Guide

Seedream 5.5: transparency, layers and OCR

Preview guide

Paste into an AI assistant to help with integration. Documentation and examples only; no real credentials.

Preview guide

Seedream 5.5: transparency, layers and OCR

Direct answerUse the native or OpenAI-compatible image endpoints with layer_decomposition and layer_decomposition_options. Requests are non-streaming; availability and prices depend on the configured channel, group and token.

Updated · Reviewed

Endpoints and access

Use model doubao-seedream-5-5-261028 with this site's API key. Provider access, channel, group, token and pricing must be configured. See the model marketplace. This guide follows the early-access specification updated on November 8, 2026; the provider may revise it.

POST /api/v3/images/generations accepts native JSON. POST /v1/images/generations and POST /v1/images/edits provide compatible access; edits accept JSON or multipart reference images. Omit stream or set it to false. Sequential generation, search tools and fast prompt optimization are unsupported.

Transparent images

{"model":"doubao-seedream-5-5-261028","prompt":"A white rabbit sticker on a transparent background","background":"transparent","output_format":"png","watermark":false}

Generation accepts up to 10 URL/Base64 reference images. Background values are auto (default), opaque and transparent; transparent output cannot use JPEG. The default format is PNG. Generation size accepts 1K, 1.5K, 2K, or explicit dimensions with 921600–4624220 pixels and an aspect ratio of 1:16–16:1. Layers accept those named tiers or auto. Default sizes are 2K for generation and auto for layers.

Layers and OCR

{"model":"doubao-seedream-5-5-261028","image":"https://your-domain.example/poster.png","layer_decomposition":true,"layer_decomposition_options":{"max_images":30,"text_recognition":true},"size":"auto","response_format":"url","safety_identifier":"stable_user_hash"}

Layers require exactly one input image, allow an omitted prompt and reject background. The total output bound includes the base image: 2–100, default 30. Text recognition defaults to false and explicit false is preserved. An optional safety identifier should be a stable user hash of at most 64 printable ASCII characters.

Results retain z_index, bounding_box, name, description, background and text_recognition, including text lines, positions, runs, fonts and styles. Equal image URLs do not merge distinct layers. The requested format controls only the base image; individual layers are PNG.

Billing and retention

Per-image pricing reserves the requested maximum and settles the actual output count, including the base image, refunding the difference. Administrator expressions can define per-request or input-image pricing. No OCR surcharge is invented; configured prices remain authoritative.

Any failed layer or response exceeding the request bound fails the whole request and follows the normal refund flow. Prefer URLs for many layers and save image and OCR results promptly. Base64 responses remain subject to the existing byte budgets.

Use cases

  • Transparent assets
  • Multiple-reference editing
  • Poster layers
  • Text recognition

API protocols

  • /api/v3/images/generations
  • /v1/images/generations
  • /v1/images/edits

FAQ

How do I start using this model?

Check availability and pricing for your group in the model marketplace, then call an image endpoint with your site API key. An administrator must enable provider access and add doubao-seedream-5-5-261028 to a Volcengine channel.

Can I use n to generate multiple images?

Ordinary generation produces one image. The native endpoint rejects n and compatible endpoints accept only n=1. For layers, max_images ranges from 2 to 100 and defaults to 30, including the base image.

Does missing text_recognition indicate failure?

No. Only supported regular text layers return OCR. Non-text, artistic or handwritten text may omit it. A failed generated layer makes the entire request fail.

Official sources

  1. Seedream 5.5 early-access API documentation Official
  2. Volcengine image generation API Official