Image Generation (Sync)

Text to Image

This page covers the Doubao plugin's synchronous OpenAI Images API. For submit-and-poll usage see Image Generation (Async) and Query Result in this section; the separate async adaptor supports different models and parameters. The completed images are returned in data; no asynchronous task polling is required.

Custom dimensions are widthxheight (also X, ×, *), with positive integer sides and aspect ratio from 1:16 to 16:1. Presets are case-insensitive. Omitted size is estimated as 2K but not inserted into the upstream request; the actual output depends on the provider. auto is only valid for layer decomposition on the edit endpoint.

JSON parameters

layer_decomposition: true requires exactly one reference image and cannot be used here. This OpenAI Images endpoint forwards only its known fields; do not assume unknown fields reach upstream.

curl -X POST "https://www.omwai.xyz/v1/images/generations" \  -H "Content-Type: application/json" \  -d '{    "model": "doubao-seedream-5-0-pro-260628",    "prompt": "一只橘猫坐在雨后的霓虹街道旁，电影感摄影",    "size": "2K",    "response_format": "url",    "watermark": false  }'

{
  "created": 0,
  "data": [
    {
      "url": "http://example.com",
      "b64_json": "string"
    }
  ]
}

Image to Image

This page covers the Doubao plugin's synchronous image-to-image endpoint POST /v1/images/edits. For async reference-image generation, use Image Generation (Async) in this section; its adaptor has different model and parameter support. Reference images may be supplied through the JSON image field as public HTTP(S) URLs or Base64 data URIs with a lowercase MIME format. Multipart requests may upload image, image[], or image[n] file fields.

Parameters and available options

The six registered models, their presets, custom pixel limits, output formats, tools and prompt optimization support are listed in the Text to Image section above.

Multipart file fields may be image, image[], or image[n], up to 30 MiB per file. Other form fields may appear only once; n, seed, and guidance_scale are parsed as numbers; stream, watermark, optimize_prompt, and layer_decomposition use true / false. Send optimize_prompt_options, sequential_image_generation_options, and tools as JSON strings. Repeat image file fields for multiple files, not ordinary fields. Results are synchronous; there is no task-query step.

curl -X POST "https://www.omwai.xyz/v1/images/edits" \  -H "Content-Type: application/json" \  -d '{    "model": "doubao-seedream-5-0-pro-260628",    "prompt": "保留主体，将背景替换为日落时的海边",    "image": "https://example.com/reference.png",    "size": "2K",    "response_format": "url"  }'

{
  "created": 0,
  "data": [
    {
      "url": "http://example.com",
      "b64_json": "string"
    }
  ]
}