FastH3 Preview v0.1: Four open-weight H3 models cut transformer calls to four
FastVideo releases four FastH3 Preview v0.1 checkpoints that reduce MiniMax H3 from 49 transformer calls to four, generating 768p video with 32 kHz stereo audio. Variants explore prompt-only and synthetic-video training, plus sparse and dense attention. Fast motion and fine audio remain weaker than Base H3; first/last-frame and reference workflows unsupported.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Podcast explores why China isn't further behind in AI
- Community thanks Unsloth for open-source contributions
- Cohere launches Parse 5 document parsing model
- Anthropic and Suno fight Round Hill's bid to relate copyright cases
- Barret Zoph joins Google as VP of research after brief OpenAI stint