MiniMax H3 Video Gen Acceleration: Spectrum New Settings Boost Speed by 45%
marres · reddit · 2026-08-07
A developer shared the latest test results of running the MiniMax H3 video generation model in ComfyUI using the Spectrum acceleration method. By aggressively lowering the forecast degree to 1, not only did it avoid the expected visual quality degradation, but it also perfectly preserved the native trajectory.
Core Data & Settings (Based on RTX PRO 6000, 992×768, 7s video):
- Speedup: In a 20-step Euler run, only 11 actual transformer evaluations are needed, with the other 9 steps forecasted. Sampler time dropped from 324.98s to 177.80s, achieving a 45% speedup (1.83x throughput).
- VRAM Cost: To store history features, VRAM usage increased, with sampler peak VRAM rising from 5.56 GB to 8.70 GB (an extra 3.2 GiB).
- Key Finding: Unlike previous video models (like WAN), MiniMax H3 shows excellent adaptability to very low-degree local feature forecasting, making it possible to start forecasting computations much earlier. The author has released node version v0.1.8 with these new default settings.
More from Multimodal
- Opus 5.5 turns heads with surprisingly polished vertical video demo — FinanceYF5 · 2026-10-16
- 'Celestial Ocean Reverie' prompt template blends starry skies with dreamlike seascapes — LudovicCreator · 2026-10-16
- 'Thicc Lil' Red': an AI-made musical meme reimagining Little Red Riding Hood — bennash · 2026-10-16
- CoWorld Reasoner uses an evolving video world model as a visual reasoning engine — ModalityDance · 2026-10-16
- PixelLab tackles ComfyUI pain points: model dedup, image triage, workflow repair — SmartRick137 · 2026-10-16
- Everything is instantly a playable game: browser Roadster Rally demo — bennash · 2026-10-16