ComfyUI-H3-FaceRefine: a ComfyUI custom node set
A ComfyUI custom node set to refine and improve the quality of small faces in MiniMax H3 video.
Verified
Core facts
Read by a script from the official API, not rewritten by a model.
- License
- MIT
- Languages
- Python
- Maintainer
- Carasibana
- Verified
What it is
ComfyUI-H3-FaceRefine is a ComfyUI custom node set to refine and improve the quality of small faces in MiniMax H3 video. The project says MiniMax H3 renders faces poorly when the head occupies a small fraction of the frame. The project is licensed under MIT.
Main features
H3 Face Track + Crop detects the face on every frame, fills gaps where detection fails, and smooths the trajectory.
H3 Load Video + Face Select loads a video and finds every face and every hard cut in one pass.
H3 Face Stitch Back warps each refined crop back onto the exact float box it came from.
H3 Face Mask (SAM) produces face-shaped paste masks instead of a rectangle, temporally smoothed.
H3 Per-Frame Denoise varies denoise strength along the temporal axis via the latent's noise mask.
In H3 Face Track + Crop, select takes largest_face, smallest_face, left_most, right_most, top_most, bottom_most, centre_most, closest_to_xy or detector_score.
Getting started
After ComfyUI is restarted, the nodes appear under MiniMax H3/Face Refine.
The Python packages ultralytics, scipy, insightface and scenedetect install automatically.
The project includes two example workflows, both annotated in-graph.
The H3 Face Refine Auto Select example workflow chooses the face to refine by a ranking rule, largest_face as shipped.
In the H3 Face Refine Manual Select example workflow, the user chooses the face to refine, one per shot.
git clone https://github.com/Carasibana/ComfyUI-H3-FaceRefine.gitWho it is for
The lipsync path in the example workflows depends on MiniMaxH3NativeAudioLock. MiniMaxH3NativeAudioLock is not redistributed here, so it has to be installed separately. H3 Face Mask (SAM) requires Impact Pack for SAMLoader.
If PySceneDetect is missing or fails to import, the run does not stop. When the environment variable H3_FACEREFINE_STRICT_PATHS is set to a non-empty value, only paths inside ComfyUI's own folder are read, and everything else is refused with a message saying so. In undetected_frames of H3 Face Stitch Back, fade_out ramps the composite to zero across the gap, skip hard-cuts to original pixels, and composite_anyway risks H3 hallucinating a face onto the back of a head.
In paste_region of H3 Face Stitch Back, full_crop pastes hair, shoulders and background too and risks a visible rectangle. In H3 Per-Frame Denoise, face_px_small is a face height in source pixels at or below which full denoise_multiplier_small_face applies. In H3 Per-Frame Denoise, face_px_large is a face height in source pixels at or above which denoise_multiplier_large_face applies.
Related tools
Related tools listed on this site.
LiblibAI·哩布哩布AI
A creative community centered on AI image models and workflows, with model sharing, online generation, and training entry points.
RunningHub
An online creative platform that brings together ComfyUI workflows, AI apps, and a canvas for generating and processing images and videos.
LiblibAI
AI image creation platform and model-sharing community whose site lists online drawing tools, downloadable models and online model training.
Genmo
The site says Genmo is developing open video world models; Mochi 1 is its open-source text-to-video model, usable online in the Playground or run locally via the open-source repository or ComfyUI.
Sources
This page is written from the official sources below. Open them to check the original text.
- GitHub READMEhttps://github.com/Carasibana/ComfyUI-H3-FaceRefineFetched
Verified ·
This page is written by AI from the official sources above and checked automatically. This site has not tested it. Found a mistake? Tell us.