Set Model Path

Let's not bury the lede: SetModelPath does not set a model path, and it doesn't download anything. You feed it a string, it hands the exact same string back. The pack's description claims it "downloads models from huggingface" - it doesn't. What it actually is: a passthrough node whose whole job is to give the pack's bigger downloader, AutoDownloadALLModels, a mapping string to chew on.

So why does it exist? ComfyUI workflows that auto-download models tend to be built by dragging outputs around, and sometimes you want to type the "which model goes in which folder" information by hand instead of having a node like GetModelsFromWorkflow guess it for you. SetModelPath is the manual entry point for that. The input modelandpath expects a mapping in the pack's internal format - roughly (modelname.safetensors~~~foldername) - and the modelpath output feeds straight into AutoDownloadALLModels' modelandpath input.

The honest critique

This node is redundant in the worst way: AutoDownloadALLModels has a plain text widget with the exact same tooltip ("Enter hugging face repo id") where you can type the mapping directly. You don't need SetModelPath as an intermediary - it just moves the typing from one widget to another and adds a wire. Its only real reason to exist is if you're chaining it before other nodes that consume a mapping string, or if you want the mapping value visible as a labeled node in the graph instead of buried in a widget.

That's the whole mechanism: return (modelandpath,). No filesystem, no HuggingFace call, no validation. Whatever malformed string you type goes through untouched - which also means a typo here will silently flow downstream into a downloader that then can't parse it.

Install

Same pack-wide story. ComfyUI Manager → search ComfyUI_AutoDownloadModels → install, or clone it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/AIExplorer25/ComfyUI_AutoDownloadModels
cd ComfyUI_AutoDownloadModels
pip install -r requirements.txt

restart, and you're done. No extra dependencies beyond the pack's standard huggingface_hub, hf_transfer, and rapidfuzz.

When you'd actually reach for it

Honestly, rarely - that's the takeaway. The package is a grab-bag of convenience nodes the author built while scripting model provisioning, and this one is the least-convenient convenience in the box. If you're following a tutorial that wires SetModelPath into AutoDownloadALLModels, it works, and now you know what's really happening under the hood. If you're building your own workflow, skip it and type the mapping into the downloader's own widget. One node fewer, same result, and one less thing to explain to yourself next month when you reopen the graph.