AI product photography prompts: a template and 8 examples

A good AI product photography prompt names five things in one or two plain sentences: the shot type, the surface or backdrop, the light, the camera angle, and, when you send a photo of your real product, what must not change. For example: “Studio packshot of the referenced candle on warm gray paper, soft light from the upper left, straight-on at eye level. Keep the jar, label, and lid identical.”

Below are a template, eight prompts by shot type, and the settings that belong in request fields. The field facts come from Sume's Image API docs, read on 2026-09-28; the prompts are illustrations, not wording any model is documented to follow.

What should a product photo prompt include?

Fill in this template and drop the parts you don't need:

• Shot type: packshot, flat lay, macro detail, hero shot, in-hand, or scene.
• Backdrop: a named color and material, such as “matte sage-green paper” or “rough slate”.
• Light: where it comes from and how hard it is, such as “one hard light from the right, long shadow”.
• Camera: “top-down”, “45 degrees above”, “eye level”, “macro close-up”.
• The keep line: which parts of your product must match the photo you send.

[Shot type] of the referenced [product] on/in [surface or backdrop],
[light: direction, softness, color], [camera: angle, distance, lens feel],
[props, if any]. Keep the [shape, label, logo, color] identical.

What are some AI product photography prompts?

Eight prompts, one per common shot. Each assumes you send a photo of the product as a reference:

• Color backdrop: “The referenced sneaker on a matte sage-green paper backdrop, soft diffused light from both sides, 3/4 view at eye level. Keep the shoe, stitching, and logo identical.”
• Flat lay: “Top-down flat lay of the referenced notebook on light oak, with a pen and a pair of glasses at the edges, even daylight, no harsh shadows. Keep the cover design identical.”
• Macro detail: “Macro close-up of the referenced watch dial, shallow depth of field, a thin highlight along the steel bezel, black background. Keep the dial markings identical.”
• Hard shadow: “The referenced perfume bottle on a white plinth, one hard sunlight from the right casting a long palm-leaf shadow across the wall. Keep the bottle and label identical.”
• Reflective hero: “The referenced headphones standing upright on a glossy black surface with a soft reflection below, rim light outlining the edges, low angle. Keep the colors identical.”
• In-hand scale: “A hand holding the referenced lip balm against a pale pink backdrop, soft front light, close crop at the fingers. Keep the tube and logo identical.”
• Texture set: “The referenced ceramic mug on rough dark slate beside coffee beans, warm side light, 45 degrees above. Keep the glaze color and handle identical.”
• Outdoor props: “The referenced water bottle on a mossy rock by a stream, dappled forest light, eye level. Keep the bottle, cap, and print identical.”

Should I send my product photo or only describe it?

Send the photo. From text alone, the model has only your words to go on, so the product it draws is invented. On POST /v1/images, put the photo in input_references, refer to it in the prompt as “the referenced” product, and set aspect_ratio: "auto": on an edit, the docs advise auto to match the reference, and omitting the field is not the same.

The reference rules (a public HTTPS URL, and a model that accepts references) are covered in AI lifestyle product photography.

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",
    "prompt": "Top-down flat lay of the referenced notebook on light oak, with a pen at the edge, even daylight. Keep the cover design identical.",
    "aspect_ratio": "auto",
    "n": 2,
    "output_format": "png",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/notebook.jpg" } }
    ]
  }'

Which settings go in fields instead of the prompt?

Frame shape, count, and file type are request fields, not prompt words. A model only accepts the values its catalog descriptors list, so read supported_parameters from GET /v1/images/models first. The full field list is in AI image prompt examples; these are the ones product shots need:

What are the limits?

• The product can still drift. No field locks its appearance, so compare every result with your photo before you publish it.
• Seedream 4.5 returns one image per call when you send a reference, whatever n says (current code). Send one call per take.
• Result URLs are Sume-hosted and signed. Download the images you keep.
• For a cutout with a transparent background, generate on a plain backdrop and remove the background afterwards, as AI image generator with a transparent background shows.
• Lifestyle scenes built from one packshot are covered in AI lifestyle product photography.