Seedance 2.5 API - Parameters, Schema & How to Call
Complete Seedance 2.5 API reference. Build the ModelArk request body with content, ratio, resolution, duration, generate_audio and output_format — copy it free.
Describe your scene — Seedance 2.5 generates up to 30 seconds of native audio-visual output.
Tip: put spoken dialogue in "double quotes" for better synced audio.
a cinematic scene. with natural ambient sound.
{
"model": "dreamina-seedance-2-5-260628",
"content": [
{
"type": "text",
"text": "a cinematic scene. with natural ambient sound."
}
],
"resolution": "720p",
"ratio": "16:9",
"duration": 10,
"generate_audio": true,
"camera_fixed": false,
"output_format": "mp4"
}Your files are processed locally in your browser. Nothing is uploaded to any server.
How to Call the Seedance 2.5 API
Build the Request
Set the model to dreamina-seedance-2-5-260628 (global) or doubao-seedance-2-5-260628 (China) and add your prompt as a content text item. The builder above writes the full JSON for you.
Set Your Parameters
Choose ratio, resolution (480p or 720p), duration up to 30 seconds, generate_audio, camera_fixed and output_format (mp4 or mov). Add a seed if you want reproducible results.
Send and Poll
POST the body to the ModelArk video generation endpoint, then poll the task or use a callback_url to receive queued, running, succeeded or failed status updates.
Seedance 2.5 API FAQ
What does a Seedance 2.5 API request look like?
The request takes a model ID and a content array. The content array holds a text item (your prompt, up to 5000 characters) plus optional image_url items with a role of first_frame, last_frame or reference_image. You then set ratio, resolution, duration, generate_audio, camera_fixed and output_format alongside it.
Which resolutions and durations does the API support?
At launch the API supports 480p and 720p (720p is the default), with 1080p and 4K planned. Duration runs from 4 to 30 seconds in a single generation. You can also drive length precisely with the frames parameter, which takes priority over duration.
Is there a cheaper tier for batch jobs?
Yes. The service_tier parameter accepts flex, an offline mode priced at roughly 50% of online inference with higher daily quota — ideal when latency is not critical. The default tier is online inference for time-sensitive requests.