> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.empiriolabs.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.empiriolabs.ai/_mcp/server.
# Seedream 5.5
> Seedream image model with native transparent output, precise multi-round editing, and decomposition into up to 99 editable layers with text recognition.
[ByteDance](/providers/bytedance) · Image Generation
`POST /v1/images/generations`
Seedream image model with native transparent output, precise multi-round editing, and decomposition into up to 99 editable layers with text recognition.
## At a glance
| Field | Value |
| ------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Model id | `seedream-5-5` |
| Model release date | 2026-11-09 |
| Input modalities | Text, Image |
| Output modalities | Image |
| Context window | - |
| Weight precision | - |
| Region | Malaysia |
| Features | image_generation, image_editing, layer_decomposition, interactive_editing, multi_reference, text_rendering, multilingual_text, transparent_background |
| Native inference | No |
| New | Yes |
| Supported endpoints | `POST /v1/images/generations` |
| Alternate model ids | `bytedance/dola-seedream-5-5-261028`, `bytedance/seedream-5-5`, `byteplus/dola-seedream-5-5-261028`, `byteplus/seedream-5-5`, `dola-seedream-5-5-261028`, `seedream-5-5-261028`, `seedream-5.5` |
## Pricing
| Charge | Spec | Rate |
| ------------------- | --------------- | -------- |
| Output up to 2.61MP | per image | \$0.0332 |
| Output above 2.61MP | per image | \$0.0581 |
| Input image | per input image | \$0.0133 |
## Example request
```bash
curl https://api.empiriolabs.ai/v1/images/generations \
-H 'Authorization: Bearer $EMPIRIOLABS_API_KEY' \
-H 'Content-Type: application/json' \
-d '{"model": "seedream-5-5", "prompt": "a misty forest at dawn", "size": "1024x1024"}'
```
## Parameters
| Parameter | Type | Required | Default | Description |
| --------------------- | ------- | -------- | -------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `prompt` | string | yes | - | Text prompt. With image attached, becomes an editing or fusion instruction. |
| `image` | array | no | - | Reference image URL, data URI, or uploaded image object. Seedream 5.5 supports up to 10 reference images, and exactly one for layer decomposition. |
| `images` | array | no | - | Alias for image. Use this for multiple uploaded image references. |
| `aspect_ratio` | enum | no | `"auto"` | Auto lets the model choose dimensions within the selected resolution tier. Explicit ratios use the preset size for that tier. · Allowed: `auto`, `1:1`, `16:9`, `9:16`, `4:3`, `3:4`, `3:2`, `2:3`, `21:9` |
| `layer_decomposition` | boolean | no | false | Split one input image into a base image plus up to 99 transparent PNG layers, each returned with its stacking order, its position in the base image, a name, and a description. Requires exactly one input image. The prompt is optional and can name the elements or coordinate regions to separate. |
| `max_images` | number | no | `30` | Layer decomposition only. Maximum images returned, the base image included, from 2 to 100; the model decides the actual count. You are billed only for images returned. · Range: 2 – 100 |
| `text_recognition` | boolean | no | false | Layer decomposition only. Also return the text, line positions, and styling (similar font, color, bold, italic, underline, strikethrough, shadow, and translucency) of each regular text layer as text_recognition. Artistic text and handwriting are not recognized. |
| `quality` | enum | no | `"2K"` | Output resolution tier. 1.5K has the same price as 1K and provides higher resolution. Defaults to 2K. In layer decomposition, leaving this unset matches the input image instead. · Allowed: `1K`, `1.5K`, `2K` |
| `size` | string | no | - | Resolution tier or explicit pixel size such as 1K, 1.5K, 2K, or 2048x1152. Explicit pixels override aspect_ratio and quality. Layer decomposition accepts tiers or auto only. |
| `use_custom_size` | boolean | no | false | Use custom_width and custom_height instead of preset aspect_ratio and quality. Applies when aspect_ratio is auto. |
| `custom_width` | number | no | - | Custom output width in pixels. Total area must be in \[0.92MP, 4.62MP] and aspect must be in \[1:16, 16:1]. · Range: 15 – 8600 |
| `custom_height` | number | no | - | Custom output height in pixels. Total area must be in \[0.92MP, 4.62MP] and aspect must be in \[1:16, 16:1]. · Range: 15 – 8600 |
| `watermark` | boolean | no | false | When enabled, the generated image includes an AI-generated provenance watermark. EmpirioLabs does not add a separate watermark. |
| `output_format` | enum | no | `"png"` | Output image file format. In layer decomposition this sets the base image format, and layers are always returned as PNG. · Allowed: `jpeg`, `png` |
| `background` | enum | no | `"auto"` | Auto lets the model decide from the prompt and output format whether the image has transparency. Transparent returns an image with an alpha channel, from text or from up to 10 reference images, and requires png output. Opaque returns an image without one. Not available with layer decomposition. · Allowed: `auto`, `transparent`, `opaque` |
| `response_format` | enum | no | `"url"` | Return signed image URLs by default, or include base64 image data when b64_json is requested. · Allowed: `url`, `b64_json` |
## Notes
**Modes**
* Text to image
* Single or multi-image to image
* Interactive editing with coordinate or sketch markers
* Layer decomposition into a base image plus up to 99 layers, with optional text recognition
**Interactive editing**
Mark the region to change directly on the input image with a sketch, circle, or arrow, then describe it in the prompt. For exact placement, use coordinate tags on a 0 to 1000 normalized grid:
```text
Place the subject from image 1 179 283 796 986 at the position in image 2 118 331 933 871.
```
**Layer decomposition**
Set layer_decomposition to true with exactly one input image to split it into a base image plus up to 99 transparent PNG layers. max_images sets the most images returned, the base image included, from 2 to 100, and defaults to 30; a higher value allows a finer decomposition. Every returned image carries its stacking order as z_index, and each layer also carries its position in the base image as bounding_box.absolute and bounding_box.normalized, plus a name and a description. Scale each layer to its box when you reassemble the image. Leave the prompt empty to let the model choose the elements, or name the elements or coordinate regions to separate. Resolution accepts 1K, 1.5K, or 2K, and matches the input image when unset.
**Text recognition**
With text_recognition on, each regular text layer also returns text_recognition: its full text, the position of each line, and the styling of each run of text (a similar font, color, bold, italic, underline, strikethrough, shadow, and translucency). Artistic text and handwriting are not recognized.
**Transparency**
background defaults to auto, which lets the model decide from the prompt and output format. Transparent returns an image with an alpha channel, from text alone or from up to 10 reference images that need no alpha channel, and requires PNG output. Opaque returns an image without one. Each returned image reports its actual transparency as background. Layer decomposition does not take background, and its layers are always transparent PNG.
**Limits**
* One output image per request, except layer decomposition, which returns the base image plus its layers
* Up to 10 reference images, and exactly one for layer decomposition
* 1K, 1.5K, and 2K resolution tiers, or an exact size from 0.92MP to 4.62MP
* PNG or JPEG output, and layers are always PNG
* No grouped output, streaming, or web search
**Billing**
Each returned image bills at its own size: up to 2.61MP, which covers 1K and 1.5K, or above 2.61MP, which covers 2K. Every input image bills, the one for layer decomposition included. Layer decomposition bills every returned image, the base image and each layer, at its own size.
**Content safety**
Generated images pass an automated content-safety check. An image flagged as explicit is not returned. Because the image is generated before the check runs, a flagged generation is still billed.
---
*Machine-readable schema:* `GET https://api.empiriolabs.ai/v1/models/seedream-5-5`.