GPT Image 2.5 is live — OpenAI's newest image model, targeted edits that leave the rest of the frame alone
All resources

· reAPI Team

Seedream 5.5 API: Availability, Parameters and Next Steps

Seedream 5.5 API guide: where you can call it today, every request parameter, layer decomposition with OCR, per-image pricing and a working request example.

Seedream 5.5 API: Availability, Parameters and Next Steps

You can call a Seedream 5.5 API today on reAPI, as doubao-seedream-5-5 on the async images endpoint. The Seedream 5.5 API is open: ByteDance announced the model on November 9, 2026, and it is also live in BytePlus's Lumina studio.[1][2]

This guide covers what the Seedream 5.5 API accepts, how its sizes and layer decomposition work, what each request costs, and a request you can run. Parameter and price details describe reAPI's API as of November 10, 2026.

TL;DR

  • On reAPI, call POST /api/v1/images/generations with "model": "doubao-seedream-5-5", then poll the task for the image URL.[3]
  • Sizes are 1K, 1.5K or 2K, ten aspect ratios, or exact pixels. Up to 10 reference images. No 3K or 4K.[3]
  • Transparent output works from a prompt or from references, and layer decomposition returns up to 100 images with optional text recognition.[3]
  • On reAPI, each output image costs $0.013 at 1K or 1.5K and $0.023 at 2K, plus $0.0052 per reference image (prices at launch, November 10, 2026).[4]

Where the Seedream 5.5 API stands

WhereStatus on November 10, 2026
reAPI images APILive as doubao-seedream-5-5[3]
BytePlus LuminaSeedream 5.5 live in the studio[2]

ByteDance's own capability claims for the model come from its announcement: multi-turn editing that keeps unedited regions intact across 100 rounds, native transparent output, and clearer small text. In ByteDance's tests, severe text corruption was 0.4% and minor corruption 1.7%.[1]

Seedream 5.5 API parameters

The request body below is reAPI's Seedream 5.5 surface.[3]

FieldWhat it does
promptThe image or the edit. Required, except for layer decomposition.
sizeA tier (1K, 1.5K, 2K; default 2K), auto, a ratio, or exact WIDTHxHEIGHT.
resolution1K, 1.5K or 2K, paired with a ratio size.
image_urlsUp to 10 public reference image URLs.
backgroundauto, opaque or transparent. Transparent needs PNG output.
output_formatpng (default) or jpeg.
layer_decompositionSplit one image into a base image plus transparent layers.
layer_decomposition_optionsmax_images (2–100, default 30) and text_recognition.
optimize_prompt_options{ "mode": "standard" } only; optimization is always on.
watermarkAdd an "AI generated" mark.
nAlways 1.

Size is the field people trip on. The ten ratios are 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 21:9, 2:1 and 1:2; pair one with resolution to pick the tier. Exact pixels must total between 921,600 and 4,624,220 with a width-to-height ratio between 1/16 and 16, and are sent without resolution.[3]

Two rules differ from Seedream 5.0 Pro. Transparent output on 5.5 works for text-to-image as well as edits, while 5.0 Pro needs exactly one reference image for it. And 5.5's layer decomposition lets you set how many images come back, up to 100, where 5.0 Pro returns at most a base image plus 16 layers.[3][5]

A Seedream 5.5 API request you can run

curl https://reapi.ai/api/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedream-5-5",
    "prompt": "A frosted glass perfume bottle with pale gold liquid, soft studio light",
    "size": "16:9",
    "resolution": "2K"
  }'

The response carries a task id with status: "processing". Poll GET /api/v1/tasks/{id} until the status is completed, then read the image from output.image_urls. Generated URLs expire, so copy results to your own storage.[3]

Layer decomposition with text recognition

Send exactly one image, turn on layer_decomposition, and leave background out:

{
  "model": "doubao-seedream-5-5",
  "image_urls": ["https://example.com/poster.png"],
  "layer_decomposition": true,
  "layer_decomposition_options": { "max_images": 10, "text_recognition": true },
  "size": "1K"
}

The finished task lists the base image first and then each layer in output.image_urls, with matching entries in output.layers: stacking order, a bounding box in the base image's coordinates and, when OCR found regular text, a text_recognition object with the text, line boxes and style estimates.[3] Layer decomposition can take several minutes, so keep polling the same task rather than resubmitting.

When a Seedream 5.5 API request fails

Most failures show up in one of four ways, and only one of them costs anything:[3]

  • A synchronous 400 on submit. A field breaks a documented rule, such as a ratio outside the list, a jpeg with a transparent background, or layer mode without exactly one image. The error names the field and nothing is charged.
  • A 402 on submit. The balance cannot cover the reserve. Layer decomposition reserves max_images images, so lowering that cap helps.
  • A failed task. Content-policy rejections, unreachable image URLs and unsupported formats such as AVIF end the task as failed, and the reserve is refunded in full.
  • A long processing state. Layer decomposition can take several minutes. Keep polling the same task instead of submitting again, or you pay twice.

What the Seedream 5.5 API costs

reAPI bills Seedream 5.5 per output image by size band, plus every reference image; the first reference is not free.[3] At launch on November 10, 2026 the rates were:[4]

ItemPrice
Output image, 1K or 1.5K (up to 2,610,000 px)$0.013[4]
Output image, 2K (above 2,610,000 px)$0.023[4]
Each reference image$0.0052[4]

Examples, rounded up to whole credits (1 credit = $0.001): a 2K text-to-image request costs 23 credits; a 1K edit with one reference image costs 19 credits.[4]

Layer decomposition reserves max_images images at the request's size band and settles to the images actually returned, with the single input image billed once. A 1K split capped at 10 images reserves 136 credits; if it returns four images, it costs 58 credits and the rest is refunded. Failed tasks are refunded in full. For comparison, Seedream 5.0 Pro cost $0.032 per 1K image and $0.063 per 2K image on reAPI on the same date.[5]

Choosing size and references for a job

The price bands reward picking the smallest size that does the job. These are our suggestions, based on how the bands and inputs are billed rather than on quality tests:

  • Drafts and social images. 1.5K costs the same as 1K, so it is the better default for the lower band.
  • Print, posters and dense text. 2K puts the most pixels into small type, the area where ByteDance reports improved rendering.[1]
  • Product edits. Start with one clean reference image. Add more only when a second image carries something the first does not, such as a setting or a logo, because each one is billed.
  • Assets for layouts. Ask for background: "transparent" with PNG output instead of generating on white and cutting it out later.

Next steps for a Seedream 5.5 API integration

This checklist is our editorial advice.

  1. Start in the playground. The Seedream 5.5 page runs the same parameters and shows the cost before you generate.
  2. Use the async pattern. Submit, poll and copy the result.
  3. Keep the model ID in one setting so you can compare Seedream 5.5 with 5.0 Pro on the same prompts.
  4. Budget for references. Every reference image adds to the price, so test how many you really need.
  5. Cap layer splits. Set max_images to what you need; the reserve is sized from it.

FAQ

Is there a Seedream 5.5 API?

Yes. reAPI serves Seedream 5.5 through its images API as doubao-seedream-5-5.[3]

Is the Seedream 5.5 API free?

No. On reAPI it is pay-as-you-go with prepaid credits and no subscription: each output image plus each reference image is billed, and failed tasks are refunded.[4]

Is there a Seedream 5.5 API tutorial?

The Seedream 5.5 API docs walk through a request in cURL, Python, Node.js and Go, every parameter, layer decomposition output and billing. The example above is the shortest working request.[3]

Can I download Seedream 5.5?

No. ByteDance has not released Seedream 5.5 weights or a downloadable model; it runs as a hosted model in ByteDance's apps and through APIs.[1][2]

Where is the Seedream 5.5 text to image documentation?

For reAPI, it is the Seedream 5.5 API docs.

How do I get a Seedream 5.5 API key?

Create a key in the reAPI dashboard and pass it as a Bearer token. The same key calls Seedream 5.5 and every other model on your account.

Calling the Seedream 5.5 API today

The model is out and callable: send doubao-seedream-5-5 to reAPI's images endpoint, and start with the playground to tune size and references before you wire the Seedream 5.5 API into your code.

References

  1. ByteDance Seed. More Realistic Images, More Stable Editing Across 100 Rounds | Introducing Seedream 5.5. Retrieved November 2026 from seed.bytedance.com/en/blog/more-realistic-images-more-stable-editing-across-100-rounds-introducing-seedream-5-5
  2. BytePlus. Seedream 5.5 Release Date: Officially released November 9, 2026. Retrieved November 2026 from ai.byteplus.com/lumina/en/resource/seedream-5-5-release-date
  3. reAPI. Seedream 5.5 API docs. Retrieved November 2026 from reapi.ai/docs/seedream-5-5
  4. reAPI. Seedream 5.5. Retrieved November 2026 from reapi.ai/models/seedream-5-5
  5. reAPI. Seedream 5.0 Pro API docs and model page. Retrieved November 2026 from reapi.ai/docs/seedream-5-0-pro and reapi.ai/models/seedream-5-0-pro
All resources