🌻 Hugging Face Download

Downloads a model or dataset from Hugging Face straight into your ComfyUI folders, from a node inside your graph. If you've ever loaded a shared workflow only to be told you're missing three checkpoints, this is one way to fetch them without leaving the interface - and, notably, it has a mirror option that's a lifesaver for anyone whose connection to Hugging Face is slow or blocked.

What it does and when it helps

Model management is the tax you pay for using ComfyUI - the community complaint about spending "90% of the time downloading models and installing nodes" is not an exaggeration. Normally you download weights in a browser or with the CLI and drop them into the right models/ subfolder yourself. AD_HFDownload moves that step into the graph: give it a repo and a destination, run it, and the file lands where you point it.

The standout input is use_mirror. Turn it on and the download routes through a mirror instead of Hugging Face directly - the reason this pack has it at all is that huggingface.co is painfully slow or unreachable from some regions, and the mirror fixes that. If your HF downloads crawl or time out, that toggle is the first thing to try.

The inputs

• repo_id - the Hugging Face repo, like stabilityai/sdxl-vae.
• filename - a specific file inside the repo, if you don't want the whole thing.
• download_url - a direct URL alternative to repo_id.
• save_path - where it goes. Point this at the correct ComfyUI/models/<type> folder (checkpoints, loras, vae…) so the file lands where ComfyUI actually looks for it.
• resource_type - model or dataset.
• use_mirror - route through the mirror (see above).
• create_folder - make the destination folder if it doesn't exist.
• hf_token - your Hugging Face access token, needed for gated repos (many Flux and other licensed weights require accepting terms and a token).

Outputs are Download_Path (where it ended up) and Status.

Things to keep in mind

This runs when the node executes, as part of the graph - so a big checkpoint means your workflow sits there downloading gigabytes before anything generates. Fine as a one-time setup step, less fine baked into a workflow you run constantly. Grab what you need, then bypass or delete the node.

Gated repos are the usual snag: if a download fails with an auth or 401-style error on a licensed model, you need to accept the license on the Hugging Face page and supply a valid hf_token. And hf_token is a credential - strip it before sharing the workflow, same as any key.

Installing ComfyUI-Addoor

ComfyUI Manager: Install Custom Nodes → search "ComfyUI-Addoor" → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Eagle-CN/ComfyUI-Addoor.git
cd ComfyUI-Addoor
pip install -r requirements.txt

Restart and find it under 🌻 Addoor / Utilities. Double-check save_path points at the right models subfolder - a checkpoint downloaded into the wrong directory is a file ComfyUI can't see.