Image Generation

Agents can generate images using DALL-E 3. The image generation tool accepts a natural language prompt and returns a URL or base64-encoded image.

Using image generation

Agents invoke the generate_image tool directly in their tool loop. The tool accepts a prompt and optional size, quality, and style parameters and returns either a temporary URL or a base64 payload depending on configuration.

json

{
  "tool": "generate_image",
  "params": {
    "prompt": "A minimalist architectural diagram of a microservices system with pastel colors",
    "size": "1792x1024",
    "quality": "hd",
    "style": "natural"
  }
}

Supported parameters

Parameter	Type	Required	Options / Default	Description
`prompt`	string	Yes	—	Natural language description of the image to generate (max 4000 characters)
`size`	string	No	`1024x1024`	Output dimensions: `1024x1024`, `1792x1024`, or `1024x1792`
`quality`	string	No	`standard`	Render quality: `standard` or `hd`
`style`	string	No	`vivid`	Visual style: `vivid` (hyper-real) or `natural` (subdued, realistic)
`n`	number	No	`1`	Number of images to generate (DALL-E 3 supports `1` only)
`persist`	boolean	No	`false`	Return base64-encoded image data instead of a temporary URL

Configuration

Image generation requires an OpenAI API key with DALL-E 3 access.

yaml

tools:
  imageGeneration:
    enabled: true

bash

export OPENAI_API_KEY=sk-...

Returned output

By default the tool returns a temporary OpenAI-hosted URL valid for one hour. Set persist: true to receive a base64-encoded image instead, which the gateway stores and serves from its own CDN.

Mode	Response field	Expiry
Default (`persist: false`)	`url` — OpenAI CDN link	1 hour
Persisted (`persist: true`)	`url` — Gateway CDN link, `base64` — raw data	No expiry

⚠HD images cost significantly more than standard quality. Use quality: "standard" for drafts and iteration, reserving quality: "hd" for final outputs where detail matters.