ComfyUI_FindModels

ComfyUI_FindModels scans the workflow currently open in ComfyUI, identifies model references that are missing locally, suggests compatible installed models, and finds candidate direct-download URLs from Civitai and Hugging Face.

Features

  • Scans model widgets and serialized workflow data automatically after a workflow opens.
  • Recognizes checkpoints, LoRAs, VAEs, ControlNet, text encoders, diffusion models, embeddings, and upscale models.
  • Uses ComfyUI's configured model paths, including paths from extra_model_paths.yaml.
  • Distinguishes installed, adaptable, and missing model references.
  • Applies only high-confidence local replacements with one click.
  • Finds candidate direct-download URLs without automatically downloading unverified files.
  • Adds FindModels to the ComfyUI top run bar/menu, with a legacy run-bar fallback.
  • Downloads selected Civitai or Hugging Face candidates into the matching configured ComfyUI model folder.
  • Provides a Quark Netdisk search link for missing model filenames.
  • Does not move, delete, or overwrite model files.

Installation

Clone this repository into ComfyUI/custom_nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/YOUR_NAME/ComfyUI_FindModels.git

Restart ComfyUI. FindModels appears in the top run bar/menu.

No extra Python packages are required.

Usage

  1. Open or drag a workflow into ComfyUI.
  2. The extension scans automatically. You can also click FindModels in the top bar.
  3. Review local replacement suggestions.
  4. Click 一键加载模型 to load safe high-confidence local matches into their nodes, or apply an individual suggestion.
  5. For missing models, click 查找下载来源. You can download a Civitai or Hugging Face candidate into its matching configured model folder, or open the Quark Netdisk search link.

The download results are search candidates. Verify the model page, license, base model, and filename before downloading. Some Civitai files require authentication, so their direct URL may redirect to a sign-in page.

Downloads are accepted only from approved Civitai and Hugging Face HTTPS hosts. Existing files are never overwritten. Quark Netdisk does not provide a stable public global-search API, so the extension opens a scoped web search rather than requesting account cookies.

Matching Rules

  • Exact configured path: installed.
  • Exact filename in a different configured subdirectory: adaptable, because the node value must be updated to the configured relative path.
  • Same normalized filename after removing common precision/version suffixes: installed.
  • Similar filename: shown as an adaptable or missing candidate depending on confidence.
  • Automatic application is limited to local matches with confidence of at least 0.86.

Development

Run the core tests from the repository root:

python -m unittest discover -s tests -v

Privacy

Workflow model filenames are sent to Civitai and Hugging Face only when you explicitly click 查找下载直链. Regular scans remain local.

License

MIT

S
Description
No description provided
Readme MIT
204 KiB
Languages
Python 67%
JavaScript 24.5%
CSS 6.1%
PowerShell 2%
Batchfile 0.4%