PixVid Turns Pretrained Latent Video Diffusion Models into Frame-by-Frame Pixel Generation

Written by

in

PixVid retains the diffusion transformer and replaces the VAE decoder with a convolutional head, enabling pretrained latent video models to generate frames directly in pixel space. The project page is chenshuo20.github.io/PixVid/.

Source:Hugging Face 社区论文精选 · Read Original

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *