Seedream Image Generation

Seedream is ByteDance's image generation model series, supporting text-to-image, high-resolution output, and fine prompt understanding (including Chinese text rendering), via the OpenAI-compatible images API with synchronous responses.

Endpoint

POST $URL/v1/images/generations

When using the OpenAI SDK, set base_url to $URL/v1.

Supported Models

Model	Description
dola-seedream-5.0-pro	Pro, highest quality, with two resolution bands (≤1.5K / >1.5K)
seedream-5.0-lite	Lite, cost-efficient

Parameters

Param	Type	Required	Description
model	string	yes	One of the models above
prompt	string	yes	Image description; include subject, style, and any text to render
n	integer	no	Number of images, default 1. Each request currently returns 1 image: larger values are accepted without error but only 1 image is returned
size	string	no	Output size, e.g. 1024x1024, 1536x1536, 2048x2048. seedream-5.0-lite requires at least 1920x1920 (≥ 3,686,400 pixels); smaller sizes return 400
seed	integer	no	Random seed for reproducibility
response_format	string	no	url (default, temporary link) / b64_json
quality	string	no	Output quality tier; support depends on the model
style	string	no	Image style; support depends on the model
background	string	no	Background mode (e.g. transparent); supported by some models only
output_format	string	no	Output file format (e.g. jpeg / png); supported by some models only
watermark	boolean	no	Whether to add a watermark

Request Example

curl

curl $URL/v1/images/generations \
  -H "Authorization: Bearer $YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dola-seedream-5.0-pro",
    "prompt": "A futuristic product poster featuring wireless earbuds on a deep-blue gradient background, bold white text at the bottom: FUTURE SOUND",
    "n": 1,
    "size": "2048x2048"
  }'

Python

from openai import OpenAI

client = OpenAI(api_key="YOUR_API_KEY", base_url="$URL/v1")
resp = client.images.generate(
    model="seedream-5.0-lite",
    prompt="A futuristic product poster featuring wireless earbuds on a deep-blue gradient background",
    n=1,
    size="2048x2048",
    seed=42,
)
print(resp.data[0].url)

Response Fields

Field	Type	Description
created	integer	Unix timestamp (seconds)
model	string	Actual model code that processed the request (vendor-normalized, e.g. seedream-5-0-260128)
data	array	Generated results (currently 1 image)
data[].url	string	Temporary image URL (download promptly)
data[].size	string	Actual generated size, e.g. 2048x2048
data[].b64_json	string	Base64 image content (returned when response_format: "b64_json")
usage.generated_images	integer	Number of images actually generated
usage.output_tokens	integer	Output tokens consumed (metering only, not billed)

Response Example

{
  "model": "seedream-5-0-260128",
  "created": 1791869249,
  "data": [
    {
      "url": "https://.../seedream-5-0/xxxx.jpeg",
      "size": "2048x2048"
    }
  ],
  "usage": {
    "generated_images": 1,
    "output_tokens": 16384,
    "total_tokens": 16384
  }
}

Pricing

Billed per output image; dola-seedream-5.0-pro is tiered by resolution:

Model	≤1.5K (≤2.61M px)	>1.5K	Input reference image
dola-seedream-5.0-pro	$0.045/image	$0.09/image	1st free, $0.003/image from the 2nd
seedream-5.0-lite	$0.035/image	$0.035/image	—

Example: dola-seedream-5.0-pro at 2048x2048 (≈4.19M px, falls into the >1.5K band) = $0.09; 1536x1536 (≈2.36M px) = $0.045.

Notes

• Result URLs are temporary — download promptly (or use response_format: "b64_json" for base64).
• Seedream handles Chinese prompts and text rendering well — great for posters, covers, and e-commerce visuals.
• Do not rely on n for multiple images: send concurrent requests instead.
• seedream-5.0-lite has a minimum size (1920x1920); dola-seedream-5.0-pro works from 1024x1024 up.
• Group multipliers apply on top; prices above are in the base currency.