Skip to main content
Seedance 2.5 is BytePlus ModelArk’s latest multimodal video generation model. It supports text-to-video, first-frame and first-and-last-frame generation, multimodal reference creation, video editing, and video extension. Compared with the Seedance 2.0 series, it extends output duration to 30 seconds and accepts up to 30 reference images, 10 reference videos, and 10 reference audio clips. Each reference video or audio clip must be 2-30 seconds long, and the combined duration for each media type must not exceed 30 seconds. In video editing mode, every reference video must be 4-30 seconds long. Use Seedance 2.5 for longer coherent sequences, richer multi-asset direction, synchronized audio, multilingual dialogue, and professional MP4 or MOV output workflows. 1080p output uses 10-bit color depth and H.265/HEVC encoding. Some browsers cannot play it; download the original and use a compatible player such as VLC, mpv, or QuickTime. The public API supports a two-step Draft workflow: create a 480p preview with providerOptions.seedance.draft: true, then create a 1080p final video with assets.draftTaskId. The draft must belong to the same API key and remains reusable for seven days from its upstream creation time. Both tasks are billed independently. Draft controls are not exposed in Playground. controls.priority accepts integers from 0 to 9. Higher values move ahead of lower-priority waiting tasks within the same upstream endpoint, without interrupting running work. GENGEN local overflow queuing retains FIFO order. For omni-reference workflows, set an explicit GENGEN mode when you know the operation: multimodal_reference maps to reference-to-video, video_modify maps to video editing, and video_extend maps to video extension. Edit and extend prompts must clearly describe the matching operation. If mode is omitted, BytePlus keeps its default automatic task classification. Even an explicitly typed task can fail asynchronously if its prompt describes a different operation. controls.outputFormat defaults to mp4. Choose mov for higher color precision and more consistent color and brightness in grading, keying, compositing, editing, and extension workflows. MOV output uses H.264 video, YUV 4:4:4 chroma sampling, and PCM audio, which is not supported by every player.

​
Model ID

Pass this exact public ID in the top-level model field. Do not substitute an upstream provider alias.

​
Supported modes

  • text_to_video
  • image_first_frame
  • image_first_last_frame
  • multimodal_reference
  • video_modify
  • video_extend

​
Inputs

  • Image
  • Video
  • Audio
  • Text

​
Outputs

  • Video

​
Capabilities

  • Multimodality-to-video
  • Video editing
  • Video extension

​
Additional capabilities

  • Up to 30-second output
  • Up to 50 reference assets
  • Explicit reference, edit, and extend task validation
  • MP4 and professional MOV output

​
Technical details

Availability and final customer prices can change. Check the Dashboard model catalog before production use.
See the API reference overview for authentication, request conventions, and response schemas.