Generate Video
generation_video_createDestructiveIdempotent
Queue video generation, image animation, source-video editing or character replacement into the media library. Editing accepts referenceVideoAssetIds and transformation controls, including original-audio preservation. Spends AI credits and returns a job readable with generation_jobs_get.
Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| seed | No | Optional seed for reproducible output. | |
| model | No | Video model. generation_options_list publishes generation type, source limits, character controls, resolution and pricing for each model. | kling-v3-pro-t2v |
| prompt | Yes | What should happen in the shot (3-5000 characters). | |
| folderId | No | Optional media library folder ID. | |
| provider | No | Explicit provider must support the model and input mode. auto uses a configured compatible provider. | auto |
| provenance | No | Original user request before prompt authoring, and previous failed job ID when retrying. Retained for review; never infer missing wording. | |
| resolution | No | Output resolution. Call generation_options_list for model-specific supported values and defaults. Unsupported settings are rejected. | |
| aspectRatio | No | Output aspect ratio. | 9:16 |
| characterId | No | Workspace character to keep consistent (characters_list). Its reference images come first (@Image1 is the primary), capped to the model limit. Only models whose parameters.character_id.supported is true in generation_options_list. | |
| workspaceId | No | Optional workspace ID. Defaults to the user's default workspace. UUID string. | |
| enhancePrompt | No | Enhance using BlitzReels model grammar and actual input context. | |
| generateAudio | No | null uses model audio behavior. Explicit true/false must be supported by the selected model. | |
| sourceAssetId | No | First-frame still. Required only when that model's parameters.source_asset_id.required is true. Optional on reference-capable I2V including Seedance 2.5. Text-to-video rejects it. Cannot combine with reference arrays and end_frame_asset_id. | |
| idempotencyKey | No | Retry key. Reuse only with identical inputs. | |
| negativePrompt | No | What to avoid in the shot. | |
| transformation | No | ||
| durationSeconds | No | Video transformations inherit the processed source duration and aspect ratio; duration_seconds is not a trim request. Choose a value from the selected model's durations_seconds. duration_seconds_min and duration_seconds_max are the bounds of that list, not a continuous range. | |
| endFrameAssetId | No | Last-frame still. Interpolates from the source first frame to this image across the full duration. Requires a source first frame; cannot combine with reference arrays. | |
| referenceAssetIds | No | Ordered reference images: Seedance 2.5 up to 30; Seedance 2.0 ref2v up to 9; other reference models up to 4. Source counts toward the limit. | |
| referenceAssetUris | No | BytePlus ModelArk real-human assets as asset://ID, for people who verified themselves in the BytePlus console. Requires provider byteplus-modelark and a model whose parameters.reference_asset_uris.supported is true in generation_options_list. They follow referenceAssetIds in @ImageN numbering and count toward the reference limit. | |
| referenceVideoRange | No | Cut this range from the reference video before a character swap or video edit; billed for the range length. | |
| referenceAudioAssetIds | No | Reference audio: Seedance 2.5 up to 10; Seedance 2.0 ref2v and fast up to 3. Other models reject them. | |
| referenceVideoAssetIds | No | Video transformations require one processed source video. Other reference modes have model-specific counts published by generation_options_list. |
Output Schema
| Name | Required | Description | Default |
|---|---|---|---|
| generation | Yes |