Skip to main content
These tables describe model limits. For the inputs and controls exposed by each format, see its creation endpoint: Ark / Seedance, OpenAI video, Gemini API, Vertex AI or Jimeng compatible.

Models

Every output is an MP4 at 24 fps with no watermark. Ark / Seedance defaults are 720p, 5 seconds, adaptive ratio and audio on. Other formats use the defaults in their respective sections. Aspect ratios: adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, 21:9. adaptive lets the model follow the prompt or the input image. Seedance 2.0 models output 864x496 and 496x864 at 480p for 16:9 and 9:16.

Pricing

A task reserves quota at creation and settles against the delivered video. Only successful tasks are charged.
output_frames is the delivered frame count (24 per second), width and height the delivered pixel size, and input_video_seconds the total length of reference videos. Ark task reads return billed usage as usage.completion_tokens. For other formats, check costs in the usage log. Rates in USD per million tokens: The seedance group ratio is shown on the console pricing page. Example: a 4 second 1280x720 Seedance 2.5 clip is 97 frames and 87,300 tokens, USD 0.93 at list rate before the group ratio.
Rates change when the model provider changes its prices. The console price at call time and the usage log are authoritative.

File and task retention

Successful video files remain available for 72 hours, after which downloads return 410. Task records remain available for 7 days. Download URLs require an API key; copy the file to your own storage before it expires. Retention does not change with the API format.