📥 InstantIDModel Download

This is a fetch-and-load node: give it a download URL and it grabs an InstantID model file, then outputs it as an INSTANTID object ready to plug into your InstantID apply pipeline. It exists so a shared workflow can carry its InstantID weights as a link instead of a "go download this and put it in the right folder first" instruction. That's the whole scope - it doesn't do the face conditioning itself, it just gets the model into your graph.

The inputs are as bare as it gets: Download_url, the direct link to the model, and Url_name, the filename to save it as (default model.safetensors). The one output, INSTANTID, is the loaded model handed downstream to the nodes that actually apply it. So think of this as the "load checkpoint" step of an InstantID setup, with an auto-download bolted on.

Now the part worth knowing before you build a workflow around InstantID at all, because it saves you real disappointment. InstantID is an SDXL tool. It works on SDXL and its descendants - Illustrious, Pony, anything SDXL underneath - and there is no Flux version and none coming. If you're on Flux, the identity job moved to PuLID; InstantID simply won't help you there. It was the January-2024 breakthrough that made people ask whether character LoRAs were finished (the answer turned out to be no - an adapter captures a face, a LoRA captures a whole character), and it's still a solid SDXL face tool, but its reference repo has been quiet since mid-2024. So this download node is useful specifically inside an SDXL graph.

The other thing InstantID drags along is dependencies and licensing. It leans on InsightFace for its face embeddings, which means you need the insightface Python package installed and working - a notorious install headache in its own right - and it means the whole thing is research/non-commercial: the InsightFace models underneath are non-commercial, so no amount of permissive licensing on the code lets you sell what you make with it. Downloading the model is easy; the constraints around it are the real gate.

As for actually using what this loads: InstantID conditions on facial landmarks (that's its edge over a plain face adapter - it holds the reference face's geometry across poses), and it likes a higher strength than you'd run an IP-Adapter, typically around 1.0–1.3, often paired with a Canny ControlNet when you also need edge structure. But those settings live on the apply nodes, not here.

Install comes with the pack. ComfyUI Manager, search SDVN_Comfy_node; or cd ComfyUI/custom_nodes && git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node, then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from your ComfyUI root and restart. The auto-download uses aria2c, so install it by hand on Windows/macOS. And separately, get insightface working - that's the dependency that actually blocks most InstantID setups.

Troubleshooting. A failed download is almost always a bad URL (it wants a direct model link) or missing aria2c on Windows. A model that downloads but errors on apply is usually the InsightFace side - the package not installed, or its buffalo face models not present. And if you got InstantID loaded but nothing recognizable comes out, check you're on an SDXL-family base model; running it against Flux or SD 1.5 is the single most common way to waste an afternoon here.