Seedream Image Generation

Use this endpoint to generate images from text descriptions or reference images. It supports text-to-image, single-image-to-image, multi-image-to-image, and sequential (group) image generation; the Seedream 5.0 pro model additionally supports layer decomposition and interactive editing.

Endpoint

POST /v1/images/generations

Authentication

Method	Description
Bearer Token	Include Authorization: Bearer <API_key> in the request header to authenticate your account.

Request Headers

Parameter	Type	Required	Description
Content-Type	string	Yes	The media type of the request body. Set it to application/json
Authorization	string	Yes	Bearer API_key, used to authenticate your account

Model Capabilities

• Seedream 5.0 pro (doubao-seedream-5-0-pro-260628): Supports layer decomposition (splitting a single image into 1 base image + up to 16 independently editable layers) and interactive editing (specifying the edit area via coordinates, box selection, arrows, etc.); single-image output only (does not support sequential_image_generation), and does not yet support group generation, web search, or streaming output
• Seedream 5.0 lite (doubao-seedream-5-0-260128): Supports group generation (when sequential_image_generation is auto: up to 15 images for text-to-group, up to 14 for single-image-to-group), multi-image-to-image, single-image-to-image, and text-to-image; use disabled for single-image output only

Request Body

Parameter	Type	Required	Default	Description
model	string	Yes	-	Model ID. See "Supported Model IDs" below
prompt	string	Yes	-	Prompt for generating the image or specifying layer-decomposition intent. Chinese prompt recommended not to exceed 300 characters, English not to exceed 600 words
image	string / string[]	No	-	Reference image(s), as a URL or Base64 (data:image/<format>;base64,<encoded>, format must be lowercase); per-image requirements below
size	string	No	-	Image size. Available tiers and pixel ranges differ per model. See "Size" below
layer_decomposition	boolean	No	false	Layer decomposition switch: true enables layer decomposition mode (Seedream 5.0 pro only; then image is required and only a single image is allowed)
optimize_prompt_options	object	No	-	Prompt optimization configuration. See subtable below
output_format	string	No	jpeg	Output image format: png / jpeg. In layer decomposition mode, this only controls the base image format; layers are always png
background	string	No	opaque	Image alpha channel: transparent (transparent background) / opaque (opaque). Seedream 5.0 pro only; image-to-image only, with a single input image that has an alpha channel
response_format	string	No	url	Return format: url (download link, valid for 24 hours) / b64_json (Base64-encoded)
sequential_image_generation	string	No	disabled	Group mode: auto (model decides whether to return a group) / disabled (group off, single image only). Seedream 5.0 lite only
sequential_image_generation_options	object	No	-	Group configuration, effective only when sequential_image_generation is auto. See subtable below
stream	boolean	No	false	Streaming switch: true returns each generated image as it is produced. Only supported by Seedream 5.0 lite
tools	object[]	No	-	Tool configuration (Seedream 5.0 lite only), e.g., enabling web search. See subtable below
watermark	boolean	No	true	Watermark switch: false no watermark, true adds an "AI generated" watermark to the bottom-right corner

optimize_prompt_options Parameters

Parameter	Type	Required	Default	Description
optimize_prompt_options.mode	string	No	standard	Prompt optimization mode: standard (higher quality, slower); fast (faster, slightly lower quality, not supported by Seedream 5.0 lite)

sequential_image_generation_options Parameters

Parameter	Type	Required	Default	Description
sequential_image_generation_options.max_images	integer	No	15	Maximum number of images generated in this request, range [1, 15]. The actual number is also limited by the reference image count (input reference images + generated images ≤ 15)

tools Parameters

Parameter	Type	Required	Description
tools[].type	string	Yes	Tool type: web_search (web search). When enabled, the model decides whether to search the web based on the prompt; the search count can be queried via usage.tool_usage.web_search

image Notes

• Image generation scenario: Seedream 5.0 pro supports up to 10 reference images; Seedream 5.0 lite supports up to 14
• Single image requirements: formats jpeg / png / webp / bmp / tiff / gif / heic / heif; aspect ratio [1/16, 16]; side length > 14px; size ≤ 30MB; total pixels [196, 6000×6000]
• Layer decomposition scenario: image is required, single image only, formats png / jpeg, total pixels [512×512, 6000×6000]

Size (size)

Each model supports two configuration methods (do not mix): Method 1 specifies a resolution tier (the model judges the aspect ratio from the prompt); Method 2 specifies the exact width×height pixel value (widthxheight).

Model	Method 1 values (default)	Method 2 constraints
Seedream 5.0 pro (image generation)	1K / 1.5K / 2K (default 2K; 1.5K costs the same as 1K with better quality)	Total pixels [1280x720, 2048x2048x1.1025], aspect ratio [1/16, 16]
Seedream 5.0 pro (layer decomposition)	1K / 1.5K / 2K / auto (default auto, output follows the input dimensions; base image and layers preserve their source aspect ratio)	Method 2 not supported
Seedream 5.0 lite	2K / 3K / 4K (no default)	Default 2048x2048, total pixels [2560x1440, 4096x4096], aspect ratio [1/16, 16]

Resolution tier to pixel size reference for Seedream 5.0 pro (image generation scenario):

Resolution	Aspect ratio	Pixel size
1K	1:1	1024x1024
1K	4:3 / 3:4 / 16:9 / 9:16 / 3:2 / 2:3 / 21:9	1152x864 / 864x1152 / 1424x800 / 800x1424 / 1248x832 / 832x1248 / 1568x672
1.5K	1:1	1536x1536
1.5K	4:3 / 3:4 / 16:9 / 9:16 / 3:2 / 2:3 / 21:9	1792x1344 / 1344x1792 / 2048x1152 / 1152x2048 / 1872x1248 / 1248x1872 / 2352x1008
2K	1:1	2048x2048
2K	4:3 / 3:4 / 16:9 / 9:16 / 3:2 / 2:3 / 21:9	2368x1776 / 1776x2368 / 2816x1584 / 1584x2816 / 2496x1664 / 1664x2496 / 3136x1344

Request Example

Text-to-Image

cURL

curl --location --request POST 'https://neolink.com/api/v1/images/generations' \
     --header 'Content-Type: application/json' \
     --header 'Authorization: Bearer <API Key>' \
     --data-raw '{
         "model": "<Model ID>",
         "prompt": "A cute orange cat wearing round-framed glasses, sitting by the windowsill reading a newspaper in warm afternoon sunlight",
         "size": "2K",
         "response_format": "b64_json",
         "watermark": true
     }'

Python

import base64
import requests

url = "https://neolink.com/api/v1/images/generations"

headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer <API Key>",
}

payload = {
    "model": "<Model ID>",
    "prompt": "A cute orange cat wearing round-framed glasses, sitting by the windowsill reading a newspaper in warm afternoon sunlight",
    "size": "2K",
    "response_format": "b64_json",
    "watermark": True,
}

resp = requests.post(url, headers=headers, json=payload, timeout=60)
resp.raise_for_status()
data = resp.json()

# Extract the returned image and save it
for item in data["data"]:
    image = item["b64_json"]
    with open("output.png", "wb") as f:
        f.write(base64.b64decode(image))

Image-to-Image (Multi-Image-to-Image)

Pass reference image(s) in image (multiple allowed) together with a text prompt to generate an image.

Base64 Encoding Note: image accepts Base64 in the format data:image/png;base64,<Base64 encoded>. See the encoding script below:

#!/usr/bin/env bash
set -euo pipefail
IMAGE_PATH="./input.png"

if [[ "$(base64 --version 2>&1)" = *"FreeBSD"* ]]; then
    B64FLAGS="--input"
else
    B64FLAGS="-w0"
fi
echo -n "data:image/png;base64," > base64.txt
base64 $B64FLAGS "$IMAGE_PATH" >> base64.txt

cURL

curl --location --request POST 'https://neolink.com/api/v1/images/generations' \
     --header 'Content-Type: application/json' \
     --header 'Authorization: Bearer <API Key>' \
     --data-raw '{
         "model": "<Model ID>",
         "prompt": "Place the cat from the reference image in a seaside sunset scene, keeping a photorealistic style",
         "image": [
             "https://example.com/reference-cat.png",
             "data:image/png;base64,<Base64 Code>"
         ],
         "size": "2K"
     }'

Python

import requests

url = "https://neolink.com/api/v1/images/generations"

headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer <API Key>",
}

payload = {
    "model": "<Model ID>",
    "prompt": "Place the cat from the reference image in a seaside sunset scene, keeping a photorealistic style",
    "image": [
        "https://example.com/reference-cat.png",
        "data:image/png;base64,<Base64 Code>",
    ],
    "size": "2K",
}

resp = requests.post(url, headers=headers, json=payload, timeout=60)
resp.raise_for_status()
result = resp.json()
print(result)

Multi-Image Fusion

Pass multiple reference images; the model fuses elements from the images into a single image based on the prompt (e.g., swapping the figure's outfit in image 1 with the outfit in image 2).

cURL

curl --location --request POST 'https://neolink.com/api/v1/images/generations' \
     --header 'Content-Type: application/json' \
     --header 'Authorization: Bearer <API Key>' \
     --data-raw '{
         "model": "doubao-seedream-5-0-pro-260628",
         "prompt": "Replace the outfit in image 1 with the one in image 2",
         "image": [
             "https://example.com/figure-1.png",
             "https://example.com/outfit-2.png"
         ],
         "size": "2K",
         "output_format": "png",
         "watermark": false
     }'

Python

import requests

url = "https://neolink.com/api/v1/images/generations"

headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer <API Key>",
}

payload = {
    "model": "doubao-seedream-5-0-pro-260628",
    "prompt": "Replace the outfit in image 1 with the one in image 2",
    "image": [
        "https://example.com/figure-1.png",
        "https://example.com/outfit-2.png",
    ],
    "size": "2K",
    "output_format": "png",
    "watermark": False,
}

resp = requests.post(url, headers=headers, json=payload, timeout=60)
resp.raise_for_status()
result = resp.json()
print(result["data"][0]["url"])

Layer Decomposition

With layer_decomposition enabled, the model splits the input image into 1 base image + up to 16 independently editable layers (all with an alpha channel). In prompt, you can use <bbox>left top right bottom</bbox> to specify the coordinates of the elements to split.

cURL

curl --location --request POST 'https://neolink.com/api/v1/images/generations' \
     --header 'Content-Type: application/json' \
     --header 'Authorization: Bearer <API Key>' \
     --data-raw '{
         "model": "doubao-seedream-5-0-pro-260628",
         "prompt": "Perform precise layer decomposition on the image. The coordinates of the text to decompose are <bbox>180 64 812 198</bbox> and <bbox>178 714 826 810</bbox>; the parrot coordinates are <bbox>347 305 642 997</bbox>.",
         "image": "https://example.com/input.png",
         "layer_decomposition": true,
         "size": "2K",
         "output_format": "jpeg",
         "response_format": "url",
         "watermark": true
     }'

Python

import requests

url = "https://neolink.com/api/v1/images/generations"

headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer <API Key>",
}

payload = {
    "model": "doubao-seedream-5-0-pro-260628",
    "prompt": "Perform precise layer decomposition on the image. The coordinates of the text to decompose are <bbox>180 64 812 198</bbox> and <bbox>178 714 826 810</bbox>; the parrot coordinates are <bbox>347 305 642 997</bbox>.",
    "image": "https://example.com/input.png",
    "layer_decomposition": True,
    "size": "2K",
    "output_format": "jpeg",
    "response_format": "url",
    "watermark": True,
}

resp = requests.post(url, headers=headers, json=payload, timeout=60)
resp.raise_for_status()
result = resp.json()

# The base image has z_index 0; layers increment from 1
for item in result["data"]:
    print(item["z_index"], item.get("name"), item["size"])

Interactive Editing

Provide a reference image with hand-drawn sketch/marked regions together with an editing instruction; the model performs precise edits based on the marked regions (e.g., adding, replacing, or removing local elements).

cURL

curl --location --request POST 'https://neolink.com/api/v1/images/generations' \
     --header 'Content-Type: application/json' \
     --header 'Authorization: Bearer <API Key>' \
     --data-raw '{
         "model": "doubao-seedream-5-0-pro-260628",
         "prompt": "Edit the image based on the hand-drawn sketch. Add a stack of realistic magazines or art books in the marked area at the bottom left, and add a ceramic cup of coffee with saucer in the marked area on the right. Remove all sketch lines. Keep the composition unchanged. Let the newly added objects blend naturally into the existing scene.",
         "image": "https://example.com/sketch.png",
         "size": "2K",
         "output_format": "png",
         "response_format": "url",
         "watermark": false
     }'

Python

import requests

url = "https://neolink.com/api/v1/images/generations"

headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer <API Key>",
}

payload = {
    "model": "doubao-seedream-5-0-pro-260628",
    "prompt": "Edit the image based on the hand-drawn sketch. Add a stack of realistic magazines or art books in the marked area at the bottom left, and add a ceramic cup of coffee with saucer in the marked area on the right. Remove all sketch lines. Keep the composition unchanged. Let the newly added objects blend naturally into the existing scene.",
    "image": "https://example.com/sketch.png",
    "size": "2K",
    "output_format": "png",
    "response_format": "url",
    "watermark": False,
}

resp = requests.post(url, headers=headers, json=payload, timeout=60)
resp.raise_for_status()
result = resp.json()
print(result["data"][0]["url"])

Group Generation (Multi-Reference-Image-to-Group)

When sequential_image_generation is auto, group mode is enabled; the model decides based on the prompt and reference images whether to generate a set of related images (input reference images + generated images ≤ 15).

cURL

curl --location --request POST 'https://neolink.com/api/v1/images/generations' \
     --header 'Content-Type: application/json' \
     --header 'Authorization: Bearer <API Key>' \
     --data-raw '{
         "model": "doubao-seedream-5-0-260128",
         "prompt": "Generate 3 images of a girl and a cow plush toy happily riding a roller coaster at the amusement park, covering morning, noon, and night",
         "image": [
             "https://example.com/reference-girl.png",
             "https://example.com/reference-cow.png"
         ],
         "sequential_image_generation": "auto",
         "sequential_image_generation_options": {
             "max_images": 3
         },
         "size": "2K",
         "output_format": "png",
         "watermark": false
     }'

Python

import requests

url = "https://neolink.com/api/v1/images/generations"

headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer <API Key>",
}

payload = {
    "model": "doubao-seedream-5-0-260128",
    "prompt": "Generate 3 images of a girl and a cow plush toy happily riding a roller coaster at the amusement park, covering morning, noon, and night",
    "image": [
        "https://example.com/reference-girl.png",
        "https://example.com/reference-cow.png",
    ],
    "sequential_image_generation": "auto",
    "sequential_image_generation_options": {"max_images": 3},
    "size": "2K",
    "output_format": "png",
    "watermark": False,
}

resp = requests.post(url, headers=headers, json=payload, timeout=60)
resp.raise_for_status()
result = resp.json()

# Iterate and print the URL and size of each image
for item in result["data"]:
    print(item["url"], item["size"])

Supported Model IDs

• doubao-seedream-5-0-pro-260628 — Supports layer decomposition (up to 16 layers), interactive editing, and multi-image-to-image (up to 10 reference images)
• doubao-seedream-5-0-260128 — Supports group generation, multi/single-image-to-image (up to 14 reference images), web search, and streaming output

<Model ID> should be replaced with the actual model ID. The actual available models, resolutions, and reference image counts depend on the platform.

Response Body

Parameter	Type	Description
created	integer	Unix timestamp (seconds) of the request creation time
model	string	The model ID used for this request
data	object[]	List of output image information; url, b64_json, size are common fields; other fields vary by scenario
data.url	string	Image download URL, returned when response_format is url; valid for 24 hours
data.b64_json	string	Base64 image data, returned when response_format is b64_json
data.size	string	Image pixel size, format widthxheight, e.g., 2048x2048
data.output_format	string	Output image format, png or jpeg
data.z_index	integer	Layer stacking index; the base image is always 0, layers increment from 1 (layer decomposition scenario)
data.name	string	Name/label of the decomposed layer element (layer decomposition scenario)
data.description	string	Detailed description of the decomposed layer element (layer decomposition scenario)
data.bounding_box	object	Bounding box of the layer in the base-image coordinate system (layer decomposition scenario)
data.bounding_box.absolute	array	Absolute pixel coordinates [left, top, right, bottom]; the base-image top-left corner is (0, 0)
data.bounding_box.normalized	array	Normalized coordinates [left, top, right, bottom] in the [0, 1000] range
data.error	object	Error info (code, message) when a single image fails in group generation; other images are unaffected
error	object	Top-level error info (code, message), returned when no image could be generated
usage	object	Usage information
usage.generated_images	integer	Number of successfully generated images; only successful images are billed
usage.input_images	integer	Number of input images
usage.output_tokens	integer	Tokens spent on generated images, computed as sum(image_width * image_height) / 256, rounded down
usage.total_tokens	integer	Total tokens consumed by this request; input tokens are not counted, so it equals output_tokens
usage.tool_usage	object	Tool usage information
usage.tool_usage.web_search	integer	Number of web searches; 0 means no search was performed

Response Example

The following response examples assume response_format is url, showing typical output structures for text-to-image / image-to-image, multi-image fusion, layer decomposition, interactive editing, and group generation (with b64_json, data.url is replaced by data.b64_json).

Note on layer decomposition: the base image has z_index fixed at 0 and does not return bounding_box; each layer returns z_index, name, description, and bounding_box. You can restore layers to the base image or a custom canvas using bounding_box.absolute / normalized.

Note on group generation: in group mode data returns multiple images; if a single image fails, that entry additionally returns an error field (code, message) without affecting other images.

Text-to-Image / Image-to-Image

{
  "created": 1722769557,
  "model": "<Model ID>",
  "data": [
    {
      "url": "https://.../image_1.png",
      "size": "2048x2048",
      "output_format": "png"
    }
  ],
  "usage": {
    "generated_images": 1,
    "output_tokens": 16384,
    "total_tokens": 16384
  }
}

Multi-Image Fusion

{
  "model": "doubao-seedream-5-0-pro-260628",
  "created": 1757323224,
  "data": [
    {
      "url": "https://.../fused.png",
      "size": "1760x2368"
    }
  ],
  "usage": {
    "generated_images": 1,
    "output_tokens": 16280,
    "total_tokens": 16280
  }
}

Layer Decomposition

{
  "model": "doubao-seedream-5-0-pro-260628",
  "created": 1784696685,
  "data": [
    {
      "url": "https://.../base.png",
      "size": "2048x2048",
      "output_format": "jpeg",
      "z_index": 0
    },
    {
      "url": "https://.../layer_title.png",
      "size": "1273x265",
      "output_format": "png",
      "z_index": 1,
      "bounding_box": {
        "absolute": [383, 120, 1655, 384],
        "normalized": [187, 59, 808, 188]
      },
      "name": "Seedream heading text",
      "description": "Yellow large serif-font Seedream heading text"
    }
  ],
  "usage": {
    "input_images": 1,
    "generated_images": 8,
    "output_tokens": 23107,
    "total_tokens": 23107
  }
}

Interactive Editing

{
  "model": "doubao-seedream-5-0-pro-260628",
  "created": 1757323224,
  "data": [
    {
      "url": "https://.../edited.png",
      "size": "2048x2048"
    }
  ],
  "usage": {
    "generated_images": 1,
    "output_tokens": 18000,
    "total_tokens": 18000
  }
}

Group Generation

{
  "model": "doubao-seedream-5-0-260128",
  "created": 1757388756,
  "data": [
    {
      "url": "https://.../image_1.png",
      "size": "2720x1536"
    },
    {
      "url": "https://.../image_2.png",
      "size": "2720x1536"
    },
    {
      "url": "https://.../image_3.png",
      "size": "2720x1536"
    }
  ],
  "usage": {
    "generated_images": 3,
    "output_tokens": 48960,
    "total_tokens": 48960
  }
}

Notes:

• Image URLs returned by url are valid for 24 hours after generation; download them in time.
• If response_format is b64_json, the generated image is Base64-encoded and must be decoded before saving.