Repository Index

We propose Prism, a dynamic sparse attention framework for natively training joint video-audio generation models at 2K.

• diffusion-models
• high-resolution
• sparse-attention
• video-generation