Models and parameters

Public model IDs, duration, image size and supported video settings.

Models

| Model | ID | Duration | Resolution | |---|---|---|---|---| | Seedance 2.0 | seedance-2.0 | 4–15 s | 480p, 720p, 1080p, 4K | | Seedance 2.5 | seedance-2.5 | 4–30 s or -1 | 480p, 720p, 1080p | | Seedance 2.0 Mini | seedance-2.0-mini | 4–15 s | 480p, 720p | | Seedance 2.0 Fast | seedance-2.0-fast | 4–15 s | 480p, 720p |

All four models accept text without references. Mini/Fast also require text when references are included; 2.0/2.5 also support reference media in their supported modes.

Use the exact public ID in model; available models depend on the key. Set seconds explicitly. For landscape or portrait 720p, use size; other resolutions and aspect ratios use the resolution and aspect_ratio extensions. Model-specific fields below are extensions to the OpenAI Videos format; access depends on your connection.

Request fields

Field	Meaning
`model`	Exact ID above
`prompt`	Text description; Mini/Fast require text; Fast accepts up to 4000 characters
`size`	`1280x720` or `720x1280` for landscape or portrait 720p
`seconds`	String containing a whole number in the model range; `"-1"` is an extension for automatic duration on 2.5
`resolution`	Extension: a resolution from the model table; must agree with `size` when both are supplied
`aspect_ratio`	`16:9`, `4:3`, `1:1`, `3:4`, `9:16`, `21:9`, `adaptive`; must agree with `size` when both are supplied
`generate_audio`	Boolean, default `true`; describe desired sound in text
`seed`	Integer `-1` through `2147483647`; matching seeds do not guarantee identical output
`watermark`	Boolean for AI Generated mark, default `false`
`return_last_frame`	Boolean, default `false`; `true` requests the final output frame as JPEG through `content.last_frame_url` on 2.0, 2.5, Mini and Fast
`output_format`	`mp4` by default; 2.5 also accepts `mov`
`omni_reference_task_type`	2.5: `auto`, `reference`, `edit`, `extend`
`draft`	2.5: `true` creates a 480p preview
`callback_url`	Public HTTPS endpoint for the final notification; see results for the payload and acknowledgement
`tools`	`[{"type":"web_search"}]` on 2.5, Mini and Fast; availability depends on the connection
`camera_fixed`	Boolean on 2.5 only; `camerafixed` is also accepted; do not send conflicting values
`nsfw_check`	Optional boolean; availability and behavior depend on the connection
`negative_prompt`, `generation_type`, `service_tier`	Optional nonempty strings; confirm accepted values for your connection; `service_tier` cannot be set on drafts or their final upgrades

Use seconds and size for the standard request. Use resolution, aspect_ratio and generate_audio explicitly for model-specific settings. JSON booleans are unquoted. See References for input media roles.

return_last_frame returns a frame from the generated video, while the input role last_frame supplies an image for generation. These are separate features. See Results for frame retrieval.

For first/last frames, editing and extension, use aspect_ratio: "adaptive". Editing on 2.5 requires seconds: "-1". First/last frames cannot be mixed with reference_image. On 2.5 they may be combined with video or audio references; on other models they cannot. A 30-second clip requires 2.5; Mini/Fast do not accept 1080p.