Spatiotemporal Diffusion Priors for Extreme Video Compression
The new codec leverages generative spatiotemporal priors to enhance video compression, significantly reducing the bitrate while preserving realistic textures and motion. This advancement allows developers to transmit high-quality video content more efficiently, which is crucial for modern gaming experiences that demand both performance and visual fidelity.
With a remarkable improvement in perceptually-oriented distortion metrics, the codec outperforms traditional video codecs like VTM, achieving an FID score improvement of up to 73.3 at the same bitrate. This could reshape how developers approach video streaming and in-game cinematics, making it a pivotal development in the field of video compression.
“Our method shows state-of-the-art performance in perceptually-oriented distortion metrics.”
- what
- Introduction of a new video codec based on spatiotemporal diffusion models
- who
- Developed by Disney Research Studios, involving authors from ETH Zurich
- when
- Presented at the Picture Coding Symposium (PCS) on December 7, 2025
- impact
- Enables extreme video compression for better performance in games
Exciting advancements in video compression technology for developers.
Follow video compression updates
See relevant stories in your personalized news feed.
Discussion