtn = (src) => {
+      console.debug(`making image ${src.filename}`)
+      const img = document.createElement('img')
+      const but = document.createElement('button')
+
+      Object.assign(but.style, {
+        height: '120px',
+        width: '120px',
+        border: 'none',
+        padding: 0,
+        margin: 0,
+      })
+      Object.assign(img.style, {
+        width: '100%',
+        height: '100%',
+        objectFit: 'cover',
+      })
+
+      img.src = `/view?filename=${encodeURIComponent(src.filename)}&type=${
+        src.type
+      }&subfolder=${encodeURIComponent(src.subfolder)}`
+
+      imageUrls.push(img.src)
+
+      console.debug(img.src)
+
+      img.onload = () => {
+        but.style.width = `${120 * (img.naturalWidth / img.naturalHeight)}px`
+      }
+
+      but.onclick = () => {
+        lightboxContainer.style.display = 'flex'
+        // add the same image to the lightbox
+        lightboxImage.src = img.src
+        // lighboxContainer.replaceChildren(lightboxButtons, img);
+      }
+
+      // add right click menu
+      but.addEventListener('contextmenu', (e) => {
+        e.preventDefault()
+
+        if (image_menu) {
+          image_menu.remove()
+        }
+
+        image_menu = document.createElement('div')
+        Object.assign(image_menu.style, {
+          position: 'absolute',
+          top: `${e.clientY}px`,
+          left: `${e.clientX}px`,
+          background: '#333',
+          color: '#fff',
+          padding: '5px',
+          borderRadius: '5px',
+          zIndex: 999,
+        })
+        const load_img = document.createElement('button')
+        load_img.textContent = 'Load'
+        load_img.onclick = () => {
+          app.handleFile(img.src)
+        }
+
+        image_menu.appendChild(load_img)
+        document.body.appendChild(image_menu)
+      })
+
+      but.append(img)
+      imageListContainer.prepend(but)
+    }
+
+    loadButton.onclick = async () => {
+      const all_history = await api.getHistory()
+      for (const history of all_history.History) {
+        if (history.outputs) {
+          for (const key of Object.keys(history.outputs)) {
+            console.debug(key)
+            if (history.outputs[key].images) {
+              for (const im of history.outputs[key].images) {
+                console.debug(im)
+                createImageBtn(im)
+              }
+            }
+          }
+          // for (const src of outputs.outputs.images) {
+          // 	console.debug(src)
+          // 	makeImage(`${src.subfolder}/${src.filename}`)
+          // }
+        }
+      }
+    }
+
+    ///////-------
+
+    // const all_history = await api.getHistory()
+    // for (const history of all_history.History) {
+    // 	if (history.outputs) {
+    // 		for (const key of Object.keys(history.outputs)) {
+    // 			for (const im of history.outputs[key].images) {
+    // 				makeImage(im)
+    // 			}
+    // 		}
+    // 		// for (const src of outputs.outputs.images) {
+    // 		// 	console.debug(src)
+    // 		// 	makeImage(`${src.subfolder}/${src.filename}`)
+    // 		// }
+    // 	}
+    // }
+
+    //- Hook into the API
+    api.addEventListener('executed', ({ detail }) => {
+      if (detail?.output?.images) {
+        for (const src of detail.output.images) {
+          console.debug(`Adding ${src} to image feed`)
+          createImageBtn(src)
+        }
+      }
+    })
+  },
+})

web/notify.js

@@ -0,0 +1,115 @@
+/**
+ * File: notify.js
+ * Project: comfy_mtb
+ * Author: Mel Massadian
+ *
+ * Copyright (c) 2023 Mel Massadian
+ *
+ */
+
+import { app } from '../../scripts/app.js'
+
+const log = (...args) => {
+  if (window.MTB?.TRACE) {
+    console.debug(...args)
+  }
+}
+
+let transition_time = 300
+
+const containerStyle = `
+position: fixed;
+top: 20px;
+left: 20px;
+font-family: monospace;
+z-index: 99999;
+height: 0;
+overflow: hidden;
+transition: height ${transition_time}ms ease-in-out;
+
+`
+
+const toastStyle = `
+    background-color: #333;
+    color: #fff;
+    padding: 10px;
+    border-radius: 5px;
+    opacity: 0;
+    overflow:hidden;
+    height:20px;
+    transition-property: opacity, height, padding;
+    transition-duration: ${transition_time}ms;
+  `
+
+function notify(message, timeout = 3000) {
+  log('Creating toast')
+  const container = document.getElementById('mtb-notify-container')
+  const toast = document.createElement('div')
+  toast.style.cssText = toastStyle
+  toast.innerText = message
+  container.appendChild(toast)
+
+  toast.addEventListener('transitionend', (e) => {
+    // Only on out
+    if (
+      e.target === toast &&
+      e.propertyName === 'height' &&
+      e.elapsedTime > transition_time / 1000 - Number.EPSILON
+    ) {
+      log('Transition out')
+      const totalHeight = Array.from(container.children).reduce(
+        (acc, child) => acc + child.offsetHeight + 10, // Add spacing of 10px between toasts
+        0
+      )
+      container.style.height = `${totalHeight}px`
+
+      // If there are no toasts left, set the container's height to 0
+      if (container.children.length === 0) {
+        container.style.height = '0'
+      }
+
+      setTimeout(() => {
+        container.removeChild(toast)
+        log('Removed toast from DOM')
+      }, transition_time)
+    } else {
+      log('Transition')
+    }
+  })
+
+  // Fading in the toast
+  toast.style.opacity = '1'
+
+  // Update container's height to fit new toast
+  const totalHeight = Array.from(container.children).reduce(
+    (acc, child) => acc + child.offsetHeight + 10, // Add spacing of 10px between toasts
+    0
+  )
+  container.style.height = `${totalHeight}px`
+
+  // remove the toast after the specified timeout
+  setTimeout(() => {
+    // trigger the transitions
+    toast.style.opacity = '0'
+    toast.style.height = '0'
+    toast.style.paddingTop = '0'
+    toast.style.paddingBottom = '0'
+  }, timeout - transition_time)
+}
+
+app.registerExtension({
+  name: 'mtb.Notify',
+  setup() {
+    if (!window.MTB) {
+      window.MTB = {}
+    }
+
+    const container = document.createElement('div')
+    container.id = 'mtb-notify-container'
+    container.style.cssText = containerStyle
+
+    document.body.appendChild(container)
+    window.MTB.notify = notify
+    // window.MTB.notify('Hello world!')
+  },
+})