FastVideo
An open-source project series from HaoAI Lab focused on improving inference efficiency for video generation. It accelerates video diffusion model inference through techniques such as distillation, cache reuse, and operator optimization. FastH3 is its latest result.
Core Facts
Timeline (last 90 days)
The FastVideo team (haoailab) shared generation examples on social media.
The FastVideo team published the open-weight model for FastH3 V2 on Hugging Face.
FastH3 V2 is the latest iteration of the video diffusion model released by FastVideo.
FastVideo provides two complete ComfyUI workflow templates: text-to-video (t2v) and image-to-video (i2v).
Model weights dedicated to Comfy are packaged in the FastVideo/FastVideo-FastH3-Comfy repository.
HaoAI Lab's FastVideo team officially released FastH3.
The FastVideo project focuses on inference acceleration for video generation, with the goal of making video generation faster and more resource-efficient.
In this demonstration, MiniMax released model weights, the FastVideo team open-sourced the acceleration solution and assisted with service integration, and NVIDIA provided hardware sponsorship and joint optimization.
FastH3 is a dedicated inference acceleration solution for the H3 model open-sourced by the FastVideo team (haoailab), which reduces the computational steps required for generation through distillation, step compression, or cache reuse.
All Facts (7)
The FastVideo project focuses on inference acceleration for video generation, with the goal of making video generation faster and more resource-efficient.
75%UnverifiedFastVideo provides two complete ComfyUI workflow templates: text-to-video (t2v) and image-to-video (i2v).
50%UnverifiedModel weights dedicated to Comfy are packaged in the FastVideo/FastVideo-FastH3-Comfy repository.
50%UnverifiedFastH3 V2 is the latest iteration of the video diffusion model released by FastVideo.
50%UnverifiedFastH3 is a dedicated inference acceleration solution for the H3 model open-sourced by the FastVideo team (haoailab), which reduces the computational steps required for generation through distillation, step compression, or cache reuse.
50%UnverifiedThe FastVideo team published the open-weight model for FastH3 V2 on Hugging Face.
50%UnverifiedHaoAI Lab's FastVideo team officially released FastH3.
50%