Vidu: a Highly Consistent, Dynamic and Skilled Text-to-Video Generator with Diffusion Models
This work addresses the challenge of scalable and consistent text-to-video generation for applications in media, entertainment, and AI-driven content creation, representing a significant but incremental advance over existing state-of-the-art models.
The authors tackled the problem of generating high-quality, long-duration videos from text by introducing Vidu, a diffusion model that produces 1080p videos up to 16 seconds with strong coherence and dynamism, matching the performance of Sora, the most powerful reported text-to-video generator.
We introduce Vidu, a high-performance text-to-video generator that is capable of producing 1080p videos up to 16 seconds in a single generation. Vidu is a diffusion model with U-ViT as its backbone, which unlocks the scalability and the capability for handling long videos. Vidu exhibits strong coherence and dynamism, and is capable of generating both realistic and imaginative videos, as well as understanding some professional photography techniques, on par with Sora -- the most powerful reported text-to-video generator. Finally, we perform initial experiments on other controllable video generation, including canny-to-video generation, video prediction and subject-driven generation, which demonstrate promising results.