Vidi2

Vidi2 is a research-grade generative video model from ByteDance that focuses on text-, image-, and video-conditioned synthesis. It converts natural-language prompts or single images into short, temporally coherent video clips while preserving appearance details and scene structure. Beyond pure generation, Vidi2 supports video-to-video editing, enabling localized adjustments through masks, prompt-based edits, and style transfer without re-shooting footage.

The pipeline typically couples base generation with temporal refinement, frame interpolation, and super-resolution to improve clarity, extend duration, and maintain consistency across frames. For added control, Vidi2 can leverage auxiliary signals such as depth, segmentation, or optical flow to guide motion, camera behavior, and object trajectories, which is useful for previsualization, motion design, and rapid content prototyping. Researchers and developers can use Vidi2 to explore diffusion-based video generation across multiple conditioning modes and to build creative tools that automate repetitive editing steps.