Models · Video
Seedance video API for businesses in Southeast Asia
Seedance, ByteDance's video generation line, is opening with early access on GotoRamp as four models behind one asynchronous API, for businesses in Singapore, Malaysia, the Philippines and Thailand.
Four models
Pick by the stage of the job.
All four share one request format and one balance. Draft on Mini or Fast, then render the approved shot on 2.0 or 2.5 by changing the model field.
| Model ID | Use it for | Trade-off |
|---|---|---|
seedance-2.5 | Hero shots and final renders | Newest model in the line; save it for shots you have already approved |
seedance-2.0 | Production ads, social clips and product video | Full multimodal references and native audio |
seedance-2.0-fast | Iterating on framing, motion and timing | Quicker turnaround for iteration rather than final delivery |
seedance-2.0-mini | Storyboards, prompt tests and bulk drafts | Lowest cost in the line; use it to find the shot, not to deliver it |
GotoRamp LLC isn't affiliated with ByteDance. The model IDs are GotoRamp's own and stay stable when the underlying model version changes; we announce version changes at least 30 days ahead. Exact limits per model are in the API reference.
Capabilities
What the Seedance 2.0 line can do.
Every mode runs through POST /v1/video/generations. You choose it by what you attach to the request.
These figures describe the Seedance 2.0 line. The console lists the exact limits for each model, including seedance-2.5.
- Clip length4 to 15 seconds, set per request
- Aspect ratios21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, plus adaptive to follow the first frame
- Resolution480p or 720p
- AudioNative audio generated with the picture; turn it on or off per request
- Text to videoDescribe subject, action, camera, light and sound in the prompt
- Image to videoAnimate a first frame (
image), with an optional last frame (metadata.last_frame_image) to set where the shot lands - ReferencesReference images, video and audio: up to 9 images, 3 video clips and 3 audio clips on
seedance-2.0
How a job runs
Submit, poll, download, store.
Video jobs are asynchronous: you get a task ID straight away and the MP4 when it's ready.
- 1 · Submit
Create the job
POST /v1/video/generationsreturns atask_idwith the statusqueued. - 2 · Poll
Check the status
Call
GET /v1/video/generations/{task_id}until it reportssucceededorfailed. - 3 · Download
Fetch the MP4
A succeeded job has a
urlto the video. The link works for 24 hours. - 4 · Store
Keep your copy
Copy the file to your own storage; we don't keep it after 24 hours.
First request
A vertical draft on Mini.
Five seconds at 480p with audio off: the lowest-cost way to test a prompt before you spend on a final render.
Step-by-step setup is in the quickstart. Prompt structure is in the Seedance prompt guide, and choosing ratio and resolution per placement is in video formats.
# Submit a draft on Mini, then poll GET /v1/video/generations/TASK_ID
curl https://api.gotoramp.ai/v1/video/generations \
-H "Authorization: Bearer $GOTORAMP_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.0-mini",
"prompt": "Batik cloth unrolling across a teak table in morning light, slow overhead push-in",
"duration": 5,
"metadata": { "ratio": "9:16", "resolution": "480p", "generate_audio": false }
}'Billing and availability
Where Seedance is offered, and how it's billed.
Billing follows the seconds of video returned, per model and resolution, from a prepaid USD balance.
Not offered in the United States, Canada, the United Kingdom, Australia or New Zealand. Seedance isn't offered in those countries through its provider's official channel, so we don't offer it there either. Requests from countries we don't serve are refused.
- What you pay forSeconds of video returned, at the rate for that model and resolution
- Failed jobsNot billed
- Video retentionDownloadable for 24 hours, then not kept
- Prompts and referencesKept 30 days, then deleted unless needed for an abuse investigation
- TrainingWe don't use your prompts, references or outputs to train models
- Also not offeredMainland China, and countries and regions under US sanctions
FAQ
Seedance questions.
Before you publish anything, read the Acceptable Use Policy and the publishing checklist.
- Which Seedance model should I start with?
- Start with
seedance-2.0-miniat 480p to test prompts at the lowest cost, then render the approved shot onseedance-2.0orseedance-2.5. Switching is a change to themodelfield. - Why isn't Seedance offered in the United States, Canada or the UK?
- Seedance isn't offered through its provider's official channel in the United States, Canada, the United Kingdom, Australia or New Zealand, so GotoRamp doesn't offer it there. Requests from those countries are refused.
- How long can a Seedance clip be?
- Between 4 and 15 seconds on the Seedance 2.0 line, set per request in whole seconds. You're billed for the seconds of video returned.
- Can Seedance generate sound?
- Yes. The Seedance 2.0 line generates native audio with the picture; set
generate_audiotofalseinmetadatawhen you score in the edit. - How long do you keep my videos?
- Video links work for 24 hours after a job succeeds, and we don't keep a copy after that. Prompts and references are kept for 30 days, then deleted unless needed for an abuse investigation.
Text models
Write the script, then the shot.
The same account and balance also call text models through the OpenAI-compatible chat API. Compare all models.
- Doubao SeedByteDance Seed · general chat and multimodal models
- DeepSeekDeepSeek · chat and reasoning models
- QwenAlibaba Cloud · chat, reasoning and vision-language models
- KimiMoonshot AI · long-context and agent-oriented models
- GLMZhipu AI (Z.ai) · general-purpose models used for coding and agents
Start with a small balance. Scale when it works.
We're onboarding early-access accounts in small batches. Tell us what you're building; we reply within one business day.