AI Product Photography Prompts: A Brief and Review Checklist

A useful product-image prompt describes what must stay accurate and what may change. Start with the actual item, then specify its setting, lighting, framing and intended finish. Treat the resulting image as a candidate to inspect, not proof that the model preserved the product.

Correction, September 20, 2026: An earlier version claimed thousands of generations, a candle-company experiment and an 86% reviewer preference. We do not have publishable evidence for those results. It also overstated how prompt words relate to model training and described automatic template capabilities that were not established. This version offers a proposed planning and review method, with no measured performance claim.

Apatero publishes this guide and offers generation workflows. A structured brief can make your decisions explicit; it does not guarantee accurate merchandise, matching angles or a usable catalogue.

Start With Product Facts, Not Style Words

For an item that customers can buy, collect reference photos and a short fact sheet before generating. Record the SKU, dimensions, material, color variant, included accessories and visible label. Mark details that cannot change: a cap shape, a clasp, the number of buttons or the wording on a package.

If you are designing a fictional product concept, label the brief as a concept. An attractive invented bottle is useful in a mood board, but it cannot stand in for a real bottle with a different dispenser or label.

Keep a reference contact sheet beside the generated candidates. A prompt can request a specific detail while the output quietly changes it. Review both the overall silhouette and the small features that determine what the buyer receives.

Use Five Parts to Make the Brief Reviewable

The five parts below are an organizational device. They are not a universal model syntax, and this guide does not establish that their order improves results.

“Photorealistic” is optional descriptive language. There is no evidence here that the word universally harms an image or activates a particular training-data category. Compare formulations with your selected model instead of adopting a banned-word rule.

Camera and lens terms can describe an intended look, but a generator is not operating a physical camera. Writing an aperture or focal length does not ensure calibrated optics, exact depth of field or correct product dimensions. A plain instruction such as “show the whole label in focus” is often easier to evaluate than a string of camera settings.

Ten Example Briefs to Adapt

These are proposed prompts for fictional or authorized subjects, not outputs from a completed benchmark. Replace the product description with verified facts and use supported reference inputs where available.

1. Serum bottle, main view: Show the supplied amber bottle and black dropper cap upright on a plain neutral surface. Keep the entire product visible, preserve the cap and bottle proportions, and leave the label legible. Use soft light from the left with a modest contact shadow.
2. Serum bottle, setting concept: Place the same authorized bottle reference on a pale bathroom shelf. Keep the dispenser and label unchanged. Leave space around it; add no accessories that could imply they are included with the purchase.
3. Stoneware mug, main view: Show a matte charcoal mug at a three-quarter angle, including its complete handle and rim. Use a light neutral background and soft side light. Preserve the reference glaze rather than adding decorative chips.
4. Stoneware mug, lifestyle concept: Place the referenced mug on a wooden table near a closed notebook. Keep the handle shape and surface finish visible. Avoid steam if it hides the product details being reviewed.
5. Watch, detail view: Use the supplied watch reference for a close view of the case, crown and strap connection. Preserve dial markings and hardware. Keep the relevant detail in focus and reject invented text or extra controls.
6. Candle, unlit view: Show the referenced candle upright on a plain surface. Preserve the wick count, container and label. Use enough side light to reveal its texture without changing its apparent color.
7. Candle, setting concept: Place the same candle reference in a simple room scene. Retain the actual vessel and proportions. Treat flame, reflections and nearby objects as additional details to inspect, not evidence of burn performance.
8. Sneakers, pair view: Show both shoes from the supplied references with the complete soles visible. Preserve the colorway, lacing and stitching. Use an uncluttered background and reject invented logos or mismatched left and right shoes.
9. Earrings, close view: Arrange the referenced pair on a neutral surface with both clasps visible. Preserve the relative dimensions and metal color. Inspect symmetry, attachment points and whether the image accidentally adds another piece.
10. Bag, three-quarter view: Show the referenced bag with its front, one side, strap attachments and hardware visible. Preserve the closure and seam positions. Do not add pockets or suggest a capacity that the actual item does not have.

For each brief, write a rejection reason before generating. “Reject if the dispenser changes” is more useful than deciding afterward that a beautiful but inaccurate output is close enough.

Build an Angle Set Without Assuming Consistency

Choose only the views needed for the task: front, three-quarter, side, rear, overhead or a detail crop. A side or rear reference is especially important when the model would otherwise have to invent unseen construction.

Reuse the verified product description, but inspect every view separately. Identical wording does not make different generations a geometrically consistent object. Compare seams, controls, label position and accessories across the set. If the rear view invents features, replace it with an actual photograph or omit it.

Use this proposed log to distinguish attempts from approved assets:

The rows illustrate decisions, not results from our testing. Record actual charges and editing time as you work. Ten attempts may produce fewer than ten approved images; a catalogue estimate must include rejection and manual correction.

Keep a Reusable Brief Outside the Generation Step

Store the five-part brief in a document or spreadsheet with the SKU and reference filenames. Copy the relevant fields into the selected operation. This method does not depend on Apatero having a named-slot template object or an automatic catalogue runner.

If you automate text assembly in your own workflow, save the completed prompt and input list for each attempt. Check what the chosen model actually supports: image references, edits, output dimensions and costs differ by operation. A saved text template documents intent; it does not lock output identity.

Review the Destination Before Publishing

A mood board, an advertisement and a shopping-feed main image have different jobs. Do not assume a decorative lifestyle concept is suitable everywhere.

Google Merchant Center's image-link guidance requires accurate product depiction, restricts promotional overlays and requires AI-origin metadata for generative images. Its best practices also call for the correct variant. Preserve applicable provenance metadata through export. Check its current specification before submitting to that feed. These are Merchant Center requirements, not a claim that every editorial blog illustration follows the same rules.

For your own release review, verify that the actual product is represented, text remains correct, rights to reference material are documented, the file meets the destination's format and dimensions, and required provenance information survives export. A white background is not transparency: inspect the output's alpha channel when a transparent file is required.

Keep the final file, its source references and the approval decision together. If the image cannot represent the item faithfully, use an actual product photograph. The purpose of the brief is to make that decision easier, not to make every generated candidate publishable.