ComfyUI-MiniMaxH3-CLSS update: per-scene refs, audio recompose, chunked latent upscale
nazgut · reddit · 2026-09-12
The author shipped a major ComfyUI-MiniMaxH3-CLSS update enabling single-go ref2vid with music via the MiniMax Music model. Key changes:
- Per-scene reference nodes: CLSSH3SceneReference(s) bind ref images/audio to one scene of a multi-scene generation (up to 9 images + 3 audios), with a 3-scene example workflow
- All-scene refs in one node: CLSSH3SceneReferencesAll attaches every ref to all scenes and auto-slices a soundtrack into per-scene windows (default 10s) — no rewiring needed
- globaltext on the prompt node: shared style/soundscape rules written once, prepended to every scene block
- Audio recompose: optional fresh audio takes from pure noise per chunk; measured that re-noising turbo audio regenerates the same take (cos 0.90–0.96), only fresh noise changes it
- Chunk-by-chunk latent upscale inside the sampler with cross-faded seams; long video never exists at high res at once (0.7 GB fp16)
- Fixes: recompose crash, bit-identical cross-chunk noise causality, audio telemetry bookkeeping
More from Multimodal
- AI-generated fake trailer imagines a 1999 Dino Crisis movie with retro VHS feel — noizlab_studio · 2026-10-31
- Spatial Hub aggregates ~5000 public COLMAP-compatible spatial datasets for world models — janusch_patas · 2026-10-31
- Single image to walkable 3D Gaussian splat scene using World Labs Atlas and WebGPU — XRarchitect · 2026-10-31
- AI video reimagines Suicide Squad's Joker as a rapper in a full music-video treatment — Silver_War5476 · 2026-10-31
- 60+ community-made ChatGPT slash-style words turn one word into a full visual style — heyshrutimishra · 2026-10-31
- Cartesia's cascaded voice agents top the Speech Agent Arena, beating native speech-to-speech — rohanpaul_ai · 2026-10-31