Realtime open video generation keeps getting faster. FastH3-Omni is out with weights + training code, and vLLM-Omni serves it day 0 — including on consumer RTX GPUs. Congrats to the @haoailab team! 📷
vLLM-Omni serves the FastH3-Omni open video model on day 0
AIvLLM says FastH3-Omni, an open video generation model, is available with weights and training code, and vLLM-Omni serves it on day 0, including on consumer RTX GPUs. The post credits the FastVideo team at haoailab, which quoted a claim of up to 12x faster generation on NVIDIA Blackwell with NVFP4 and FP8 support.
AIWhy it matters.
The post shows that the open video model runs day 0 in vLLM-Omni, including on consumer RTX GPUs, which matters for anyone deploying it locally.
@vllm_project
As promised, FastVideo open-sources @MiniMax_AI FastH3 Omni, training code included! 🚀 - Real-time gen @ 12x faster on NVIDIA Blackwell - NVFP4/FP8 for Spark + RTX! - ComfyUI day-0 Collab with @NVIDIAAI , Nuva Lab @nuvalab and Reactor @reactorworld Open source must win! ♥️
Source details
vLLM · x.comPublished
Original title: Realtime open video generation keeps getting faster. FastH3-Omni is out with weights + training code, and vLLM-Omni serves it day 0 — inc...