Skip to content

Latest commit

 

History

History
48 lines (32 loc) · 4.51 KB

File metadata and controls

48 lines (32 loc) · 4.51 KB

Detector and segmentation setup

MATRIX LAB NODES does not bundle, download, or install model weights or optional detector runtimes. Package import can succeed while MATRIX_SkinMask and MATRIX_EyeMask remain unavailable. Acquire each asset from its cited source only after reviewing its license and suitability for your use, then verify the exact SHA-256 before execution.

Paths below are relative to the active ComfyUI root. Do not rename a different model to match an expected filename; the runtime checks both logical identity and content hash.

Required assets

Used by ComfyUI-relative path Expected SHA-256 Source License evidence and limit
Eye Mask detection models/ultralytics/bbox/Eyeful_v2-Individual.pt 278fee230b1be01cfb8c47c3f9ad7118c7aa33149a2249b80cea4da236361fbe Pinned Hugging Face file A license for this weight has not been established from the retained source evidence. Review before use or redistribution.
Eye Mask SAM refinement models/sams/sam_vit_b_01ec64.pth ec2df62732614e57411cdcf32a23ffdf28910380d03139ee0f4fcbe91eb8c912 Official model file The Segment Anything repository publishes Apache-2.0 terms. Confirm those terms cover your intended use of the weight.
Skin Mask person gate models/onnx/rembg/u2net_human_seg.onnx 01eb6a29a5c4d8edb30b56adad9bb3a2a0535338e480724a213e0acfd2d1c73c Pinned Hugging Face file The host and upstream U-2-Net repository identify Apache-2.0 terms; the conversion lineage has not been fully established here.
Skin Mask human parts models/onnx/human-parts/deeplabv3p-resnet50-human.onnx a6e823a82da10ba24c29adfb544130684568c46bfac865e215bbace3b4035a71 Hugging Face file The host model card uses CC0 but questions the underlying model license. Treat the underlying license as unresolved.

SAM is needed only when Eye Mask's sam_refine is enabled. The person-segmentation weight is needed when Skin Mask's person_gate is enabled. Skin-part inference still needs the human-parts weight whenever any body-part switch is enabled.

Verify the files

From the active ComfyUI root, calculate SHA-256 without opening or modifying the weights.

PowerShell:

Get-FileHash -Algorithm SHA256 -LiteralPath .\models\ultralytics\bbox\Eyeful_v2-Individual.pt
Get-FileHash -Algorithm SHA256 -LiteralPath .\models\sams\sam_vit_b_01ec64.pth
Get-FileHash -Algorithm SHA256 -LiteralPath .\models\onnx\rembg\u2net_human_seg.onnx
Get-FileHash -Algorithm SHA256 -LiteralPath .\models\onnx\human-parts\deeplabv3p-resnet50-human.onnx

Linux or macOS shell:

sha256sum models/ultralytics/bbox/Eyeful_v2-Individual.pt
sha256sum models/sams/sam_vit_b_01ec64.pth
sha256sum models/onnx/rembg/u2net_human_seg.onnx
sha256sum models/onnx/human-parts/deeplabv3p-resnet50-human.onnx

Every result must match the table exactly. Stop on a missing or mismatched file. Do not bypass the registry, substitute a similarly named model, or edit _core/runtime-assets.json.

Optional runtime packages

The detector paths may need onnxruntime, ultralytics, and segment-anything in the same Python environment that launches ComfyUI. Development validation observed versions 1.29.0, 8.4.142, and 1.0 respectively; these observations are not minimum supported versions or installation pins. Inspect the environment and package requirements before making changes, especially where Torch/CUDA could be affected.

Ultralytics publishes AGPL-3.0 and Enterprise licensing options in its official licensing guidance. Which terms apply to this proprietary pack and a particular deployment has not been established by this release documentation. Review the current upstream terms before installation or distribution.

After the exact assets and compatible runtimes are present, restart ComfyUI and test each mask node with non-sensitive local media. A successful import or hash check alone does not prove inference compatibility, output quality, or license suitability.