Enhance Image Generation with Presets

Goal: a preset that wraps a text model around the image generation helper. The text model rewrites a vague request into a detailed picture prompt, then calls the helper. One call, a better picture.

Outcome: a preset any app calls like a model — @preset/image-enhancer. A user sends "a couscous poster for Yennayer"; the text model writes the scene — subject, composition, light, colours, style — before the picture model draws it.

Each request through this preset runs two models: the text model, on its tokens, and the picture model, on its own meter. Both are billed in DA, and the request's usage.cost carries the whole charge.

How the pattern works#

1. The user sends a short request to @preset/image-enhancer.
2. The text model, guided by the preset's system prompt, expands it into a detailed picture prompt.
3. The text model calls opendunes:image_generation with that prompt.
4. The picture is drawn, and its address goes back to the text model.
5. The text model answers with the picture, and a word on what it made.

Step 1: Save the preset#

Send the body you would send to /v1/chat/completions, to the preset route. The system prompt, the tools and the parameters are kept; the rest of the conversation is not.

curl https://opendunes.com/v1/presets/image-enhancer/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-5",
    "messages": [
      {
        "role": "system",
        "content": "You are an expert visual prompt writer. When the user asks for a picture, expand the request into a detailed prompt — subject and action, composition, lighting, colour palette, style — then call the image generation tool with it. Reply with the picture and one sentence on what you made."
      }
    ],
    "tools": [
      {
        "type": "opendunes:image_generation",
        "parameters": { "model": "openai/gpt-5-image-mini", "aspect_ratio": "3:4" }
      }
    ],
    "temperature": 0.7
  }'

Saving again under image-enhancer adds a new version and makes it current; every version is kept.

Step 2: Use it#

curl https://opendunes.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "@preset/image-enhancer",
    "messages": [{ "role": "user", "content": "A couscous poster for Yennayer" }]
  }'

Anything the request sets itself wins over the preset — a different temperature, or a tools array of its own.

Next steps#

• Presets reference — what a preset keeps, and its routes.
• Image Generation helper — the picture model's parameters.
• Images API — drawing a picture directly, without a text model.