{"id":"fasth3-omni","slug":"fasth3-omni","name":"FastH3 Omni","orgId":"hao-ai-lab","family":"fasth3-omni","releaseDate":null,"availability":"announced","openWeights":true,"license":null,"paramsTotal":null,"paramsSource":null,"contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://x.com/haoailab/status/2112625528805208257","hfRepo":null,"modalities":["video","audio"],"aliases":["FastH3 Omni","fasth3 omni","fasth3-omni","fasth3 omniref","fasth3-omniref","fastvideo fasth3 omni","fastvideo-fasth3-omni-8-step-v1","fastvideo/fastvideo-fasth3-omni-8-step-v1"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-20","announced":"2026-10-20","blurb":"Hao AI Lab's FastVideo team's 8-step reference-to-video model, distilled from MiniMax's MiniMax-H3: video with synchronized audio from a text prompt plus image, video and audio references, which the lab says runs in real time. Open weights under the MiniMax-H3 Community License, with training code, NVFP4 / FP8 variants and day-0 ComfyUI support.","api":"none","openWeights":"yes","facts":[{"label":"Lab","value":"Hao AI Lab at UC San Diego (FastVideo), with NVIDIA, Nuva Lab and Reactor"},{"label":"Base model","value":"MiniMaxAI/MiniMax-H3 (distilled derivative; not a MiniMax release)"},{"label":"Sampling","value":"8 transformer evaluations, guidance 1.0"},{"label":"Inputs","value":"Text prompt plus up to 9 images, 3 videos and 3 audio clips (12 references max; at least one image or video)"},{"label":"Output","value":"Video at 24 FPS with synchronized 32 kHz stereo audio"},{"label":"Weights","value":"Hugging Face (FastVideo): BF16 release, NVFP4 QAD and FP8 variants, NVFP4 VAE decoders, ComfyUI packages"},{"label":"License","value":"MiniMax-H3 Community License Agreement (card license: other)"},{"label":"Runtimes","value":"UniServe for GB200 / B200 / RTX PRO 6000 / Hopper servers; FastVideo for DGX Spark and RTX 5090"}],"claims":["Lab claim: real-time generation, 12× faster on NVIDIA Blackwell."],"sources":[{"kind":"lab","title":"Hao AI Lab announcement on X","url":"https://x.com/haoailab/status/2112625528805208257","date":"2026-10-20"},{"kind":"repo","title":"FastVideo-FastH3-Omni-8-Step-V1 on Hugging Face","url":"https://huggingface.co/FastVideo/FastVideo-FastH3-Omni-8-Step-V1","date":"2026-10-20"},{"kind":"repo","title":"FastH3 Omni collection on Hugging Face","url":"https://huggingface.co/collections/FastVideo/fasth3-omni","date":"2026-10-20"},{"kind":"press","title":"Hugging Face (multimodalart) post on X","url":"https://x.com/multimodalart/status/2112635820611977250","date":"2026-10-20"}],"firstBenchmark":null},"organization":{"id":"hao-ai-lab","name":"Hao AI Lab (UC San Diego)","slug":"hao-ai-lab","country":null,"url":"https://haoailab.com","color":"#11110F","mark":"H"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Oct 20, 2026; no independent results yet.","lead":"As of Oct 22, 2026, FastH3 Omni is pending benchmarks on isitsota: announced Oct 20, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}