AI product angles from one photo: front, three-quarter and back views

To get extra product angles from one photo, send the packshot as a reference and ask for the same product from a new viewpoint, but expect the model to invent sides it cannot see. If the back carries a label, a barcode or print, give the model a photo of the back as a second reference. On Sume this is a POST /v1/images with two entries in input_references, and GPT Image 2.5 takes up to 16 (Image API).

Request

Describe each reference by role and ask for one view per call. One call, one view, is easier to check than a grid of views in one image.

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 the front of a tin. Image 2 is the back. Show the same tin in a three-quarter view from the left, plain white background, soft shadow. Keep the label art and text exactly as in the images.",
  "input_references": [
    {"type": "image_url", "image_url": {"url": "https://example.com/front.jpg"}},
    {"type": "image_url", "image_url": {"url": "https://example.com/back.jpg"}}
  ],
  "aspect_ratio": "1:1"
}''

Which views are safe

A three-quarter view that shows mostly the front is the safest. A full back view without a back reference is a guess. Top-down views of complex objects are the next riskiest.

Check and fix

Compare label art, proportions and small parts against your photos at full zoom. If a view fails, rerun it with a stronger preserve list. If the marketplace needs strict angles, use real photos for those slots and generated ones for lifestyle (marketplace sets).