FastVideo open-sources MiniMax FastH3 Omni with training code, 12x faster on Blackwell
multimodalart · x · 2026-10-21
Hao AI Lab's FastVideo has open-sourced MiniMax's FastH3 Omni, training code included. Key points: real-time generation at 12x speedup on NVIDIA Blackwell; NVFP4/FP8 support running on consumer hardware like Spark and RTX cards; day-0 ComfyUI integration. Built in collaboration with NVIDIA, Nuva Lab, and Reactor. Open training code means developers can fine-tune and deploy it themselves.
Related event: FastVideo Open-Sources FastH3 Omni with 12x Real-Time Speedup on Blackwell(2 posts)→
More from Infra
- Voice models should run on-device: Phonon open-source release sparks debate — timshi_ai · 2026-10-21
- Modal benchmarks NVIDIA's Vera CPU: up to 2x x86 throughput on agentic workloads — graceisford · 2026-10-21
- Anthropic's new S-1 draft shows potential $750B compute spend, up from $518B — firstadopter · 2026-10-21
- PyTorch Foundation ecosystem grows 2.4x contributors across six projects including vLLM and Ray — PyTorch · 2026-10-21
- TorchTPU launches today — Rob Mulla publishes explainer video on how it works — Rob_Mulla · 2026-10-21
- DeepMind infra lead calls Gartner's 'compute surplus' claim nonsense: we're massively constrained — appenz · 2026-10-21