Images (Seedream)

Generations

POST/v1/images/generations

curl https://www.link42.ai/v1/images/generations \
  -H "Authorization: Bearer $LINK42_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "dola-seedream-5-0-lite", "prompt": "Sunrise over the sea, film grain look", "size": "2K", "watermark": false}'

{
  "created": 1788300000,
  "data": [{"url": "https://…/image.png", "size": "2048x2048"}],
  "usage": {"generated_images": 1, "output_tokens": 0, "total_tokens": 0}
}

Edits

POST/v1/images/edits

For Volcengine Ark and BytePlus image-to-image, call /v1/images/generations with image set to a URL, Base64 data URI, or your own asset://<id>. These providers do not expose a separate upstream edits endpoint. /v1/images/edits applies only to other providers explicitly configured for that endpoint.

Asynchronous generation

POST/v1/images/generations/async

Takes the same body as a synchronous generation (stream is not supported) and answers 202 with a task id at once. Use it when a large or high-quality image takes long enough to run into the idle timeout of a CDN or reverse proxy. The task runs as the API key that submitted it: authentication, IP allowlist, rate limits, billing and routing are those of a synchronous call.

Poll GET /v1/images/tasks/{id}; status moves through queued, running, then succeeded or failed. A succeeded task carries result in the shape of the synchronous response, where data[].url is a temporary download link to the image in platform storage, signed afresh on every read (valid for 1 hour by default); tasks and their images are kept for 72 hours by default. A failed task carries error with the code and message a synchronous call would have returned.

A task turned away by a rate or concurrency limit does not fail: it goes back to the queue and runs later, and ends with that limit's error only after waiting more than 30 minutes. Each user may have 10 tasks queued or running at once by default; beyond that the submission answers 429 too_many_running_tasks. Submissions honour the Idempotency-Key header: a retry with the same key while the task is retained returns the task the first submission created (with the response header Idempotency-Replayed: true) instead of rendering and charging again; the same key with a different body answers 409 idempotency_conflict.

curl https://www.link42.ai/v1/images/generations/async \
  -H "Authorization: Bearer $LINK42_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "dola-seedream-5-0-lite", "prompt": "Sunrise over the sea, film grain look", "size": "4K"}'

{
  "id": "imgtask_7c10e3ddf2dd27209947d8b9c0b603b2",
  "object": "image.task",
  "model": "dola-seedream-5-0-lite",
  "status": "queued",
  "created_at": 1788300000
}

curl https://www.link42.ai/v1/images/tasks/imgtask_7c10e3ddf2dd27209947d8b9c0b603b2 \
  -H "Authorization: Bearer $LINK42_API_KEY"

{
  "id": "imgtask_7c10e3ddf2dd27209947d8b9c0b603b2",
  "object": "image.task",
  "model": "dola-seedream-5-0-lite",
  "status": "succeeded",
  "created_at": 1788300000,
  "finished_at": 1788300042,
  "expires_at": 1788559242,
  "result": {
    "created": 1788300041,
    "data": [{"url": "https://…/image-tasks/…/0.png?X-Amz-Signature=…"}],
    "usage": {"generated_images": 1}
  }
}

Model reference

Seedream extensions are validated per model: n>1 for Lite / 4.5 is adapted to Ark image-set fields, and model is rewritten to the bound upstream id. The maximum output count is held, while actual images are settled. Older rows below aid migration; the live catalogue determines what is available.

• data[].url is a signed upstream address with a limited lifetime. Copy it to your own storage if you need to keep it, or turn on result archiving in the console.