FaceRefine Plus v0.2.0 for MiniMax H3: 2.7x faster, one-dropdown V1/V2 modes, stronger default
retained text
Hi all, v0.2.0 of \*\*ComfyUI-H3-FaceRefine-Plus\*\* is out. It's an add-on pack for Carasibana's\[ComfyUI-H3-FaceRefine\](https://github.com/Carasibana/ComfyUI-H3-FaceRefine) (MiniMax H3 video face refiner) that fixes the refined face wobbling on the head, the pasted-on look and colour drift. \*\*Side-by-side tests first:\*\* [https://annicodex.github.io/FaceRefine-Plus-Tests/](https://annicodex.github.io/FaceRefine-Plus-Tests/) \### What's new \*\*1. +170 % speed (2.7x faster).\*\* The heavy steps now run on the GPU. Same output, within 1/255 of the old CPU nodes. On the \`mara\` example (107 frames, RTX 5090, model loaders cached) a full refine went from \*\*244 s to 90 s\*\*. https://preview.redd.it/lfj9kmv3tg2i1.png?width=1800&format=png&auto=webp&s=5b1f56dcaa3f6632a2248b87b072e344d187ae99 \*\*2. One Mode dropdown.\*\* \`H3 FaceRefine Plus Mode\` picks the whole recipe, and lazy switches run only the branch you pick: \- \*\*V1\*\* (default): face crop re-drawn with prompt + 3 reference pictures, Stitch Plus, Flow Lock, Tone Match. Best identity and the least wobble. \- \*\*V2\*\*: SAM3 mask on the face (or face + hair), only that part is re-drawn, character sheet only. About 1.4x faster than V1, more wobble. Optional VOSR2 pass for extra sharpness. \*\*3. Default strength is now 0.30\*\* (was 0.15, which changed too little on small faces). Old workflows with the v0.1 labels still load. \### Install \*\*Recommended right now: install with git.\*\* v0.2.0 is on the Comfy Registry, but the ComfyUI Manager list can take a few days to catch up and may still offer v0.1.0. cd ComfyUI/custom\_nodes git clone [https://github.com/AnniCodex/ComfyUI-H3-FaceRefine-Plus](https://github.com/AnniCodex/ComfyUI-H3-FaceRefine-Plus) Already on v0.1.0 from git? Run \`git pull\` inside the folder and restart ComfyUI. Installed v0.1.0 from the Manager? Update it there once it shows v0.2.0, or remove it and use git. It needs the official ComfyUI-H3-FaceRefine. V2 also needs SAM3 nodes, and the \`+ VOSR2\` modes need \[ComfyUI-VOSR2\](https://github.com/ylchen333/ComfyUI-VOSR2). Example workflow: \`examples/face\_refine\_plus\_modes\_workflow.json\`. Repo: [https://github.com/AnniCodex/ComfyUI-H3-FaceRefine-Plus](https://github.com/AnniCodex/ComfyUI-H3-FaceRefine-Plus) Feedback and test clips welcome!
document fetched 2026-11-20 00:42 · canonical https://reddit.com/r/comfyui/comments/1xzvw7k/facerefine_plus_v020_for_minimax_h3_27x_faster