Seedance 2.0 video generation

Seedance 2.0 is the current video generation model on ATP. It supports text-to-video, image-to-video, first and last frame control, reference images, optional generated audio, and async task polling.

Model id

Use seedance-2-0 in the model field. Other Seedance variants may exist in internal pricing data, but they are not exposed in the public catalog yet.

Parameters

Field	Accepted values	Notes
`model`	`seedance-2-0`	Required.
`resolution`	`480p`, `720p`, `1080p`, `4k`	`480p` is best for draft tests. 4K uses its own billing tier.
`ratio`	`16:9`, `9:16`, `4:3`, `3:4`, `1:1`, `21:9`, `adaptive`	Cost follows rendered pixels, so square video is cheaper than 16:9 at the same tier.
`duration`	integer seconds, `4` to `15`	Use `4` for the cheapest functional test.
`generate_audio`	`true`, `false`	Ask the model to generate audio with the video.
`return_last_frame`	`true`, `false`	When enabled, completed task output can include the final frame URL.
`content`	text plus optional media parts	Use `role` on image/video/audio parts to explain how the asset should be used.

Media roles

Role	Purpose
`first_frame`	The starting frame for image-to-video.
`last_frame`	The ending frame for controlled transitions.
`reference_image`	Visual, style, or character reference. This is not the same as a boundary frame.
`reference_video`	Motion or style reference video.
`reference_audio`	Audio reference when supported by the selected workflow.

Text to video

curl "$ATP_BASE_URL/v1/media/v1/contents/generations/tasks" \
  -H "Authorization: Bearer $ATP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2-0",
    "content": [
      {
        "type": "text",
        "text": "A cinematic tracking shot of Taipei at night, neon reflections, smooth camera motion"
      }
    ],
    "resolution": "480p",
    "ratio": "16:9",
    "duration": 4,
    "generate_audio": false
  }'

The create response returns a task id such as cgt_....

Image to video

Use an HTTPS image URL, base64 image URL, or an uploaded asset://... reference.

{
  "model": "seedance-2-0",
  "content": [
    {
      "type": "text",
      "text": "Animate this product shot with a slow premium studio push-in."
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://example.com/product.jpg"
      },
      "role": "first_frame"
    }
  ],
  "resolution": "720p",
  "ratio": "16:9",
  "duration": 4,
  "generate_audio": false,
  "return_last_frame": true
}

First and last frame

{
  "model": "seedance-2-0",
  "content": [
    {
      "type": "text",
      "text": "Create a smooth transition from morning to night with consistent camera framing."
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "asset://tya_first_frame"
      },
      "role": "first_frame"
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "asset://tya_last_frame"
      },
      "role": "last_frame"
    }
  ],
  "resolution": "720p",
  "ratio": "16:9",
  "duration": 6,
  "return_last_frame": true
}

Poll the task

curl "$ATP_BASE_URL/v1/media/v1/contents/generations/tasks/cgt_xxx" \
  -H "Authorization: Bearer $ATP_API_KEY"

Poll until the task reaches succeeded, failed, or canceled. Successful responses include the generated video URL and usage metadata.

Billing

Video is billed on video tokens:

video_tokens = width × height × seconds × 24 ÷ 1024
cost = video_tokens × resolution_rate

Failed tasks are not billed. Use 480p, duration: 4, and generate_audio: false for the cheapest end-to-end smoke test.

← PreviousMedia models Next →Provider routing