Doubao Seedream Image Generation API Guide

Fhddos exposes Doubao’s Seedream and SeedEdit image models through a single OpenAI-compatible endpoint: POST /v1/images/generations. You can generate images from text prompts, edit images using a reference, fuse multiple input images, and even stream results as they’re generated — all with your Fhddos API key and no VolcArk console setup.

export BASE_URL="https://aiapi.fhddos.com/v1"
export TOKEN="oh-xxxxxxxxxxxxxxxx"

Authorization: Bearer <TOKEN>
Content-Type: application/json

Available Models

Model Name	Capabilities
`doubao-seedream-4.5`	Text-to-image, image-to-image, multi-image fusion, sequential generation, streaming (SSE) — same feature set as 4.0 with improvements
`doubao-seedream-4.5-n`	Same as above — per-image billing variant, ideal when using `n` for batch generation
`doubao-seedream-4.0`	Text-to-image, image-to-image, multi-image fusion (2–10 inputs), sequential generation (up to 15 images), streaming (SSE)
`doubao-seedream-4.0-n`	Same as above — per-image billing variant, ideal when using `n` for batch generation
`doubao-seedream-3.0-t2i`	Text-to-image with `seed` and `guidance_scale` control for reproducible outputs
`doubao-seededit-3.0-i2i`	Dedicated image-to-image editing

Use doubao-seedream-4.5 or doubao-seedream-4.0 as your default. Switch to doubao-seedream-3.0-t2i when you need deterministic outputs via seed, or to doubao-seededit-3.0-i2i for precise image editing tasks.

Endpoint

POST /v1/images/generations

Request Parameters

Standard OpenAI Fields

Parameter	Type	Required	Description
`model`	string	✅	Model name, e.g. `doubao-seedream-4.0`
`prompt`	string	✅	Image description. Keep under 300 Chinese characters or 600 English words.
`size`	string	❌	Output dimensions, e.g. `1024x1024`, `2K`, `4K`
`n`	integer	❌	Number of independent images to generate (default: 1)
`response_format`	string	❌	`url` (default) or `b64_json`
`quality`	string	❌	`high` or `standard`
`style`	string	❌	`vivid` or `natural`

VolcArk Extension Fields

Parameter	Type	Supported Models	Description
`image`	string or string[]	Seedream 4.0, SeedEdit 3.0	Reference image(s) — URL or base64. Pass an array for multi-image fusion (2–10 images).
`seed`	integer	Seedream 3.0, SeedEdit 3.0	Random seed in `[-1, 2147483647]`. Use `-1` for random.
`sequential_image_generation`	string	Seedream 4.0	Enable sequential composition: `auto` or `disabled`
`sequential_image_generation_options.max_images`	integer	Seedream 4.0	Max images in a sequential set (range: 1–15)
`stream`	boolean	Seedream 4.0	Enable SSE streaming for progressive output
`guidance_scale`	float	Seedream 3.0, SeedEdit 3.0	Text adherence strength (range: 1–10)
`watermark`	boolean	All	Add a watermark to the output (default: `true`)
`optimize_prompt_options.mode`	string	Seedream 4.0	Prompt optimization mode: `standard` (default) or `fast`

All unrecognized fields are forwarded as-is to the VolcArk upstream. Billing is calculated from resolution and image count — your client doesn’t need to provide extra metadata.

Examples

Text-to-Image

cURL
Python (OpenAI SDK)

curl -X POST "$BASE_URL/images/generations" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedream-4.0",
    "prompt": "A cute panda eating bamboo in a sunlit forest, dappled light through the leaves",
    "size": "2048x2048",
    "response_format": "url"
  }'

from openai import OpenAI

client = OpenAI(
    base_url="https://aiapi.fhddos.com/v1",
    api_key="oh-xxxxxxxx"
)

response = client.images.generate(
    model="doubao-seedream-4.0",
    prompt="A cute panda eating bamboo in a sunlit forest, dappled light through the leaves",
    size="2048x2048",
    response_format="url"
)

print(response.data[0].url)

Image-to-Image Editing (Single Reference)

Pass a image URL with doubao-seedream-4.0 to apply style or content transformations:

curl -X POST "$BASE_URL/images/generations" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedream-4.0",
    "prompt": "Convert this photo to a watercolor painting style",
    "image": "https://ark-project.tos-cn-beijing.volces.com/doc_image/seelite_first_frame.png",
    "size": "2048x2048"
  }'

For precise editing with adaptive sizing, use the dedicated SeedEdit model:

curl -X POST "$BASE_URL/images/generations" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seededit-3.0-i2i",
    "prompt": "Change the background to a snowy mountain scene",
    "image": "https://example.com/original.jpg",
    "size": "adaptive",
    "guidance_scale": 7.5
  }'

Multi-Image Fusion (2–10 Inputs)

Pass an array of image URLs to blend styles or elements from multiple sources:

curl -X POST "$BASE_URL/images/generations" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedream-4.0",
    "prompt": "Blend the visual styles of these images into a single composition",
    "image": [
      "https://example.com/style-ref-1.jpg",
      "https://example.com/style-ref-2.jpg",
      "https://example.com/style-ref-3.jpg"
    ],
    "size": "2048x2048"
  }'

Sequential / Composition Generation

Generate a narrative-linked set of images in one request:

curl -X POST "$BASE_URL/images/generations" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedream-4.0",
    "prompt": "Four seasons landscape: spring blossom, summer green, autumn leaves, winter snow",
    "size": "1024x1024",
    "sequential_image_generation": "auto",
    "sequential_image_generation_options": {
      "max_images": 4
    }
  }'

Reproducible Output with Seed Control

Use doubao-seedream-3.0-t2i with a fixed seed to reproduce the same image:

curl -X POST "$BASE_URL/images/generations" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedream-3.0-t2i",
    "prompt": "Van Gogh style starry night over a mountain village",
    "size": "1024x1024",
    "seed": 12345,
    "guidance_scale": 7.5
  }'

Streaming Sequential Generation (SSE)

Use stream: true with doubao-seedream-4.0 to receive images progressively:

curl -N -X POST "$BASE_URL/images/generations" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -H "Accept: text/event-stream" \
  -d '{
    "model": "doubao-seedream-4.0",
    "prompt": "A city skyline at sunset from multiple angles",
    "size": "1024x1024",
    "stream": true,
    "sequential_image_generation": "auto",
    "sequential_image_generation_options": {
      "max_images": 4
    }
  }'

Streaming events follow this pattern:

Event Type	Meaning
`image_generation.partial_succeeded`	One image in the set is ready
`image_generation.partial_failed`	One image failed (others continue)
`image_generation.completed`	All images processed, includes `usage`

Response Format

Standard (Non-Streaming)

{
  "created": 1589478378,
  "data": [
    {
      "url": "https://...",
      "size": "2048x2048"
    }
  ],
  "usage": {
    "output_tokens": 16384,
    "total_tokens": 16384
  }
}

Base64 Response

Set "response_format": "b64_json" to receive the image as a base64 string instead of a URL:

{
  "created": 1589478378,
  "data": [
    {
      "b64_json": "iVBORw0KGgoAAAANSUhEUgAA..."
    }
  ]
}

Frequently Asked Questions

What is the difference between sequential generation and the n parameter?

Sequential generation (sequential_image_generation: auto) produces a set of content-related images — the model decides how many to generate (up to max_images) based on the prompt’s narrative. The n parameter generates n independent images from the same prompt, similar to a batch sample.

How do I set the size for each model?

Seedream 4.5 / 4.0: Use resolution modes (1K, 2K, 4K) or exact dimensions like 2048x2048. Total pixel count must be between 921,600 and 16,777,216.
Seedream 3.0-t2i: Choose between 512x512 and 2048x2048.
SeedEdit 3.0-i2i: Use adaptive to match the input image dimensions automatically.

How long are generated image URLs valid?

Generated image URLs are typically valid for 24 hours. Download and store images to your own storage before the expiry window.

How do I remove the watermark?

Add "watermark": false to your request body. This requires the model and your account configuration to allow it.

Why can't my reference image be uploaded?

Ensure image contains a publicly accessible HTTPS URL or a correctly encoded base64 string. For large images, compress them first to avoid exceeding upstream size limits.

Billing

Billing is based on the number of successfully generated images. Failed images are generally not charged. Check Model Pricing in the Fhddos console for the specific per-image rates for each Seedream model.

​Available Models

​Endpoint

​Request Parameters

​Standard OpenAI Fields

​VolcArk Extension Fields

​Examples

​Text-to-Image

​Image-to-Image Editing (Single Reference)

​Multi-Image Fusion (2–10 Inputs)

​Sequential / Composition Generation

​Reproducible Output with Seed Control

​Streaming Sequential Generation (SSE)

​Response Format

​Standard (Non-Streaming)

​Base64 Response

​Frequently Asked Questions

​Billing

Available Models

Endpoint

Request Parameters

Standard OpenAI Fields

VolcArk Extension Fields

Examples

Text-to-Image

Image-to-Image Editing (Single Reference)

Multi-Image Fusion (2–10 Inputs)

Sequential / Composition Generation

Reproducible Output with Seed Control

Streaming Sequential Generation (SSE)

Response Format

Standard (Non-Streaming)

Base64 Response

Frequently Asked Questions

Billing