comfyg-models

comfyg-models

A ComfyUI custom node that adds a dedicated web interface for managing local AI models and exploring generated images — available at /comfyg-models inside your ComfyUI instance.

Features

🗂️ Library (Models)

• Tabbed navigation by model type (Checkpoints, LoRAs, VAEs, ControlNets, Embeddings, Upscalers, CLIP, CLIP Vision)
• Search, sort, and filter by name, file size, CivitAI rating, base model, or tags
• Thumbnail gallery per model — upload reference images via drag-and-drop or file picker
• CivitAI integration: auto-sync metadata, trigger words, descriptions, and example images

🖼️ Results (Generated Images)

• Gallery view of all generated images found in configured scan folders
• Supports PNG, WEBP, and AVIF formats
• Automatic metadata extraction from image files:
  • ComfyUI prompt and workflow chunks (PNG tEXt/zTXt/iTXt)
  • WEBP EXIF and XMP chunks
  • AVIF Exif and xml ISO boxes
• Extracted data stored efficiently — workflow JSON is not stored in the database, only indexed metadata (prompt text, model/LoRA refs, base model, tags)
• Workflow JSON available on-demand via GET /comfyg-models/api/images/{id}/workflow
• Filter by base model, checkpoint, LoRA, metadata presence, or free-text search
• Auto-tagging: source type, base model, model/LoRA names, prompt terms, subfolder
• Image deduplication via SHA256 — the same file added twice is stored once

⚙️ Settings

• CivitAI API key (stored server-side only — never exposed to the frontend)
• Configurable generated image scan paths with folder picker

Access

After installing the plugin, open:

http://127.0.0.1:8188/comfyg-models

A direct-access button is also available in the ComfyUI top bar and the legacy floating menu.

Installation

Clone or copy this repository into your ComfyUI custom nodes folder:

cd ComfyUI/custom_nodes
git clone https://github.com/guerreiro/comfyg-models

Dependencies are installed automatically by ComfyUI Manager. To install manually:

pip install blake3 aiosqlite watchdog

Runtime Data

Plugin data is stored in:

ComfyUI/user/comfyg-models/
├── models.db        # Models, hashes, CivitAI data, tags, notes, ratings
├── images.db        # Generated image index (metadata only, no image blobs)
├── settings.json    # Plugin settings
└── user-images/     # Uploaded reference images for models

The two SQLite databases are intentionally split to keep the model database compact regardless of how many generated images are indexed. Image blobs are never stored — only the file path and extracted metadata.

API Endpoints

| Method | Path | Description | |--------|------|-------------| | GET | /comfyg-models/api/models | List models (supports filters, pagination) | | GET | /comfyg-models/api/models/{id} | Model detail with gallery and prompts | | POST | /comfyg-models/api/models/{id}/sync | Force re-sync model with CivitAI | | POST | /comfyg-models/api/models/{id}/images | Upload image to model gallery | | GET | /comfyg-models/api/images | List generated images (filters, pagination) | | GET | /comfyg-models/api/images/{id} | Image detail | | GET | /comfyg-models/api/images/{id}/content | Serve image file (PNG/WEBP/AVIF) | | GET | /comfyg-models/api/images/{id}/workflow | Read ComfyUI workflow JSON from file on-demand | | POST | /comfyg-models/api/images/{id}/reveal | Open image folder in file manager | | GET | /comfyg-models/api/images/filters | Available filter buckets for the Results UI | | POST | /comfyg-models/api/scan | Start model library scan | | POST | /comfyg-models/api/scan/stop | Stop running model scan | | POST | /comfyg-models/api/results/scan | Start generated image scan | | POST | /comfyg-models/api/results/scan/stop | Stop running image scan | | GET | /comfyg-models/api/settings | Read settings | | PUT | /comfyg-models/api/settings | Update settings | | GET | /comfyg-models/api/export-data | Download raw SQLite database |

Notes

• Requires Python ≥ 3.8 and aiosqlite, blake3, watchdog
• CivitAI API key is stored server-side only and never sent to the frontend
• SHA256 deduplication prevents storing the same image twice
• Workflow JSON is read from the source file on-demand — not cached in the DB — keeping images.db small even with tens of thousands of images
• Scan jobs can be stopped at any time without corrupting the database