MiniMax H3 ecosystem roundup: multi-GPU recipes, 24GB VRAM runtime, turbo LoRAs
optimisticalish · reddit · 2026-09-12
A Reddit roundup of the MiniMax H3 ecosystem:
- Inference: RunningHub's H3 Lightning multi-GPU acceleration recipe; FastVideo's FastH3-Preview-v0.2 and FastH3-Live 1.2.0.
- Low-end: New VideoDeltaNet weights plus a ComfyUI runtime that runs on 8×B200 or a single 24GB card.
- Training & speed: Experimental DMD turbo 8-step LoRAs; an Image-Training-Adapter aiming at LoRA concept training without degrading video knowledge.
- Workflows: ComfyUI-MM-1Frame single-frame extractor (auto-picks best of five frames, accepts an OpenPose reference with a one-word pose trick to avoid stick-figure artifacts); a no-RefMod multi-image reference workflow; a Portuguese 3D camera-control widget; and Depthcat for depth blockouts (Apache-2.0, 111MB extra model).
More from Infra
- SemiAnalysis: OpenAI reportedly negotiating ~1GW TPU commitment while seeking sites — dylan522p · 2026-10-28
- I tested 11 video upscaling methods for AI video: local, Topaz and cloud compared — lopas99 · 2026-10-28
- Thermo-backed CTO says AI is in its dial-up phase, touts disaggregated compute for an ultrafast 'broadband moment' — beffjezos · 2026-10-28
- DeepSeek-V4.1-Flash Q2 runs at 41 tok/s on one RTX 3090, 90% token match vs API — musicurgy · 2026-10-28
- USearch v3 RC ships after 1,000 days, its libraries now embedded in 550K-star projects — viglovikov · 2026-10-28
- SK Hynix says memory demand will outpace supply through 2030 — Chemical_Hawk_6307 · 2026-10-28