AI sneaker product photo on a colored backdrop, from one packshot

To put a sneaker on a coloured studio backdrop, send your real packshot as a reference and tell the model to keep the shoe unchanged while it replaces the background. On Sume that is one edit call with openai/gpt-image-2.5 or another reference model. The shoe is the product, so the prompt must protect its stitching, logo, sole and laces, and you must check each output against the original.

Prompt structure

Order the prompt as scene, subject, details, constraints. Open with the backdrop and light, then say the shoe in Image 1 stays exactly as photographed, then list what must not change.

A preserve list works better than a vague "keep it the same": name the logo, sole pattern, lace colour and stitching, and say not to add text or a second shoe.

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": "Place the sneaker from Image 1 on a seamless coral studio backdrop, soft key light from the left, gentle shadow on the floor. Keep the shoe exactly as photographed: logo, sole pattern, laces, stitching. No added text, one shoe only.",
  "input_references": [
    {"type":"image_url","image_url":{"url":"https://example.com/sneaker-packshot.jpg"}}
  ],
  "aspect_ratio": "4:5",
  "quality": "high"
}'

Sizes and checks

Ratio depends on where the still goes. Instagram portrait is 4:5 and is available on the GPT family; a square marketplace image is 1:1 (Image API). A GPT custom size needs both edges as multiples of 16, so 1080x1350 is not a legal box; ask for 4:5 and resize afterwards.

Check the logo shape, sole pattern and laces against the packshot at full zoom. If one detail drifted, mask only that region and redo it (masked edits).