Causal Forcing Framewise: Video Generation by THU-ML (ICML 2026)
Causal Forcing Framewise is a diffusion model from THU-ML (ICML 2026) that generates video from text and image prompts, natively unifying T2V and I2V.
C
Causal Forcing Framewise
VideoText-to-VideoImage-to-VideoTHU-MLICML 2026Real-TimeDiffusion model from THU-ML (Tsinghua University) that generates images or video from text and image prompts. Uses causal forcing to significantly outperform self-forcing in visual quality and motion dynamics. Natively unifies T2V and I2V, enabling real-time streaming video generation on a single RTX 4090.
| Developer | THU-ML (Tsinghua University) |
| Release Date | 2026 (ICML 2026) |
| Architecture | Diffusion Transformer with Causal Forcing |
| License | Apache-2.0 |
| Capabilities | Text-to-video, image-to-video, real-time streaming |
Guides and workflows related to this model series.
No articles found.