Merge pull request #4 from Limbicnation/fix/instance-id-validation

fix(validation): instance_id type + combo field validation for legacy…

.github/publish_action.yml

@@ -0,0 +1,20 @@
+name: Publish to Comfy registry
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - main
+    paths:
+      - "pyproject.toml"
+
+jobs:
+  publish-node:
+    name: Publish Custom Node to registry
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out code
+        uses: actions/checkout@v4
+      - name: Publish Custom Node
+        uses: Comfy-Org/publish-node-action@main
+        with:
+          personal_access_token: ${{ secrets.REGISTRY_ACCESS_TOKEN }} ## Add your own personal access token to your Github Repository secrets and reference it here.

README.md

@@ -1,18 +1,22 @@
 # ComfyUI Face Detection Node
 
-A ComfyUI custom node for face detection and cropping using OpenCV Haar cascades, with full ComfyUI v3 schema support and backward compatibility.
+A ComfyUI custom node for face detection and cropping using OpenCV Haar cascades — optimized for H100 cloud pipelines and LTX-Video avatar workflows. Full ComfyUI v3 schema support with backward compatibility for v1/v2.
 
 ![Face Detection Output Example](images/combined_strip.jpg)
 ![Face Detection Output Example](images/comfyui_face_detection_node.png)
 
 ## Features
 
-- **Face Detection**: Uses OpenCV Haar cascade classifiers for robust face detection
-- **Flexible Cropping**: Crop largest face or all detected faces
-- **Adjustable Parameters**: Configurable detection threshold, minimum face size, and padding
-- **Multiple Classifiers**: Choose between default and alternative Haar cascades
-- **ComfyUI v3 Ready**: Full schema support with backward compatibility for v1/v2
-- **Async Execution**: Stateless execution pattern for better performance
+- **Auto-Padding**: Adaptive padding based on detected face size (no hardcoded values)
+- **Temporal Smoothing**: Exponential moving average of bbox coordinates across video frames — eliminates jitter in batch/video processing
+- **Aspect Ratio Presets**: `1:1`, `9:16`, `16:9`, `4:3`, `auto` — with forced crop recalculation
+- **Full Batch Processing**: Iterates all batch items, outputs aligned batch tensors
+- **GPU-First**: `torch.no_grad()` everywhere, minimal CPU transfers, stateless class-level caching
+- **Face Output Format**: `strip` (horizontal layout) or `individual` (separate batch items) for multi-face output
+- **Dual Classifiers**: Choose between default and alternative Haar cascades
+- **Proper Error Signaling**: Returns flagged tensor + metadata when no face detected
+- **ComfyUI v3 Ready**: Full schema support with async execution
+- **Legacy Workflow Compat**: Handles old v1.x workflows with positional `widgets_values` misalignment gracefully
 
 ## Installation
 
@@ -22,9 +26,9 @@ A ComfyUI custom node for face detection and cropping using OpenCV Haar cascades
 3. Click Install
 
 ### Manual Installation
-1. Navigate to your ComfyUI custom nodes directory
-2. Clone this repository:
+1. Navigate to your ComfyUI custom nodes directory:
    ```bash
+   cd ComfyUI/custom_nodes
    git clone https://github.com/Limbicnation/ComfyUI_FaceDetectionNode.git
    cd ComfyUI_FaceDetectionNode
    pip install -r requirements.txt
@@ -32,41 +36,78 @@ A ComfyUI custom node for face detection and cropping using OpenCV Haar cascades
 
 ## Usage
 
-1. Add the "Face Detection and Crop" node to your workflow
-2. Connect an image input
-3. Adjust parameters:
-   - **Detection Threshold**: Confidence threshold (0.1-1.0)
-   - **Min Face Size**: Minimum face size in pixels (32-512)
-   - **Padding**: Padding around detected faces (0-256)
-   - **Output Mode**: "largest_face" or "all_faces"
-   - **Face Output Format**: "strip" (horizontal layout) or "individual" (separate batch items)
-   - **Classifier Type**: "default" or "alternative"
+1. Add the **"Face Detection and Crop v2"** node to your workflow
+2. Connect an image batch input
+3. Adjust parameters as needed
 
-## Parameters
+### Parameters
 
 | Parameter | Type | Range | Default | Description |
 |-----------|------|-------|---------|-------------|
-| detection_threshold | Float | 0.1-1.0 | 0.8 | Face detection confidence threshold |
-| min_face_size | Int | 32-512 | 64 | Minimum size for detected faces |
-| padding | Int | 0-256 | 32 | Padding around detected faces |
-| output_mode | Combo | - | largest_face | Output mode for detected faces |
-| face_output_format | Combo | - | strip | Format for multiple faces (strip/individual) |
-| classifier_type | Combo | - | default | Haar cascade classifier type |
+| detection_threshold | Float | 0.1–1.0 | 0.8 | Face detection confidence threshold (0.1=lenient, 1.0=strict) |
+| min_face_size | Int | 32–512 | 64 | Minimum face dimension in pixels |
+| auto_padding_ratio | Int | 0–100 | 35 | Padding as percentage of detected face size |
+| aspect_ratio | Combo | — | auto | Crop aspect ratio: `auto`, `1:1`, `9:16`, `16:9`, `4:3` |
+| output_mode | Combo | — | largest_face | `largest_face` or `all_faces` |
+| temporal_smoothing | Int | 0–100 | 0 | 0=disabled (image mode) · 1–100=EMA smoothing strength for video |
+| output_height | Int | 256–2048 | 512 | Output height for cropped faces (width derived from aspect ratio) |
+| instance_id | String | — | "0" | Unique ID for temporal smoothing (share across frames). Use "0" for image mode |
+| classifier_type | Combo | — | default | Haar cascade: `default` or `alternative` |
+| face_output_format | Combo | — | strip | `strip` (horizontal) or `individual` (separate batch items) |
+| padding | Int | 0–256 | 0 | Legacy padding in pixels — if >0, overrides `auto_padding_ratio` |
+
+### Outputs
+
+| Output | Type | Description |
+|--------|------|-------------|
+| cropped_faces | IMAGE | Batch of cropped face tensors `[B, H, W, C]` |
+| face_metadata | FLOAT | Per-face metadata `[x, y, w, h, score, detected]` normalized to image dims. Shape: `[B, 6]` — NOT an image, use for downstream bbox logic only |
+
+### Temporal Smoothing for Video
+
+When processing video frames through ComfyUI, face detection bboxes can jitter frame-to-frame. Enable temporal smoothing to stabilize:
+
+- Set `temporal_smoothing` to 1–100 (higher = more smoothing)
+- Use a consistent `instance_id` across all frames in the same video sequence
+- Set to `"0"` for single-image mode (no smoothing)
+
+## Changelog
+
+### v2.1.3
+- **FIX**: `VALIDATE_INPUTS` now includes required combo fields (`aspect_ratio`, `output_mode`, `classifier_type`) in its signature, forcing ComfyUI to delegate their validation to our method instead of doing strict framework-level "Value not in list" checks. Legacy workflows with misaligned positional `widgets_values` (e.g. `'auto'` landing on `output_mode`, `0` landing on `classifier_type`) are now caught and replaced with defaults.
+- **FIX**: Added defensive combo sanitization in both v1 and v3 execute methods — invalid combo values are replaced with defaults at runtime as a safety net.
+
+### v2.1.2
+- **FIX**: `instance_id` input type changed from INT to STRING. Legacy workflows pass `instance_id="default"` (string) from old v1 nodes — ComfyUI's `validate_inputs` runs before `execute()`, so `_coerce_int` never fires. Since `instance_id` is only a dict key, STRING is the correct type.
+
+### v2.1.1
+- **FIX**: `temporal_smoothing` input validation error — moved to optional section in `INPUT_TYPES` to prevent ComfyUI framework-level `int()` coercion crash when legacy workflows pass string `"default"` (from `classifier_type`) into this slot via positional `widgets_values` mapping
+- **FIX**: Added `_coerce_int()` helper for safe string→int conversion with fallback to defaults, applied defensively in both v1 and v3 execute methods
+- **FIX**: Enhanced `VALIDATE_INPUTS` to handle type mismatches gracefully (was only handling `face_output_format` before)
+
+### v2.1.0
+- **BACKWARD-COMPAT**: Re-added optional `face_output_format` param (`strip`/`individual`) — old workflows now work; invalid values auto-fallback with warning
+- **BACKWARD-COMPAT**: Re-added optional `padding` param — auto-converts to `auto_padding_ratio`
+- **FIX**: `all_faces` mode now actually detects ALL faces (not just largest)
+- **FIX**: `OUTPUT_NODE=True` on v3 schema
+- **FIX**: `VALIDATE_INPUTS` on v1 wrapper to catch invalid combo values early
+
+### v2.0.0
+- Auto-Padding, Temporal Smoothing, Aspect Ratio Presets, Full Batch Processing, GPU-First, Proper Error Signaling
 
 ## Compatibility
 
-- **ComfyUI v3**: Full schema support with async execution
-- **ComfyUI v1/v2**: Backward compatibility via wrapper class
-- **Auto-detection**: Automatically selects appropriate implementation
+- **ComfyUI v3**: Full schema support with async execution (`DEFINE_SCHEMA`)
+- **ComfyUI v1/v2**: Backward compatibility via `FaceDetectionNodeV1` wrapper class
+- **Auto-detection**: Automatically selects appropriate implementation based on available ComfyUI API
 
 ## Requirements
 
-- Python ≥ 3.8
+- Python ≥ 3.10
 - OpenCV ≥ 4.5.0
-- PyTorch ≥ 1.9.0
+- PyTorch ≥ 2.0.0
 - NumPy ≥ 1.21.0
-- Pillow ≥ 8.0.0
 
 ## License
 
-Apache License Version 2.0, January 2004 - see LICENSE file for details.
+Apache License Version 2.0, January 2004 — see [LICENSE](LICENSE) file for details.

comfyui-manager-entry.json

@@ -1,12 +1,12 @@
 {
     "author": "Limbicnation", 
-    "title": "ComfyUI Face Detection Node",
+    "title": "ComfyUI Face Detection Node v2",
     "id": "comfyui-face-detection-node",
     "reference": "https://github.com/Limbicnation/ComfyUI_FaceDetectionNode",
     "files": [
         "https://github.com/Limbicnation/ComfyUI_FaceDetectionNode"
     ],
     "install_type": "git-clone",
-    "description": "A ComfyUI custom node for face detection and cropping using OpenCV Haar cascades, with full ComfyUI v3 schema support and backward compatibility. Features adjustable detection threshold, minimum face size, padding, and multiple classifier options.",
+    "description": "H100-optimized face detection with Auto-Padding, Temporal Smoothing, Aspect Ratio Presets (1:1/9:16/16:9/4:3/auto), and full batch processing for LTX-Video avatar pipelines. Backward compatible with v1.x workflows.",
     "nodename_pattern": "FaceDetectionNode"
-}
+}

node_list.json

@@ -2,17 +2,17 @@
     "custom_nodes": [
         {
             "author": "Limbicnation",
-            "title": "ComfyUI Face Detection Node",
+            "title": "ComfyUI Face Detection Node v2",
             "reference": "https://github.com/Limbicnation/ComfyUI_FaceDetectionNode",
             "files": [
                 "https://github.com/Limbicnation/ComfyUI_FaceDetectionNode"
             ],
             "install_type": "git-clone",
-            "description": "A ComfyUI custom node for face detection and cropping using OpenCV Haar cascades, with full ComfyUI v3 schema support and backward compatibility.",
+            "description": "H100-optimized face detection with Auto-Padding, Temporal Smoothing, Aspect Ratio Presets, and full batch processing for LTX-Video avatar pipelines.",
             "nodes": [
-                "Face Detection and Crop"
+                "Face Detection and Crop v2"
             ],
             "nodename_pattern": "FaceDetectionNode"
         }
     ]
-}
+}

pyproject.toml

@@ -1,15 +1,15 @@
 [project]
 name = "ComfyUI_FaceDetectionNode"
-description = "A ComfyUI custom node for face detection and cropping using OpenCV Haar cascades, with full ComfyUI v3 schema support and backward compatibility."
-version = "1.1.0"
+description = "H100-optimized face detection with Auto-Padding, Temporal Smoothing, Aspect Ratio Presets, and full batch processing for LTX-Video avatar pipelines. Backward compatible with v1.x workflows."
+version = "2.1.3"
 license = {text = "Apache-2.0"}
-dependencies = ["opencv-python>=4.5.0", "torch>=1.9.0", "torchvision>=0.10.0", "numpy>=1.21.0", "Pillow>=8.0.0"]
+dependencies = ["opencv-python>=4.5.0", "torch>=2.0.0", "numpy>=1.21.0"]
 
 [project.urls]
 Repository = "https://github.com/Limbicnation/ComfyUI_FaceDetectionNode"
 #  Used by Comfy Registry https://comfyregistry.org
 
 [tool.comfy]
-PublisherId = "Limbicnation"
-DisplayName = "Face Detection Node"
-Icon = ""
+PublisherId = "limbicnation"
+DisplayName = "Face Detection and Crop v2"
+Icon = "https://raw.githubusercontent.com/Limbicnation/ComfyUI_FaceDetectionNode/main/icon.png"

requirements.txt

@@ -1,5 +1,3 @@
 opencv-python>=4.5.0
-torch>=1.9.0
-torchvision>=0.10.0
+torch>=2.0.0
 numpy>=1.21.0
-Pillow>=8.0.0