Skip to main content
Video generation is asynchronous. First create a task, then retrieve it until it reaches a terminal state.

Request structure

Generation modes

If mode is omitted, GENGEN can infer the request shape from the supplied assets. Set it explicitly when you want predictable validation.

Seedance 2.5 Flash

Use dreamina-seedance-2-5-flash-261115 with any of the six modes above. Flash supports 4–30 second output (or duration: -1), 480p/720p, and MP4/MOV. It accepts up to 10 images, 5 videos including the source, and 5 audio clips. Video/audio clips are 2–30 seconds each, with at most 30 seconds total for each media type; video editing requires 4–30 second input videos. Audio-only references are supported. Flash follows the 2.5 task-type, adaptive-ratio, priority, and smart-duration rules below. It does not support Draft, 1080p, 4K, or flex. See Flash pricing and capabilities.

Seedance 2.5 task-type validation

For Seedance 2.5 omni-reference workflows, GENGEN maps an explicitly supplied normalized mode to BytePlus’s task-type hint: An edit prompt must clearly describe the intended edit, such as adding, removing, replacing, or changing content. An extension prompt must clearly ask to extend or continue the source video. When mode is omitted, the task type is determined automatically. BytePlus classifies the request from its assets and prompt during processing. Constraint failures found after automatic classification are asynchronous and can use InvalidParameter.TaskTypeConstraint.
An explicit task type enables submission-time constraint validation, but BytePlus still checks prompt intent while processing the task. If the prompt describes a different operation, the task can fail asynchronously with InvalidParameter.TaskTypeMismatch.

Media inputs

assets accepts HTTPS URLs and owned asset:// references where supported.
  • firstFrameImage and lastFrameImage control boundary frames.
  • referenceImages accepts up to 30 references for Seedance 2.5 and up to 9 for Seedance 2.0 variants.
  • referenceVideos accepts up to 10 references for Seedance 2.5 and up to 3 for Seedance 2.0 variants.
  • referenceAudios accepts up to 10 audio references for Seedance 2.5. Audio can be its only multimodal reference input.
  • referenceAudio is the compatibility form for a single audio reference.
  • sourceVideo is used by video modification and extension.
Use GENGEN asset references for files already registered in the same workspace. Requests cannot use private assets owned by another workspace.

Seedance reference audio

For Seedance 2.0-series and 2.5 models, send audio clips in assets.referenceAudios and describe how each clip should guide the result: voice, background music, sound effects, or rhythm. Use mode: "multimodal_reference" for reference generation.

Refer to clips in the prompt

In API prompts, use Audio 1, Audio 2, and so on, following the order in assets.referenceAudios. Numbering starts at 1 and is independent of images and videos: Audio 1 is assets.referenceAudios[0], even when images or videos are also present. The number refers to the submitted audio order, not the file name. This example uses the first audio as a voice reference and the second as a music reference. Replace the example URLs with accessible files that meet the limits below.
For a single clip, assets.referenceAudio is a compatibility alternative to the array and corresponds to Audio 1. Use one form per request. In legacy content[] requests, count only the audio_url entries, in their submitted order.

Formats and limits

Reference audio must be WAV or MP3, with a maximum size of 15 MB per file. AAC and M4A are not supported Seedance reference-audio formats. Submit a public HTTP(S) URL or an authorized asset:// reference; inline/base64 audio is not supported by the public Seedance task API’s asset-preparation path. The combined audio limit is separate from the combined reference-video limit. Seedance 1.5 Pro does not support reference-audio input.

Does reference audio change the price?

For Seedance 2.0-series and 2.5 models, reference audio has no separate audio-input fee and its duration does not add billable input-video seconds. Adding audio alone does not select the video-input pricing rate; supplying a source or reference video does. Final generation charges still use the reported video token usage and the applicable model and resolution price. See Pricing for current rates. controls.generateAudio controls audio in the generated video. It is separate from supplying reference clips, and switching it on or off does not change the unit price for Seedance 2.0-series or 2.5 models. When using model-selected output duration (duration: -1), the final duration and charge can vary with the generated result.

Output controls

Common controls include:
  • duration: model-dependent output duration in seconds; supported Seedance models also accept -1 for model-selected duration.
  • resolution: 480p, 720p, 1080p, or 4k, subject to model support.
  • ratio: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, or adaptive.
  • generateAudio: request synchronized audio when supported.
  • watermark: add a provider watermark.
  • returnLastFrame: return a last-frame image when supported.
  • outputFormat: mp4 or mov for Seedance 2.5. The default is mp4; this control is not available for other Seedance models.
  • serviceTier: default or flex, subject to model support.
Reference videos must contain 407,696–8,847,360 pixels per frame for the Seedance 2.5 series or 407,696–8,295,044 for the 2.0 series. Asset registration accepts up to 8,295,044 pixels, including when the public API automatically registers a reference URL. A 3840×2160 clip fits both limits. Seedance 2.5 supports 4–30 second output at 480p, 720p, 1080p, or 4K. Its first-frame, first/last-frame, edit, and extend modes require adaptive ratio; video editing uses model-selected duration (-1). mp4 uses standard color precision and has broad compatibility with browsers, mobile devices, media players, and distribution platforms. mov preserves higher color precision and more consistent color and brightness for grading, keying, compositing, and other professional post-production workflows. For editing and extension, using MOV for both input and output is recommended.
Seedance 2.5 MOV output uses H.264 video encoding, YUV 4:4:4 chroma sampling, and PCM audio. Some players may not support this combination. VLC, mpv, and ffplay support it on macOS and Windows; IINA supports it on macOS.

Seedance 2.5 video editing with MOV output

Task lifecycle

Poll the retrieve endpoint at a reasonable interval. Treat succeeded, failed, cancelled, and the compatibility status completed as terminal. Do not treat a Seedance provider status of expired as terminal: continue polling the same task because it can resume and later succeed. GENGEN normalizes new responses where possible.
Creating a task reserves wallet balance. Retrieving a successful terminal result can finalize the corresponding charge. Deleting or cancelling a task changes remote and local task state.

Compatibility

Older integrations may still send provider-native content[] bodies or snake_case fields. They remain compatibility inputs, but new integrations should use assets, controls, and providerOptions.

Seedance 2.5 Draft workflow

Draft generation is available through the public API; Playground does not expose Draft controls. Use the standard Seedance 2.5 endpoint with content filtering enabled. Flash does not support Draft workflows. Create a 480p preview:
Poll the returned GENGEN task ID until completion, then use that ID to create a final video:
Both requests use POST /api/gengen/v1/contents/generations/tasks and the same API key. The draft must have completed successfully and remain within its validity period, measured from its creation timestamp. Drafts created at or after October 28, 2026, 10:00 (UTC+8) are valid for 14 days; earlier drafts are valid for seven days. Use the exact public task ID returned by the create request, including for queued tasks. Final responses include draftTaskId containing the original public ID.
  • Draft creation defaults to and only accepts 480p. Finalization accepts 1080p (default) or 4k.
  • Omit prompt, mode, original media, duration, aspect ratio, seed, audio generation, and task-type hints from finalization requests, even if their values match the draft.
  • You may set output format, watermark, priority, return-last-frame, callback URL, safety identifier, task expiry, and the default service tier again. Omitted optional settings use the documented defaults rather than the draft’s values.
  • The draft and final video are separate billable tasks with separate balance reservations. Final-video pricing uses whether the original draft request contained input video. The original input-video duration is reused; the generated draft is not additional video input.
Provider-native draft: true and content: [{"type":"draft_task","draft_task":{"id":"seedance25:..."}}] are accepted for compatibility. Prefer the normalized fields above for new integrations.

Seedance priority and playback

controls.priority accepts an integer from 0 to 9 for Seedance 2.0 and 2.5. Higher values request earlier processing of waiting tasks, without interrupting running work. Priority is a scheduling hint and does not guarantee start order. Seedance 2.5 does not support the flex service tier. Seedance 2.5 1080p and 4K MP4 outputs use 10-bit H.265/HEVC. Browser playback support varies; a successful generation may require downloading the original and opening it in VLC, mpv, or QuickTime. Professional MOV output may also require a compatible player.