What it does

This is a ComfyUI custom node that wraps IDM-VTON, a diffusion-based virtual try-on model. Feed it a person photo, a garment image, and a mask; it returns the person wearing the new clothes. The node handles model downloads automatically from HuggingFace and expects you to bring your own segmentation and DensePose preprocessing via other ComfyUI extensions.

The interesting bit

The heavy lifting isn’t new — it’s a thin adapter layer. The value is in the plumbing: it wires IDM-VTON into ComfyUI’s node graph so you can swap segmentation models, tweak ControlNet preprocessors, or reroute the pipeline without touching Python. Think of it as a translation layer between a research checkpoint and a visual workflow tool.

Key highlights

• Requires a GPU with at least 16GB VRAM — no CPU fallback, no quantization mentioned
• Auto-downloads model weights from yisol/IDM-VTON on first run
• Segmentation via ComfyUI Segment Anything, DensePose via ControlNet Auxiliary Preprocessors
• Installable through ComfyUI Manager or manual clone into custom_nodes
• Licensed under the original’s CC BY-NC-SA 4.0 (non-commercial)

Caveats

• 16GB VRAM floor puts this firmly in enthusiast territory; laptop GPUs need not apply
• README doesn’t specify which ComfyUI versions are compatible, or whether the node breaks with recent updates
• No mention of inference speed, batching, or memory optimization strategies

Verdict

Worth a look if you’re already living in ComfyUI and need virtual try-on without leaving the node graph. If you’re not invested in ComfyUI’s ecosystem, the standalone IDM-VTON repo is probably less friction.