Photo to watercolor with ChatGPT Image 2: one reference, one prompt

The Sume Image API docs show an image-to-image example on openai/gpt-image-2: one input_references entry with a public HTTPS photo URL and the prompt "make this scene look like a watercolor painting". That is the whole request. The output keeps the scene and changes its look, and you decide by eye whether it kept the details you care about.

The request

The body is the one in the Image API docs, with a placeholder URL.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2",
    "prompt": "make this scene look like a watercolor painting",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
    ],
    "aspect_ratio": "auto"
  }'

Fields worth setting

aspect_ratio auto matches the reference; leaving it out is not the same. quality can be low, medium or high on this model; xhigh is for the 2.5 rows. The URL must be public HTTPS.

What to check

Style transfer rewrites many pixels. Faces, text on signs and fine patterns often change, so compare against the original before you publish. If you need a region left alone, use a 2.5 row with a mask, remembering that a mask is guidance.