Running MiniMax H3 on an RTX 3090: Sage Attention + Triton cut 10s video gen from 18 to 10 minutes
Fun-Helicopter21 · reddit · 2026-09-22
A newcomer asks for advice on further speeding up local MiniMax H3 video generation. Setup: RTX 3090 with ComfyUI, generating at 1280×736 (0.9MP).
So far, using Sage Attention + Triton, they cut generation time for a 10-second video from 18 minutes down to 10. They're now exploring deeper tweaks (block-skipping, spectrum, and other methods) inside the ComfyUI node graph but struggle with node placement and connections, and are asking the community for further optimization tips.
More from Multimodal
- AI-generated bull statue swaps into a museum, fooling viewers online — umesh_ai · 2026-09-22
- AI influencer vlogs level up: 30-second consistent clips with same face and story — aftahi_ai · 2026-09-22
- Canva Gets 70% More Video Generations per GPU Hour on NVIDIA Blackwell — pbaylies · 2026-09-22
- Manually adjust camera angles with Viggle's H3 Meridian, for both image and video — linoy_tsaban · 2026-09-22
- Startup claims a neuron-inspired software layer makes AI video 5x faster and 80% cheaper — The Decoder · 2026-09-22
- AI-Generated Short Film Depicts a Cinematic 'Glass' Delivery Rider — NoVeterinarian5438 · 2026-09-22