A ComfyUI extension that relinks missing models in shared workflows: it finds the closest matches among your local files using fuzzy matching, and can download what you don't have from HuggingFace or CivitAI.
- Workflow scanning — finds model references in all nodes, including nested subgraphs
- Custom node support — detects model fields of third-party loader nodes automatically by tracing which model folder each input reads from (works even when that folder is empty); no hardcoded node list
- Multi-LoRA loaders — rgthree Power Lora Loader and LoraManager LoRAs are detected, linked and swappable like any other model, with strength and on/off shown
- Fuzzy matching — intelligent similarity scoring finds files despite renames, different separators, or capitalization; knows that
fp8/bf16/Q4_K_Mvariants are the same model but a 4B is not an 8B; 100% matches shown first, otherwise best matches ≥70% confidence - Cross-platform paths — workflows authored on Windows match on Linux and vice versa (
\vs/is handled) - Folder-aware suggestions — a file sitting in a folder the node can't load from is flagged with a "wrong folder" warning instead of silently failing at prompt time
- Auto-resolve — one click links every perfect match; also available directly from ComfyUI's native Missing Models popup
- Browse & swap — the All models tab lists every model in the workflow grouped by category; swap any of them for another local file without hunting through the node tree, applied to all referencing nodes at once
- Downloads — fetches missing models from HuggingFace/CivitAI (URLs from the workflow, a model database, a bundled index of every Comfy-Org and Kijai file, or online search), with progress, speed display, bulk download, and cancel; HuggingFace files download over multiple connections for much higher speeds (see HuggingFace downloads)
- Hash lookups — when a workflow carries a model's SHA-256, it's identified on HuggingFace, CivitAI or CivArchive even if the file was renamed or deleted from CivitAI
- API tokens — add a HuggingFace token and CivitAI API key in the ⚙ Settings tab for gated and login-only models; stored on the ComfyUI server, never sent back to the browser (or set
HF_TOKEN/CIVITAI_API_TOKEN) - Export for sharing — the Export tab finds a download link for every model (or lets you paste one), then adds a "Model links" note like ComfyUI's official templates, writes the links into the loader nodes so ComfyUI's own downloader can use them, and adds a Model Linker node
- Safe updates — resolved models are applied to the live graph in place (no canvas rebuild); you save the workflow yourself when ready
ComfyUI-Manager: search for Model Linker and install it, then restart ComfyUI. It's published on the ComfyUI Registry as comfyui-model-linker.
Manually:
- Clone or download this repository into your ComfyUI
custom_nodes/directory - Install the requirements with ComfyUI's Python:
pip install -r requirements.txt - Restart ComfyUI
The requirements are requests, aiohttp, rapidfuzz and huggingface_hub. All but rapidfuzz normally come with ComfyUI already. huggingface_hub is listed without a version, so installing never upgrades the one ComfyUI (via transformers) depends on. Without rapidfuzz, matching falls back to Python's difflib.
- Open a workflow with missing models
- Open the Model Linker via the button in ComfyUI's top menu bar, the
Ctrl+Shift+Lshortcut, the button injected into ComfyUI's Missing Models popup, or the buttons on a Model Linker node in a shared workflow - Review missing models and their suggested matches
- Link individual matches, use Auto-Link for all 100% matches, or Download All Missing for models with known sources
- Save your workflow when ready
Files from huggingface.co are downloaded with huggingface_hub (multiple connections via hf_xet where available), which is much faster than a single stream on fast links. CivitAI and other URLs use the built-in downloader. If huggingface_hub is missing or older than 1.1.0 (for example with transformers 4.x), HuggingFace downloads use the built-in downloader too.
Gated models need a HuggingFace token: add one in the ⚙ Settings tab, or set HF_TOKEN in the environment ComfyUI runs in. A Read-scoped token from huggingface.co/settings/tokens is enough. You also need to accept the model's terms on its HuggingFace page once. Downloads never use a token cached by hf auth login.
Files are staged in a hidden folder next to the destination and moved into place when complete. hf_xet wrote no chunk cache to HF_HOME in testing, only about 60 KB of logs per run. If ComfyUI is killed mid-download, the hidden .model_linker_tmp_* folder in the destination can keep the partial file (up to the full model size) and is safe to delete.
See CHANGELOG.md.