Adds a "Paste Link" flow so a model found anywhere can be downloaded to the
correct ComfyUI folder without hunting for it in the repo browser.
New link_resolver.py recognises:
- Civitai /models/<id>, ?modelVersionId=<v> and /api/download/models/<v>,
mapping the Civitai model type to a folder (LORA/LoCon/DoRA -> loras,
Checkpoint -> checkpoints, TextualInversion -> embeddings, VAE,
Controlnet, Upscaler, Hypernetwork), with per-file overrides for a VAE
shipped alongside a checkpoint and filename fallback for unknown types
- HuggingFace /blob/, /resolve/ and /raw/ file links, classified through
the existing classify_file + README-hint logic; repo and /tree/ links
hand off to the normal repo browser; bare "owner/repo" ids also work
- GitHub /blob/ and raw.githubusercontent.com links
- Plain direct URLs, with the filename taken from the path or
Content-Disposition
Sizes come from a HEAD probe with a ranged-GET fallback, and unreachable or
gated links now report why (404, auth required, unreachable) instead of
silently resolving a 14-byte error page as a model.
Security: url_headers() now picks auth by hostname. The download worker
previously used `_hf_headers() if "github" not in url else {}`, which sent
HF_TOKEN to every non-GitHub host including civitai.com. Tokens are now only
ever sent to the host they belong to, and CIVITAI_TOKEN is supported for
gated Civitai downloads.
Frontend: new modal accepts several links at once (one per line), shows each
resolved file with its size and detected destination, and lets that
destination be overridden before downloading. Files already on disk are
marked and skipped.
Also hardens POST handling — a non-JSON error body (aiohttp's
"405: Method Not Allowed", HTML error pages) previously surfaced as an
opaque "JSON.parse: unexpected non-whitespace character" instead of a
readable message, and a failed download left the button stuck on a spinner.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
📦 Advanced Model Manager for ComfyUI
A powerful model browser, downloader and manager built directly into ComfyUI. Browse hundreds of HuggingFace repositories, search across every file in every repo instantly, download models straight to the right folder, manage GitHub workflows, and track everything you have installed — all without leaving ComfyUI.
✨ Features at a Glance
| Feature | Description |
|---|---|
| 🔗 Paste a link | Paste any HuggingFace, Civitai, GitHub or direct URL — the file type is recognised and it lands in the right folder |
| 🔍 Instant cross-repo search | Type any word and find matching files across ALL repos simultaneously |
| 📁 Repository browser | Repos grouped by author with model-family badges (WanVideo, Flux, LTX, etc.) |
| ⬇️ One-click download | Files go straight to the correct ComfyUI model subfolder automatically |
| 📋 Workflow browser | Browse and download ComfyOrg + Kijai example workflow JSONs from GitHub |
| 💾 Downloaded tab | See every model you have installed, delete files, open folders |
| 🌐 HuggingFace live search | Search HuggingFace directly to discover new repos not in the index |
| 📊 Node widget | Shows live model counts (Diffusion, Text Encoders, VAE, LoRAs) on the graph node |
| ⌨️ Keyboard shortcut | Ctrl + Shift + M to open/close the panel from anywhere |
🖼️ Screenshots
Browse Repositories
All HuggingFace repos listed in the left panel, grouped by author and tagged by model family. Click any repo to load its files.
View & Download Files
Select a repo to see all its files with sizes and download status. Files already on disk are marked. Click Download to save a file directly to the correct ComfyUI folder.
Instant Search Across All Repos
Type any word in the search box — results appear instantly across all repos. Results are grouped by repo in the left panel. Matching files are shown on the right. Case-insensitive, multi-word, ignores underscores and dashes.
🚀 Installation
Option 1 — ComfyUI Manager (Recommended)
- Open ComfyUI Manager → Install Custom Nodes
- Search for Advanced Model Manager
- Click Install and restart ComfyUI
Option 2 — Manual
cd ComfyUI/custom_nodes
git clone https://github.com/BISAM20/ComfyUI-advanced-model-manager
cd ComfyUI-advanced-model-manager
pip install -r requirements.txt
Restart ComfyUI.
🎮 How to Use
Adding the node to your graph
Double-click on an empty area of the ComfyUI canvas to open the node search, type Advanced Model Manager, then click the result to place it.
Once placed, the node shows a live count of all your installed models and a button to open the manager panel:
Browsing & Downloading Models
- The left panel lists all HuggingFace repos by author
- Click any repo name to load its files on the right
- Each file shows its size and whether it is already downloaded (✓ checkmark)
- Click ↓ Download to save a file — it goes to the correct ComfyUI subfolder automatically (e.g.
models/diffusion_models/,models/loras/, etc.)
Downloading from a Link (🔗 Paste Link)
Found a model somewhere else? Click 🔗 Paste Link, paste the URL, and press Resolve (or Ctrl+Enter). The file is identified, its size is looked up, and the correct ComfyUI folder is picked automatically — then click Download.
Recognised link types:
| You paste | What happens |
|---|---|
https://civitai.com/models/122359 |
Latest version's files, folder from the Civitai model type (LORA → loras/, Checkpoint → checkpoints/, …) |
https://civitai.com/models/4384?modelVersionId=128713 |
That specific version |
https://civitai.com/api/download/models/62833 |
The version behind that download link |
https://huggingface.co/<repo>/blob/main/path/file.safetensors |
That single file, classified by path + repo README |
https://huggingface.co/<repo>/resolve/main/… |
Same — ?download=true suffixes are fine |
https://huggingface.co/<repo> |
Opens the repo in the normal file browser |
Comfy-Org/flux1-dev |
Bare repo ids work too |
https://github.com/<owner>/<repo>/blob/<branch>/file.json |
Downloaded via raw.githubusercontent.com |
Any direct .safetensors / .gguf / .ckpt / .pt / .json URL |
Filename taken from the URL or Content-Disposition |
You can paste several links at once, one per line. Every row has a destination dropdown, so if the auto-detection guesses wrong you can correct it before downloading. Files already on disk are marked and skipped.
Searching Across All Repos
Just type in the search box — no button needed. The tool searches every file in every repo:
animate→ finds all files with "animate" in the name across all reposwan fp8→ finds Wan FP8 files even if the name uses underscores likeWan_fp8ltx lora→ multi-word search, both words must be present
Search is case-insensitive and normalises underscores, dashes and dots to spaces.
HuggingFace Live Search
Click 🌐 Search HuggingFace to search the live HuggingFace index for repos not already in the local index. Results are clickable and load their file list exactly like local repos.
Workflows Tab
Switch to 📋 Workflows mode to browse and download workflow JSON files from:
- ComfyOrg official example workflows
- Kijai WanVideoWrapper example workflows
Workflows are saved directly to your ComfyUI/user/default/workflows/ folder.
Downloaded Tab (💾)
Switch to the 💾 Downloaded tab to see all models currently on disk, organised by category. From here you can:
- Click 📂 to open the containing folder in your file manager
- Click 🗑️ to delete a file
Category Filter
Use the All Categories dropdown to filter the repo list and search results to a specific type: Checkpoints, LoRAs, VAE, Text Encoders, Upscalers, etc.
Refresh / Build Index
Click Refresh to rebuild the local file index. The index is what makes instant cross-repo search possible. Progress is shown live. Once built, search is instant even across hundreds of repos.
📊 Graph Node
Add the Advanced Model Manager node to your workflow graph to get a live dashboard showing how many models of each type you have installed. Click Open Model Downloader on the node to open the full panel.
Model counts update automatically every 8 seconds.
📂 Where Files Are Saved
The tool automatically classifies files and saves them to the right folder:
| File type | Saved to |
|---|---|
.safetensors, .ckpt checkpoint files |
models/checkpoints/ or models/diffusion_models/ |
| LoRA files | models/loras/ |
| VAE files | models/vae/ |
| Text encoder / CLIP files | models/text_encoders/ or models/clip/ |
| Upscaler files | models/upscale_models/ |
| ControlNet files | models/controlnet/ |
| Embeddings / textual inversions | models/embeddings/ |
| Hypernetworks | models/hypernetworks/ |
Workflow .json files |
user/default/workflows/ |
Any path you have configured in extra_model_paths.yaml is respected. When downloading from a pasted link you can override the detected destination before confirming.
⚙️ Requirements
- ComfyUI (any recent version)
- Python 3.10+
huggingface_hub >= 0.20.0requests >= 2.28.0
🔑 Tokens (Optional)
HuggingFace
For downloading gated models (e.g. Meta Llama), set your HuggingFace token in your environment:
export HF_TOKEN=hf_your_token_here
Or log in via the CLI:
huggingface-cli login
Civitai
Some Civitai models require an account to download. Create an API key in your Civitai account settings and set:
export CIVITAI_TOKEN=your_civitai_api_key
Tokens are only ever sent to the host they belong to — your HuggingFace token is never sent to Civitai, and vice versa.
📄 License
MIT License — see LICENSE for details.
🙏 Credits
Built for the ComfyUI community. Repos and models sourced from HuggingFace. Workflow JSONs from ComfyOrg and Kijai.




