Popular repositories Loading
-
ComfyUI-MiniMax-H3-W4A4-VSA
ComfyUI-MiniMax-H3-W4A4-VSA PublicMiniMax H3 Ref2VA acceleration for ComfyUI on an RTX 4070 12GB: 4-bit FC1 weights + Streaming VSA sparse attention. 1024x1792 video with audio in about 3 min 17 s, about 3.4x faster than dense atte…
-
ComfyUI-MiniMax-H3-TaoMate-C3
ComfyUI-MiniMax-H3-TaoMate-C3 PublicTaoMate-H3 3-step acceleration package for ComfyUI on an RTX 4070 12GB: 4-bit (W4A4) FFN weights, 10% sparse attention and extra INT8 weight caching, with Windows setup and sample workflows.
-
ComfyUI-H3-X2-Stream
ComfyUI-H3-X2-Stream PublicComfyUI nodes for MiniMax H3: INT8 X2 VAE and asynchronous NVENC saving. 5-second 1088x1920 video with audio in about 60 s on an RTX 4070 12GB, about 30% faster than the FP16 X2 VAE.
-
minimax_h3_workflows
minimax_h3_workflows PublicComfyUI workflows for MiniMax H3 (Ref2VA / FL2VA / VideoRef / Pose Control) at Full HD on 12GB VRAM.
-
h3-3D-camera-guide
h3-3D-camera-guide PublicLocal 3D camera keyframes and RGB motion guides for ComfyUI and MiniMax H3 Ref2VA
-
h3-alphaflow
h3-alphaflow PublicTransparent character video workflow for MiniMax H3 and ComfyUI — from a 4×4 character sheet to alpha WebM / ProRes 4444.
If the problem persists, check the GitHub status page or contact support.
