Seedance 2.5

Call the domestic Doubao and overseas BytePlus Dreamina Seedance 2.5 through GregAPI. Seedance 2.5 offers both a domestic Doubao and an overseas BytePlus Dreamina variant. The two share GregAPI’s task interface and request structure, but differ in model names, pricing, and web-search capability. When you need to use real-person imagery that has been verified by the person themselves, first complete the H5 verification and asset ingestion for the relevant region through the Real-Person Asset Library API, then create tasks with asset://<AssetId>.

1. Models and pricing

Region	GregAPI stable model	Official version ID	No video input	With video input
Domestic	`doubao-seedance-2-5`	`doubao-seedance-2-5-260628`	¥70/M tokens	¥42/M tokens
Overseas	`dreamina-seedance-2-5`	`dreamina-seedance-2-5-260628`	$10.70/M tokens	$6.40/M tokens

We recommend using the stable model name in requests. The platform also accepts the corresponding official version ID and maps it to the stable model automatically; query and list responses still echo the stable model name. The video tier means content[] contains video_url; otherwise the novideo tier applies. 480p and 720p share the same unit price; only successfully generated videos are billed, and the actual video usage is based on usage.completion_tokens in the query response. The domestic model additionally supports tools: [{"type":"web_search"}] for text-only tasks. Web search is billed separately according to the actual usage.tool_usage.web_search count in the response; the overseas model does not support this tool, and specific customer pricing is subject to the GregAPI console.

2. Endpoint

export BASE_URL="https://api.gregapi.com"
export TOKEN="oh-xxxxxxxxxxxxxxxx"

Feature	Path
Create task	`POST /volcark/api/v3/contents/generations/tasks`
Query task	`GET /volcark/api/v3/contents/generations/tasks/{task_id}`
List tasks	`GET /volcark/api/v3/contents/generations/tasks`
Cancel or delete task	`DELETE /volcark/api/v3/contents/generations/tasks/{task_id}`

All requests use the GregAPI API Token:

Authorization: Bearer <TOKEN>
Content-Type: application/json

3. Create task

Domestic text-to-video example:

curl -X POST "$BASE_URL/volcark/api/v3/contents/generations/tasks" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2-5",
    "content": [
      {
        "type": "text",
        "text": "A 5-second cinematic product clip with soft natural light and a slow push-in."
      }
    ],
    "resolution": "720p",
    "ratio": "16:9",
    "duration": 5,
    "generate_audio": true,
    "output_format": "mp4",
    "watermark": false
  }'

For overseas calls, simply change model to dreamina-seedance-2-5; the endpoint still uses GregAPI’s BASE_URL. A successful creation returns the task ID:

{
  "id": "cgt-202608130001-example",
  "model": "doubao-seedance-2-5"
}

Multimodal references

content[] supports the following inputs:

Type	`role`	Max count
`image_url`	`reference_image`, `first_frame`, `last_frame`	30
`video_url`	`reference_video`	10
`audio_url`	`reference_audio`	10

Images, videos, and audio together may total at most 50 items. GregAPI-managed assets can be referenced with asset://<Asset_Id>; assets must be uploaded and processed before submitting a generation task.

{
  "model": "doubao-seedance-2-5",
  "content": [
    {
      "type": "text",
      "text": "Keep the camera motion of the reference video, use the product look of the reference image, and follow the rhythm of the reference audio."
    },
    {
      "type": "video_url",
      "role": "reference_video",
      "video_url": {"url": "asset://video-asset-id"}
    },
    {
      "type": "image_url",
      "role": "reference_image",
      "image_url": {"url": "asset://image-asset-id"}
    },
    {
      "type": "audio_url",
      "role": "reference_audio",
      "audio_url": {"url": "asset://audio-asset-id"}
    }
  ],
  "resolution": "720p",
  "ratio": "adaptive",
  "duration": -1
}

First-frame/last-frame tasks must use ratio=adaptive; last_frame must appear together with first_frame, and first/last frames cannot be mixed with reference_* assets.

Domestic web search

Web search is supported only by the domestic doubao-seedance-2-5, and content[] must contain text only:

{
  "model": "doubao-seedance-2-5",
  "content": [
    {
      "type": "text",
      "text": "Generate a tech-news-style short clip based on recent public information."
    }
  ],
  "resolution": "720p",
  "duration": 5,
  "tools": [
    {"type": "web_search"}
  ]
}

4. Parameter boundaries

Field	Seedance 2.5 constraint
`resolution`	Only `480p`, `720p`; `1080p` and `4k` are not supported
`duration`	Integer from `4-30`, or `-1` for model auto-selection
`ratio`	`adaptive`, `21:9`, `16:9`, `4:3`, `1:1`, `3:4`, `9:16`
`output_format`	`mp4`, `mov`
`generate_audio`	Boolean; synchronous audio generated by default
`return_last_frame`	Boolean; a successful response may return a last-frame URL
`priority`	`0-9`
`execution_expires_after`	`3600-259200` seconds
`service_tier`	Only `default` is supported
`draft`	Only `false` is supported

seed, frames, frames_per_second, frame_rate, fps, and camera_fixed do not apply to Seedance 2.5; passing them returns HTTP 400 before the request reaches the model service.

5. Query result

curl "$BASE_URL/volcark/api/v3/contents/generations/tasks/cgt-202608130001-example" \
  -H "Authorization: Bearer $TOKEN"

Task statuses include queued, running, succeeded, failed, and expired. A successful task returns content.video_url, the actual specs, and usage:

{
  "id": "cgt-202608130001-example",
  "model": "doubao-seedance-2-5",
  "status": "succeeded",
  "content": {
    "video_url": "https://example.com/generated-video.mp4"
  },
  "usage": {
    "completion_tokens": 108000
  }
}

Successful video URLs are typically valid for 24 hours and are subject to download limits; please transfer them promptly. Failed tasks should read the structured error in the response.

6. OpenAI-style endpoint

Seedance 2.5 can also create tasks via POST /v1/videos:

curl -X POST "$BASE_URL/v1/videos" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2-5",
    "prompt": "A paper boat floating on the water in the early morning",
    "seconds": 5,
    "resolution": "720p",
    "ratio": "16:9",
    "generate_audio": true
  }'

The OpenAI-style endpoint is suitable for basic text-to-video and reference-asset calls. When you need full native capabilities such as content[].role, output_format, and tools.web_search, use /volcark/api/v3/contents/generations/tasks.

Large Language Models (LLM)

Image Models

Video Models

Speech Models

1. Models and pricing

2. Endpoint

3. Create task

Multimodal references

Domestic web search

4. Parameter boundaries

5. Query result

6. OpenAI-style endpoint

​1. Models and pricing

​2. Endpoint

​3. Create task

​Multimodal references

​Domestic web search

​4. Parameter boundaries

​5. Query result

​6. OpenAI-style endpoint

1. Models and pricing

2. Endpoint

3. Create task

Multimodal references

Domestic web search

4. Parameter boundaries

5. Query result

6. OpenAI-style endpoint