comfyui-model-manager

Download and browse models in ComfyUI.

(Viewing metadata and deleting are WIP.)

Fork

About this fork

I made this fork because the original repo was inactive and missing many things I needed to make ComfyUI more usable. Also, many other custom nodes bundle unrelated features together or search the internet in the background.

Currently it is still missing a way to view/download model metadata and delete a model.

Fork Improvements

Download Tab

• Remade download tab.
  • View multiple models connected to url.
  • Select download directory.
  • Optionally download default or custom preivew image.
• Civitai and HuggingFace API token configurable in server_settings.yaml.

Models Tab

• Search bar in models tab.
• Advanced keyword search using "multiple words in quotes" or a minus sign to -exclude.
• Search /subdirectories of model directories based on your file structure (for example, /0/1.5/styles/clothing).
  • Add / at the start of the search bar to see auto-complete suggestions.
• Include models listed in ComfyUI's extra_model_paths.yaml or added in ComfyUI/models.
• Sort for models (Date Created, Date Modified, Name).

ComfyUI Node Graph

• Button to copy a model to the ComfyUI clipboard or embedding to system clipboard. (Embedding copying requires secure http connection.)
• Button to add model to ComfyUI graph or embedding to selected nodes. (For small screens/low resolution.)
• Right, left, top and bottom toggleable sidebar modes.
• Drag a model onto the graph to add a new node.
• Drag a model onto an existing node to set the model field.
• Drag an embedding onto a text area to add it to the end.

Settings Tab

• Correctly change colors using ComfyUI's theme colors.
• Settings tab saved in ui_settings.yaml.
  • Hide/Show 'add' and 'copy-to-clipboard' buttons.
  • Text to always search.
  • Show/Hide add embedding extension.

TODO:

Code

• Javascript cleanup.
  • Seperate into classes per tab?
  • HTML generation all inside main class?
  • More server driven, HTMX-like HTML generation? (Avoid x2 states)

Model Copying

• Copy model path?

Model info window/panel (server load/send on demand)

• Info icon ⓘ button to view.
• Optional (re)download 📥︎ model info from the internet and cache the text file locally. (requires checksum?)
  • Radio buttons to swap between downloaded and server view.
• Delete model with warning popup.
• Set preview image.

Settings

• Exclude hidden folders with a . prefix.
• Enable optional checksum to detect if a model is already downloaded.
• Sidebar width.

Search filtering and sort

• Real-time search
  • Check search code is optimized to avoid recalculation on every minor input change
• Directory dropdown
  • Use always filter to filter directory content auto-suggest dropdown
• Filters dropdown
  • Stable Diffusion model version/Clip/Upscale/?
  • Favorites
• Sort-by dropdown
  • Recently used (ascending/decending)
  • Frequently used (ascending/decending)
• or vs and type of keyword search (currently and)

Image preview

• Support multiple preview images. (swipe?)
• Show preview images for videos.
  • If ffmpeg or cv2 available, extract the first frame of the video and use as image preview.
  • Play preview video?

Accessibility

• Proper naming and labeling.
• Tool tips?

Sidebar

• Drag sidebar width/height dynamically.