AI lifestyle product photography from one packshot

AI lifestyle product photography starts from one clean packshot: send it as a reference image to an image model that edits from references, describe one scene per request (the room, the surface, the props, the light), and ask for a few versions of each. Keep the images where the product still matches the real thing, and re-run the scenes that work.

Amazon's seller guide describes lifestyle shots as ones that show "your products in action". The Sume facts below come from the Image API docs and the model catalog that GET /v1/images/models serves, read on 2026-09-27.

How do I turn a packshot into a lifestyle scene?

Send the packshot in input_references to POST /v1/images, describe the scene in prompt, set n for several candidates, and pick an aspect_ratio for the place the image will run.

• The packshot must be at a public HTTPS URL. Localhost, private-network, and non-HTTPS URLs are rejected before submission.
• The ratio must be on the model's list. The docs note that 4:5 is Instagram portrait (1080×1350), not 4:3, and aspect_ratio: "auto" matches the output to the reference on models that list it.
• A text-to-image-only model, one whose input_references descriptor is {"min": 0, "max": 0}, rejects the packshot.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/nano-banana-pro",
    "prompt": "The referenced coffee grinder on a kitchen counter by a window, morning light, a mug and a bag of beans beside it. Keep the grinder and its logo identical.",
    "aspect_ratio": "4:5",
    "n": 4,
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/grinder-packshot.png" } }
    ]
  }'

What should each scene prompt include?

One scene per request keeps each prompt specific. Refer to the product as "the referenced" item and spell out the rest:

• Setting: "a bathroom shelf with folded towels", "a desk by a window", "a picnic blanket in a park".
• Surface and props that suit the product and leave it in full view.
• Light and time of day: "soft morning light", "warm lamp light in the evening".
• People, if any: "a hand reaching for the bottle".
• What must not change: "Keep the bottle, label, and cap identical."

How many images can I get per call?

The docs allow n up to 10 per request and say per-model ceilings are lower; today the catalog ceiling on the reference-capable models below is 4. Seedream 4.5 is the exception today: with a reference it returns one image per call, so send one call per candidate. Slow configurations, a large n among them, are the most likely to answer 202 with a job instead of 200 with the images.

Will the product look exactly like mine?

Not guaranteed. The model generates a new picture, so shapes, labels, and logos can drift. Compare each image with the packshot before you publish it, and keep plain product shots too: Amazon's guide advises always having white-background images of a product, and then considering colorful or detailed ambient backgrounds. White background product photos covers that half.

For photos of a model with the product, AI model holding your product covers the catalog Format sume-model-product-portrait. For beauty splash and pour shots, see AI product photography API.

How much does it cost, and what are the limits?

Each completed image is billed at the model's catalog rate, so four candidates cost four images, plus a 5.5% agent fee by default. A failed generation is not billed. Each model's per-image rate is in GET /v1/images/models/{model_id}/endpoints.

• Image API result URLs are Sume-hosted and signed. Download the images you keep.
• Inputs are URLs, not file bytes, and each must be public HTTPS.
• No request field locks the product's appearance; the prompt and your review are the controls.