What it does
Seedance 2.0 is a model from ByteDance Seed built on a single multimodal architecture that produces audio and video jointly. It accepts text, image, audio and video as inputs, and the site describes broad options for referencing and editing with those inputs. Creators can supply images, audio and video as references to shape the performance, lighting, shadow and camera movement of a result. The page highlights steady motion and combined audio-video output, and describes the results as cinematic. The page shows results from ByteDance's internal SeedVideoBench-2.0 evaluation as radar charts for text-to-video, image-to-video and multimodal tasks, plus a gallery of showcase examples. The page links to Dola and a playground, and also offers a Get API option.
Capabilities
- Joint generation of audio and video in one model
- Accepts text, image, audio and video inputs
- Image, audio and video references for control over performance, lighting, shadow and camera movement
- Multimodal reference and editing support
- Evaluated on text-to-video, image-to-video and multimodal tasks
- API access offered
When to use it
- Creating cinematic video where the audio should be generated together with the visuals
- Steering a result with reference images, audio or video instead of text alone
- Adding video generation to an application through the API
