generate-image-by-seedream

Doubao Seedream Image Generation

Use the included script to generate or edit images viaByteDanceimage generation models:

Doubao Seedream (Jimeng / Dreamina) series of image generation models
Supports text-to-image, image-to-image, and multi-image reference editing

Default version:4.5 → corresponds to doubao-seedream-4-5-251128. Default size:2K。

Model performance recommendations

Performance ranking:4.5 > 5.0 > 4.0
It is recommended to prioritize the 4.5 version for the best results

Aspect ratio

The aspect ratio must be explicitly specified in the image generation prompt, for example: "Generate a cute puppy, 9:16 vertical ratio"

Text-to-image

uv run {baseDir}/scripts/generate_image.py --prompt "你的图片描述" --filename "可爱小狗.jpg"

Using a YAML configuration file

You can write all image generation parameters into a YAML configuration file, then use it via the --config parameter:

Text-to-image configuration example (`text2image.yaml`)：

prompt: 一只可爱的小狗在草地上，9:16竖屏比例
filename: 可爱小狗.jpg
size: 2K
version: 4.5

Image-to-image configuration example (with a reference image list,`image2image.yaml`）：

prompt: 将图1的服装换为图2的服装
filename: 换装女孩.jpg
images:
  - https://ark-project.tos-cn-beijing.volces.com/doc_image/seedream4_imagesToimage_1.png
  - https://ark-project.tos-cn-beijing.volces.com/doc_image/seedream4_5_imagesToimage_2.png
size: 2K
version: 4.5

Local reference image configuration example (`local_images.yaml`)：

prompt: 多图融合成一张插画
filename: 多图融合插画.jpg
images:
  - /path/to/参考图1.jpg
  - /path/to/参考图2.png
size: 2K
version: 4.5

Usage:

uv run {baseDir}/scripts/generate_image.py --config image2image.yaml

Mix configuration files and command-line parameters (command-line parameters take higher priority):

uv run {baseDir}/scripts/generate_image.py --config prompt.yaml --filename "自定义文件名.jpg"

All parameters supported by the YAML configuration file:

prompt: image description
filename: output file name
images: list of reference images (can be URLs or local file paths)
size: output size (such as "2K", "4K")
version: Seedream version (such as "4.5")
model: full model name (advanced option)
api_key: API key

Image-to-image / reference images (multiple images: URLs or local files)

uv run {baseDir}/scripts/generate_image.py \
  --prompt "将图1的服装换为图2的服装" \
  --filename "换装女孩.jpg" \
  -i "https://ark-project.tos-cn-beijing.volces.com/doc_image/seedream4_imagesToimage_1.png" \
  -i "https://ark-project.tos-cn-beijing.volces.com/doc_image/seedream4_5_imagesToimage_2.png" \
  --size 2K \
  --version 4.5

Local reference images (multiple local files)

uv run {baseDir}/scripts/generate_image.py \
  --prompt "多图融合成一张插画" \
  --filename "多图融合插画.jpg" \
  -i "/path/to/参考图1.jpg" \
  -i "/path/to/参考图2.png" \
  --size 2K \
  --version 4.5

Notes

API key

You should first automatically detect whether there is a ARK_API_KEY environment variable; if not, you need to ask the user for it or have the user apply for it on Volcano Engine

Version options

4.0 → doubao-seedream-4-0-250828
4.5(default) → doubao-seedream-4-5-251128
5.0 → doubao-seedream-5-0-260128
Advanced options: the default mapping can still be overridden via parameters such as --model doubao-seedream-5-0-lite-260128 .

Size options (by version)

4.0: 1K, 2K, 4K
4.5: 2K, 4K
5.0: 2K, 3K

File name suggestions (for the caller's reference)

Do not include implementation details such as "seedream" in the file name.
The file name language should be consistent with the user's Prompt language; for example, for a Chinese Prompt use a short Chinese file name (such as 可爱小狗.jpg）。
The name should be semantic but as short as possible; avoid overly long sentences.

Keyword hints (for search discovery)

"doubao", "豆包", "即梦", "Seedream", "dreamina"
"ByteDance image model", "字节跳动生图模型"
"text-to-image", "image-to-image", "reference image", "image generation", "image generation", "outfit swap", "image editing"