vLLM-Omni serves open FastH3-Omni video model on day 0, including consumer RTX GPUs

vLLM says it serves FastH3-Omni, an open video generation model, on day 0 of its release, including on consumer RTX GPUs.

The model is released with weights and training code, according to the vLLM post on X.

vLLM credits the FastVideo team at haoailab for the model. That team claims generation up to 12x faster on NVIDIA Blackwell, with NVFP4 and FP8 support. These speed figures are the developer's claims, and the vLLM post does not independently verify them.