Seedance 2.5 · Video API · Asynchronous tasks

Seedance 2.5 API: access, pricing and production workflow

Use Seedance 2.5 for 15- or 30-second video tasks through an asynchronous workflow. On island AI Coding, Seedance is not part of the default public model list: contact support on WeChat GPT774 to confirm access, the endpoint, your API key permissions and the exact model ID before sending production traffic.

Reviewed September 19, 2026·Site availability and billing boundaries documented
Model IDseedance2.5
Durations15 / 30 seconds
WorkflowCreate / poll / download
AccessSeparate activation required

1. What this Seedance API endpoint is for

Seedance is useful when a team needs controlled text-to-video or image-to-video generation for product shots, social clips, storyboards and short narrative sequences. The production pattern is different from a chat completion: a client creates a video task, receives an identifier, checks that task later and downloads the result only after it reaches a successful terminal state.

The public website verifies that the video endpoint has an authentication gate, but that check alone does not prove rendering speed, quality, capacity or a service-level agreement. Those claims require successful tasks and measured samples. This page therefore separates verified site configuration from results that must be tested inside your own account. For provider-level product context, see the official Seedance 2.5 overview; account-specific fields and limits still come from your current access instructions.

Authentication probe for the island AI Coding video generation endpoint
An unauthenticated request returned HTTP 401 with API_KEY_REQUIRED. This verifies the authentication boundary, not model quality or generation speed.

2. Model IDs, durations and site pricing

Model IDDurationSite directory priceRecommended use
seedance2.0fast10 secondsCNY 1.50 per successful requestComposition and motion drafts
seedance2.515 secondsCNY 2.50 per successful requestShort social and product shots
seedance2.530 secondsCNY 3.50 per successful requestLonger continuous actions after validation

These are fixed per-successful-request directory prices, not a promise of per-second billing. A successful duplicate submission is a second billable task. Failed requests are described as non-billable in the current site directory, but the control panel remains the final billing record.

Prices are displayed in CNY because the current recharge flow supports WeChat Pay and Alipay. Confirm payment options and the final charged amount before committing a production budget.

3. Text, image, first-frame and reference-image workflows

A Seedance task can start from text alone or include an image. In practice, the image workflow is most useful when the frame already defines the product, character, composition or visual style. The prompt should then describe limited motion and explicitly name the elements that must remain stable.

  • Text-to-video: define the subject, one main action, camera movement, setting and final state.
  • Image-to-video: provide a reachable image URL and say what may move and what must not change.
  • First frame: prepare the target aspect ratio before submission instead of relying on destructive cropping later.
  • Reference image: use it to preserve identity or style, but still validate faces, hands, product structure and embedded text.

For practical prompt structures, use the Seedance 2.5 prompt guide.

4. Minimal request shape

The exact endpoint and supported fields must come from the access information provided for your account. The following object illustrates the task data that a client should keep together; it is not a substitute for the current control-panel documentation.

Image-to-video task object
{
  "model": "seedance2.5",
  "content": [
    {
      "type": "text",
      "text": "Keep the product shape unchanged. Slow push-in, one continuous shot, no text."
    },
    {
      "type": "image_url",
      "image_url": {"url": "https://example.invalid/first-frame.png"}
    }
  ],
  "duration": 15
}

Validate the image MIME type, size and URL lifetime before creating a task. Use only the duration values enabled for the selected model. A made-up value such as 12 or 20 seconds should fail locally rather than consume a network request.

5. Create, poll and download

StageClient responsibilityEvidence to store
CreateSubmit model, input, duration and idempotency keyRequest ID, task ID, prompt version
PollCheck status with bounded backoffStatus, error code, attempt count
DownloadFetch the result after a successful terminal stateMIME type, duration, file size, checksum
ReviewSeparate technical success from usable creative outputQA decision and rejection reason

Queueing and rendering can outlive the original HTTP request. Keep task state outside the browser or request worker. If the download expires or fails, retry the download path first; do not immediately create another video.

6. Retries, idempotency and error boundaries

Handle authentication, validation, rate limits and upstream failures as different classes. A 401 usually means key or activation problems; a 400 points to the payload; a 429 requires slower submission or polling; a 5xx may justify a bounded retry. Retrying every error with the same policy creates duplicate work and hides the actual cause.

  1. Create an idempotency key from the project, input asset, prompt version and requested duration.
  2. Persist the task ID before returning control to the caller.
  3. Use increasing polling intervals for queued or running tasks.
  4. Limit create retries and reconcile uncertain requests before resubmitting.
  5. Record download failures separately from generation failures.

7. Plan cost per usable video, not per request

A directory price tells you the cost of a successful task, not the cost of a publishable asset. If four 15-second tasks cost CNY 10 in total and only two pass creative review, the effective cost is CNY 5 per usable clip. Track structural defects, poor hooks, unsafe crops and failed text rendering as separate rejection reasons.

Use 10-second drafts to validate framing, then 15-second tasks for a single hook or product action. Reserve 30-second tasks for a sequence that genuinely benefits from continuous motion. For a social workflow, see the Seedance TikTok video guide.

8. Production launch checklist

  • Confirm separate Seedance activation with support.
  • Copy the endpoint, API key permission and model ID from the current account instructions.
  • Run one small task and verify create, poll and download.
  • Store task IDs and use idempotency keys before increasing concurrency.
  • Test the real aspect ratio and image URL lifetime.
  • Separate technical success, content review and billing reconciliation.
  • Set a daily spend limit and a maximum number of create retries.

9. Frequently asked questions

Can I use Seedance immediately after registering?

No. Seedance is not in the current default public model list. Contact support on WeChat GPT774 and confirm activation before integrating it.

Is Seedance billed per second?

The current island AI Coding directory lists fixed prices for successful 10-, 15- and 30-second requests. Do not convert those prices into a contractual per-second rate.

Does this page claim tested rendering quality?

No. The public authentication probe verifies the endpoint boundary only. Quality, latency and usable-output rate must be measured with successful tasks from your own account.

Where should I start with prompts?

Start with one subject, one achievable action and one camera movement. The English prompt guide includes templates for products, TikTok clips and short drama shots.

Reviewed September 19, 2026. Site model IDs, availability and charges remain subject to the control panel and access confirmation. Official product material is used for capability context; it is not treated as a guarantee of this site's account limits.