Shypd.aiThe largest open index of AI toolsCausVid is an open-source video generation tool that transforms slow bidirectional video diffusion models into fast autoregressive ones. It enables few-step text-to-video generation and streaming video-to-video translation at 9.4 FPS.
Trending
About
CausVid is an innovative open-source video diffusion model designed to overcome the latency issues of traditional bidirectional models. It achieves fast autoregressive video generation by adapting pretrained bidirectional diffusion transformers to generate frames on-the-fly. To further enhance speed, CausVid extends distribution matching distillation (DMD) to videos, condensing a 50-step diffusion model into a 4-step generator. This process is stabilized by a student initialization scheme based on the teacher's ODE trajectories and an asymmetric distillation strategy that supervises a causal student with a bidirectional teacher. This approach effectively mitigates error accumulation, allowing for long-duration video synthesis from short training clips. CausVid supports fast streaming generation of high-quality videos at 9.4 FPS on a single GPU, leveraging KV caching. It also facilitates streaming video-to-video translation, image-to-video generation, and dynamic prompting in a zero-shot manner.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
