Skip to main content
Use ai-stats:image_generation when the model should be able to create an image while it is answering. AI Stats exposes a managed tool to the model, executes the image request server-side, and feeds the image result back into the model loop.

​
Request

curl https://api.phaseo.app/v1/responses \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5-nano",
    "input": "Design a wide product concept image for a compact desktop synthesizer.",
    "tools": [
      {
        "type": "ai-stats:image_generation",
        "parameters": {
          "model": "openai/gpt-image-2",
          "quality": "high",
          "aspect_ratio": "16:9",
          "output_format": "png"
        }
      }
    ]
  }'

​
Parameters

ParameterTypeDescription
modelstringOptional image-capable model ID. Defaults to openai/gpt-image-2 when neither the tool nor tool call provides a model.
qualitystringOptional provider quality hint, such as low, medium, or high.
sizestringOptional image size hint.
aspect_ratiostringOptional aspect ratio, such as 1:1, 16:9, or 9:16.
backgroundstringOptional background mode hint.
output_formatstringOptional output format, such as png, jpeg, or webp.
output_compressionintegerOptional provider compression hint.
moderationstringOptional provider moderation hint.
The model supplies the final prompt when it calls the tool. Tool-level parameters provide defaults and constraints for that call. The tool call may also send description instead of prompt. AI Stats treats description as a prompt alias for compatibility with model-generated image descriptions.

​
Tool Result

The result is returned to the model as JSON: 
{
  "status": "ok",
  "model": "openai/gpt-image-2",
  "imageUrl": "https://...",
  "mime_type": "image/png"
}
Some providers return base64 image data instead: 
{
  "status": "ok",
  "model": "openai/gpt-image-2",
  "b64_json": "...",
  "mime_type": "image/png"
}

​
Usage And Pricing

Image generation usage is included in the parent response usage when the image model reports token usage. AI Stats also records: 
{
  "usage": {
    "server_tool_use": {
      "image_generation_requests": 1
    }
  }
}
Pricing cards can bill this with the server_tool_image_generation_requests meter. The underlying image model usage can still be billed through its normal model pricing.
Last modified on June 11, 2026