@dhee_ai/runner-matte
v0.1.1
Published
SAM 3 concept-segmentation matte runner — extract a product (on white) + mask from an image, for compositing.
Downloads
294
Keywords
Readme
dhee-runner-matte
comfy.matte — a SAM 3 concept-segmentation matte runner for
Dhee.
Given an input image and a text concept (e.g. "jar of cookies",
"the product", "round logo badge"), it returns:
<outputPath>— the segmented subject composited on a clean white background (real pixels — used as a reference for model-compositing).<outputPath>_mask.png— the binary mask (for alpha composites).
Native SAM 3 (comfy_extras.nodes_sam3) under the hood:
CheckpointLoaderSimple(sam3.1_multiplex) → CLIPTextEncode(concept) →
SAM3_Detect, with the product-on-white built in-graph
(GetImageSize → EmptyImage → ImageCompositeMasked) so it never depends on
ComfyUI SaveImage preserving alpha.
Config
| key | meaning |
|-----|---------|
| concept / conceptInput | the text concept to segment |
| imageInput | input id holding the image path |
| outputPath (maskOutputPath) | where to write the cutout (+ mask) |
| threshold, refineIterations, ckptName, endpoint | SAM3 / Comfy knobs |
Requires a ComfyUI endpoint (COMFYUI_BASE_URL / ENDPOINT_self_local) with the
native SAM 3 nodes + the sam3.1_multiplex_fp16 checkpoint.
