Make Pixels Dance: High-Dynamic Video Generation
This addresses a key limitation in AI video generation for applications requiring rich motion, though it appears incremental by combining existing techniques.
The paper tackles the problem of generating high-dynamic videos with complex motions, which current text-to-video methods often fail at, by introducing PixelDance, a diffusion-based approach that uses image and text instructions, resulting in significantly better synthesis of videos with intricate scenes and motions.
Creating high-dynamic videos such as motion-rich actions and sophisticated visual effects poses a significant challenge in the field of artificial intelligence. Unfortunately, current state-of-the-art video generation methods, primarily focusing on text-to-video generation, tend to produce video clips with minimal motions despite maintaining high fidelity. We argue that relying solely on text instructions is insufficient and suboptimal for video generation. In this paper, we introduce PixelDance, a novel approach based on diffusion models that incorporates image instructions for both the first and last frames in conjunction with text instructions for video generation. Comprehensive experimental results demonstrate that PixelDance trained with public data exhibits significantly better proficiency in synthesizing videos with complex scenes and intricate motions, setting a new standard for video generation.