AI product photo retouching: fix defects, keep the product

AI product photo retouching works as a prompt edit on your real photo: send the shot to an image-edit model, list the defects to fix (dust, fingerprints, a glare spot, a creased or crooked label), and say the product, logo, and colors must not change. The model returns a redrawn image, so every retouched file needs a check against the original before it goes on a listing.

Below is the workflow with Sume's Image API and the Image Upscale 1.0 schema in the Sume API reference, served by the API reference docs, all read on 2026-09-29. The general change-one-thing pattern is in AI photo editing prompts; this post is the product-shot cleanup job.

How do I retouch a product photo with AI?

Host the photo at a public HTTPS URL and send it in input_references to POST /v1/images on a model that takes references, such as ChatGPT Image 2.5 (openai/gpt-image-2.5). Send aspect_ratio: "auto" so the result keeps the photo's shape; on edit calls, leaving the field out is not the same. n returns more than one version to pick from.

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": "Retouch this product photo. Remove the dust specks and fingerprints on the glass, soften the bright glare spot on the cap, and smooth the wrinkle in the label. Keep the bottle shape, logo, label text, colors, background, and framing exactly the same.",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/products/serum.jpg" } }
    ],
    "aspect_ratio": "auto",
    "n": 2
  }'

What should a product retouching prompt list?

Two lists: what to fix and what to keep. The second list matters more, because the model redraws the whole frame.

• The defects, by name and place: dust on the lid, a fingerprint on the left side, a scratch on the base, a reflection of the photographer in the glass.
• What fixes them: remove, soften, smooth, straighten. Avoid open words like “enhance” or “make it pop”, which invite changes you didn't ask for.
• What must not change: product shape, logo, label text, colors, background, and framing.
• For a catalog, the same wording on every photo of a product line, so the set stays consistent. Batch edit photos with AI covers running one prompt across many files.

Can AI enhance a small product photo?

Yes, as a separate step after the retouch. Sume Image Upscale 1.0 (POST /v1/image-upscale-1.0/upscale) takes a public HTTPS image_url and an upscale_factor from 1 to 4 (default 2), with output_format png (the default) or jpg. It is listed at $0.20 per image, plus a 5.5% agent fee by default. AI image upscale API covers the call.

What can AI retouching get wrong?

Because the model redraws the image rather than painting over pixels, the details a shopper checks can shift. Sume's terms say generated outputs may be inaccurate or unsuitable for a particular use, and that you are responsible for reviewing outputs before publication or commercial use. Check each file for:

• Logos and fine label text: letters can change or blur.
• Shape and proportions: caps, handles, and edges should match the real product.
• Colors: compare against the product itself, not only the source photo.
• Defects that came back, or new ones the edit added.

How much does AI product photo retouching cost?

Each retouch is an ordinary image edit, billed per completed image at the model's endpoint price; an upscale is a second job at its listed per-image price.