AI whiteboard photo to a clean diagram: an image edit, step by step

To turn a whiteboard photo into a clean diagram, send the photo to an image model as a reference and ask it to redraw the same boxes, arrows and labels on a plain background. On Sume that is a single POST /v1/images edit call. It works best for simple flows of a few boxes; for a dense diagram, treat the result as a draft and compare every label against the photo.

Prepare the photo

Shoot straight on, in even light, with the board filling the frame. Glare is the most common reason text is lost. Crop out the room. Upload the photo to a public HTTPS URL, because Sume rejects private and non-HTTPS URLs (Image API).

Prompt

Say what is on the board, then what to change. For example: "Image 1 is a whiteboard photo of a three-step flow. Redraw it as a clean flat diagram on a white background. Keep every box, arrow and label. Use sans-serif text. Do not add anything." List the labels in quotes if you can read them; the model reads handwriting less reliably than typed text.

Use aspect_ratio: "auto" so the diagram keeps the board's shape. Omitting the field is not the same as auto on edit calls.

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-2.5",
  "prompt": "Image 1 is a whiteboard photo of a three-step flow. Redraw it as a clean flat diagram on white. Keep every box, arrow and label: 'Upload', 'Review', 'Publish'. Do not add anything.",
  "input_references": [
    {"type":"image_url","image_url":{"url":"https://example.com/whiteboard.jpg"}}
  ],
  "aspect_ratio": "auto",
  "quality": "high"
}'

Check the result

Read the output against the photo, label by label. Missing or reordered arrows are the usual failure. Fix one thing per follow-up call, passing the previous output as the reference and saying what must not change.

If you need an editable file, an image model is the wrong end product. Use it for a clean picture and rebuild in a diagram tool.