HF Model Selector

HuggingFace is where the open-weights ecosystem actually ships - the KB's HF panel calls it "the git-based upstream of open AI," the place base models and research weights land before they become CivitAI pages. The problem for a ComfyUI workflow is that HF URLs are long, error-prone strings you'd otherwise be typing by hand. This node is a dropdown that lists the .safetensors files in the repos this pack is configured for, and spits out a ready-to-use URL. It's a small node, but it's the front door to the pack's HuggingFace corner.

Mechanically: the node pulls from the pack's configuration (the supported_models.yaml / schema stack) to know which HF repos it's allowed to browse, lists the safetensor files in each, and hands you a single dropdown (safetensor_file). Pick a file, and the output model_url is a direct download URL of the form https://huggingface.co/<repo>/resolve/main/<file>. That's it - one input, one output.

Here's the honest part, and it's the same trap every auto-generated picker falls into: if the dropdown only shows -- No safetensors available --, you haven't configured anything yet. The choice list is only as good as the pack's repo allowlist. The README ships with a few repos pre-listed - like renderartist/Technically-Color-Z-Image-Turbo and starsfriday/FLUX.2-klein-AC-Style-LORA - but if your model isn't in there, this node won't show it. The pack's HuggingFace setup also wants HF_TOKEN set (the README marks it optional, for private repos, but the node prints a warning when it's absent), and it may need a refresh pass - the __init__.py mentions running python refresh_models.py to populate the HuggingFace model list.

Where it fits in a workflow: it's a URL source. Wire model_url into the pack's HF Lora URL Builder or HF Lora Combiner if you're assembling arrays for a LoRA-capable generation node, or into any node that accepts a model URL string. In this pack's mental model - CPU-only ComfyUI, everything offloaded to APIs - URLs are the currency, and this node just makes them less painful to spend.

Install is the shared pack flow - ComfyUI Manager (search ComfyUI-API-DockerCPU), or:

cd ComfyUI/custom_nodes
git clone https://github.com/trustypangolin/ComfyUI-API-DockerCPU
cd ComfyUI-API-DockerCPU
pip install -r requirements.txt

Set HF_TOKEN (at least for private repos) and restart. It's under 🎨 DockerCPU API/🤗 HuggingFace.

The practical trap is expectation-setting: people wire this up, see a one-entry dropdown, and assume the node is broken. It's not - it's just showing you what the pack knows about. If you need a model that isn't listed, your options are to add the repo to the pack's supported_models.yaml (the README documents the format) or skip this node and use HF Lora URL Builder, which takes any repo ID and filename and constructs the URL manually. Both paths work; one of them doesn't require editing YAML.