Running Minimax H3 locally on 12GB VRAM: 5s low-res clips take 10 minutes, anything bigger takes hours
burnswhenireddit · reddit · 2026-08-25
A ComfyUI user reports hitting walls running Minimax H3 locally on a 4070 Super (12GB VRAM, 64GB RAM): 3–5 second low-resolution clips take 10 minutes, but respectable resolutions or clips beyond 5 seconds blow up to many hours (usually aborted). The portable ComfyUI install has blocked them from ever getting Triton or Sage Attention working despite many tutorials, limiting workflow options. A turbo LoRA helps somewhat but not enough to justify higher resolutions. They're asking for workflows suited to this hardware.
More from Multimodal
- Fun video: AI blends cats, dogs, and fish into bizarre creatures — littleteckmonkey · 2026-08-25
- AI music experiment: DJ FL-AI releases new track "I am AI" — AIandDesign · 2026-08-25
- Magnific Launches Wan 3.0: Consistent Character Generation Across 22 Years — iamfakhrealam · 2026-08-25
- TLive-Omni: Omni-Modal Model for E-Commerce Live Streaming Understanding — TaoLiveAIGC · 2026-08-25
- Alibaba Wan 3.0 launches on Merge: 30s 1080p video in one pass — shensi · 2026-08-25
- Which model provides the utmost in fine detail for landscapes? — Dekes1 · 2026-08-25