Seedance API Parameters: Model IDs, Settings and Output
Seedance API parameters explained: which model id to pick, resolution and duration limits, Native vs Upscaled, safe create requests and output file handling.
By OceanToken Team

Editor's note: Most Seedance API problems we hear about are not code problems. Someone picked a model id that cannot do what the request asks, left a setting to chance, or trusted a status field instead of the file. The create call is the easy part. This guide is about the choices around it: which id, which resolution and duration, how to send a create safely, and what to check once the video comes back.
Short answer: The Seedance API parameters that matter most are the model id, resolution and duration. On our API, Upscaled ids take 720p or 1080p at 4 to 15 seconds, and Seedance 2.5 Native takes 480p or 720p up to 30 seconds. Send an Idempotency-Key on create, then download and check the file.
If you just want a first clip working in Python, start with our Seedance API quickstart. This page assumes the basic create, poll and download loop already runs and goes one level deeper.
Seedance API parameters: start with the model id
Seedance is ByteDance's video model, and on our API the model id decides more than any other field. It sets which version you get, how the picture is produced, and which resolutions and durations the request may ask for. Four ids are listed on the /seedance-api page (checked 2026-11-02):
| Model id | Type | Resolution | Duration |
|---|---|---|---|
bytedance/seedance-2.0-upscaled |
Upscaled | 720p, 1080p | 4 to 15 s |
bytedance/seedance-2.5-upscaled |
Upscaled | 720p, 1080p | 4 to 15 s |
bytedance/seedance-2.5 |
Native | 480p, 720p | Up to 30 s |
bytedance/seedance-2.0 |
Native | See the model page | See the model page |
4K is not self-serve. The same page says it goes through sales, so a public id will not give you a 4K file however you set the fields.
Native vs Upscaled: what the suffix changes
Native ids call ByteDance's official API, one request per official task. Upscaled ids take the official render and run it through our own super-resolution step, as the Upscaled vs Native comparison explains.
That difference tells you when to use each. If you need long shots, Seedance 2.5 Native is the id that lists durations up to 30 seconds, while both Upscaled ids stop at 15. If you need 1080p from a self-serve id, the Upscaled ids are the ones that list it. And if a client wants output that comes straight from ByteDance's renderer with nothing added, choose Native and confirm that the id lists the resolution you need.
Because both variants sit behind the same endpoint and key, switching is a one-line change to model. That makes it cheap to render the same prompt both ways and let the people who will approve the clip decide.
Resolution and duration: set them on purpose
Always send resolution and duration yourself instead of relying on whatever the model would pick. Explicit values make cost and framing predictable from one job to the next, and they make your logs readable when you compare runs a week later.
A two-pass habit saves most of the money. Draft at 720p and a short duration, because problems such as drifting motion or a badly framed subject already show at 720p, and a higher resolution only makes the same mistake more expensive. Render at 1080p only the shots that pass that first round.
Keep each request inside the limits of its id. Asking an Upscaled id for 20 seconds, or 2.5 Native for 1080p, is a request the model does not offer. If a job fails right after you submit it, an out-of-range setting is one of the first things to check, as our Seedance failure checklist walks through.
Request options: a create you can safely retry
Video jobs are slow, so network timeouts on the create call happen. The danger is a client library that quietly retries the create and starts a second job for the same clip. The video guide covers this with an Idempotency-Key header on create: a retried create with the same key does not start a second job.
curl https://api.oceantoken.ai/v1/videos \
-H "Authorization: Bearer $OCEANTOKEN_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: shot-042-take-1" \
-d '{"model": "bytedance/seedance-2.5", "resolution": "720p", "duration": 10, "prompt": "Handheld walk through a night market, warm lantern light, shallow focus"}'
Build the key from something stable in your own system, such as a shot id plus a take number, so the same intent always maps to the same key. A new take gets a new key on purpose. For reference inputs and any other fields, check the field list in the video guide rather than guessing names.
Output handling: the file is the source of truth
Once the create returns an id, poll GET /v1/videos/{id}. The /seedance-api page shows a job moving from queued to completed, or ending as failed. When it completes, fetch the file from GET /v1/videos/{id}/content:
curl -sS https://api.oceantoken.ai/v1/videos/$VIDEO_ID/content \
-H "Authorization: Bearer $OCEANTOKEN_API_KEY" -o shot-042-take-1.mp4
ffprobe -v error -select_streams v:0 \
-show_entries stream=width,height,duration -of csv=p=0 shot-042-take-1.mp4
Copy the file into your own storage right away, so the rest of your pipeline never depends on downloading it again. Then check its pixel size and length with ffprobe against what you asked for. A status of completed tells you the job ended; only the file tells you what you got.
Log one row per job: video id, model id, resolution, duration, prompt, idempotency key and file path. When a job fails, store the whole response body as well. That log is what lets you answer "which settings produced the take the client approved?" without guessing.
How to test this on your workload
- Pick three prompts you actually plan to ship.
- Render each at 720p with an explicit duration on one Native and one Upscaled id.
- Run
ffprobeon every file and compare width, height and length with the request. - Send one create twice with the same
Idempotency-Keyand confirm only one job appears. - Have the person who approves clips pick between Native and Upscaled without being told which is which.
Our take: Choose the model id before you touch any other parameter, because it fixes the resolution and duration you are allowed to ask for. Set both explicitly, draft at 720p, and send an idempotency key on every create. Then trust the downloaded file, not the status field, when you decide whether a render is done.
Running it on OceanToken
Point any OpenAI-style client at https://api.oceantoken.ai/v1 and pass one of the four ids above to POST /v1/videos. Native and Upscaled Seedance sit behind one key, so switching between them, or from Seedance to Kling, DeepSeek or Qwen, means changing the model string rather than adding another integration. The /seedance-api page also says failed generations are not charged: the cost is held while a clip renders and released if it fails. See the current rate on /seedance-api.
FAQ
Which Seedance model id supports the longest clips? bytedance/seedance-2.5 (Native) lists durations up to 30 seconds. The Upscaled ids stop at 15 seconds.
Can I get 1080p Seedance output? Yes, from bytedance/seedance-2.0-upscaled or bytedance/seedance-2.5-upscaled. Seedance 2.5 Native lists 480p and 720p.
Can I request 4K? Not from the public ids. The /seedance-api page says 4K goes through sales.
What does the Idempotency-Key header do? A retried create with the same key does not start a second job, so a network timeout does not turn into two renders.
How do I know the output matches my request? Download the file and check width, height and duration with ffprobe. Do not rely on the status alone.
Sources
- OceanToken Seedance API page (checked 2026-11-02)
- OceanToken video guide (checked 2026-11-02)
- OceanToken: Seedance 2.0 Upscaled vs Native (checked 2026-11-02)
Get your OceanToken API key
- Create an OceanToken account and an API key.
- Copy the snippet below, paste your key, and run it.
- Check the current rate on /seedance-api.
curl https://api.oceantoken.ai/v1/videos \
-H "Authorization: Bearer $OCEANTOKEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "bytedance/seedance-2.5", "resolution": "720p", "duration": 10, "prompt": "Handheld walk through a night market, warm lantern light, shallow focus"}'
More from the blog
A Unified API for Chat, Image and Video
Building on a unified API for chat, image and video: sync vs async calls, billing units, retries, timeouts and storage, with three calls and a launch checklist.
Kimi K3 API International Access Guide
Use the Kimi K3 API internationally: sign up with email, pay by card, create one key and call moonshotai/kimi-k3 with an OpenAI SDK. Steps, code and checks.
Call Tencent Hy3 With the OpenAI SDK
Call Tencent Hy3 with the OpenAI SDK outside China: base URL, model id, streaming, tool calls, reasoning_effort values, Hy3 vs Hy3 preview, and token billing.