AI product photography prompts

How often the product changes, and what the vendors’ docs state

Product alteration is a recurring complaint in public reviews of product-photo tools, which we counted per tool on 28 September 2026:

• ChatGPT: edits change parts of the image that should stay the same (7 of 34 sources: Trustpilot).
• Pixelcut: generative edits change the product, with jewellery misplaced and colours altered (9 of 68 sources: App Store, Trustpilot).
• Photoroom: generative features alter the product, labels distorted among them (6 of 76 sources: Trustpilot, App Store).
• Pebblely: prompts and reference images ignored and the product altered (3 of 16 sources: Trustpilot, Neurodigital).
• Flair: logos and patterns come out wrong (2 of 24 sources: Trustpilot, Shotkit).

Each vendor’s documents state a limit of their own. Ideogram’s text layers page says of generated designs: “The text on your poster, book cover, or social graphic is baked in. Fixing a typo means regenerating.”

OpenAI’s image generation guide says of edit masks: “Masking with GPT Image is entirely prompt-based. The model uses the mask as guidance, but may not follow its exact shape with complete precision.”

Google’s API docs cap references per model: Nano Banana 2 takes “Up to 10 images of objects with high-fidelity to include in the final image”. Reference photos and OpenAI’s fidelity setting apply when you start from your own photo, so start there.

One prompt structure for any tool

Seven slots cover a product shot in any tool: subject, surface, light, lens, angle, finish and a constraint line. The templates below join lens and angle on one line and split the constraint into what may change and what must not.

Google publishes its own version for Nano Banana, a sentence template that opens “A high-resolution, studio-lit product photograph of a [product description] on a [background surface/description]” (Developers Blog, 28 August 2025). The filled-in Nano Banana versions, with per-image costs, are on how to use Nano Banana for product photography.

AI product photography prompts by shot

Each template is ours, in brackets where you fill it. Paste it after attaching the product photo. The last two lines differ by shot: they say what the tool may change and what it must keep.

White-background packshot

Subject: the product in the attached photo
Surface: seamless white sweep, soft contact shadow
Light: large overhead softbox, low contrast
Lens and angle: [85mm], straight on at label height
Finish: catalogue, true-to-life colour
Change only: the background and the shadow
Keep unchanged: label text, logo position, cap and body colour

Attach the front photo and a label close-up.

Lifestyle scene

Subject: the product in the attached photos
Surface: [oak kitchen counter beside a cutting board]
Light: [morning window light from the left]
Lens and angle: [50mm], three-quarter view at eye level
Finish: natural, slight depth of field behind the product
Change only: the setting and props around the product
Keep unchanged: product shape, proportions, label and printed colours

Attach two or three angles: Google’s docs count each as an object reference “with high-fidelity”.

Flat lay

Subject: each item in the attached photos, one of each
Surface: [matte stone slab]
Light: single diffused overhead source, short even shadows
Lens and angle: top-down, all items fully in frame
Finish: editorial, generous spacing between items
Change only: the arrangement and the surface
Keep unchanged: item count, each item's packaging and text

Attach one photo per item and stay inside the model’s object limit (10 on Nano Banana 2, 6 on Pro, per Google’s docs).

On-model or in-hand

Subject: the product in the attached photos, held in [a right hand]
Person: [adult, short nails, no rings], face out of frame
Light: [soft daylight, outdoor shade]
Lens and angle: [35mm], close crop on the hand and product
Finish: phone-camera look, no retouching
Change only: the hand, the background and the light
Keep unchanged: the product's label facing camera and its scale in the hand

Describe a generic person. Creator-style ads and real likenesses have rules of their own, set out on AI UGC prompts.

Seasonal and campaign

Subject: the product in the attached photos
Surface: [frosted wooden table], [two pine sprigs]
Light: [cool key light with a warm rim]
Lens and angle: [70mm], slightly elevated front view
Finish: banner, empty space on the [right] third for copy
Change only: props, colour mood and the empty space
Keep unchanged: the product and every word printed on it

Leave headline space empty and add the copy in an editor, where the words stay exact.

Keeping the label and logo right

Reference limits

Google’s Gemini API docs list “Up to 14 images of objects with high-fidelity” on Nano Banana 2 Lite, 10 on Nano Banana 2 and 6 on Nano Banana Pro. Front, back, side and label close-up make four references, inside every model’s limit.

Input fidelity

OpenAI’s guide says “The input_fidelity parameter controls how strongly a model preserves details from input images during edits and reference-image workflows.” For gpt-image-2 you “omit this parameter”, because “the model processes every image input at high fidelity automatically”. OpenAI positions its GPT Image 2.5 Sunburst model “for workflows where editing precision matters most”.

Text layers

Ideogram says that after “Layerize Text”, “every line of text in your design becomes a separate, selectable layer” and “The visual design stays exactly as generated.” The feature is “available on all plans, including the free tier” and through the API.

Ideogram AI Review 2026 – The AI That Finally Gets Text Right (4.0 & Editable Layers Tested)Tech Storm

None of these documents promises an exact label, so the last step is a manual check. Open the output at 100 per cent and compare the label to the product photo. Paste the real label artwork over any wrong letters in your editor.

An AI product photography workflow

1. Photograph the product plainly. Use a phone in daylight against a plain background, and shoot both sides plus a label close-up.
2. Remove the background. Photoroom and Pixelcut both list background removal.
3. Generate or edit the scene. Attach the cut-out and a shot template in Nano Banana, ChatGPT or Flair.
4. Check the label at 100 per cent. Compare every word and the logo against step 1.
5. Fix the text. Composite the real label artwork in an editor. For text generated in Ideogram, its text layers edit the words.
6. Export at the placement’s ratio.

This $8 AI Tool Is Replacing Product Photographers (Flair AI)Apexa AI

The best AI product photography tools list compares the tools for each step, and product photo to video takes the finished still into motion.

Free prompts, paid rights

A free prompt can still produce an image you cannot run as an ad. Photoroom’s terms grant Free accounts a licence “for your own personal, non commercial purposes” (clause 2.1.b, terms dated 29 July 2026).

For images made after 28 February 2025, Flair’s terms say “[i]mages created under the Free, Pro, and Pro+ plans will show up in Flair Gallery and other Flair users may also use them in personal and commercial settings” (Privacy, 19 July 2024).

Check the plan before the prompt. Every tool’s rights by plan and use sit in the rights matrix.

Questions sellers ask about AI product photo prompts

What is a good prompt for product photography?

Name the surface, light, lens, angle and finish, then list what must stay unchanged. Google’s template covers surface, lighting, angle and key detail.

Can AI keep my product’s label exact?

No vendor we read promises it. Composite the real label for exact print; Ideogram’s text layers edit only Ideogram-generated words.

Where can I see AI product photography examples?

The videos on this page show a phone photo turned into a scene and Ideogram’s text layers. Google’s prompting blog shows its template’s coffee-mug output.