AI prompts for better product photos
How to describe lighting, angle and background so generated product imagery looks deliberate.

Generated product imagery goes wrong in a predictable way: it looks like a photo of something similar to your product, taken by nobody, in a place that doesn't exist. The fix is to prompt like an art director rather than a shopper.
Name the four variables
- Subject: what it is, its material, its actual colour.
- Lighting: soft window light, hard studio flash, warm golden hour.
- Angle: straight-on, three-quarter, top-down flat lay.
- Background: seamless studio backdrop, linen surface, concrete, cream paper.
Miss any one of these and the model invents it, usually badly. State all four in one sentence and quality jumps immediately.
Borrow the vocabulary of real photography
"Shot at 85mm, shallow depth of field, soft shadow to the left" produces a far more coherent image than "professional and beautiful". Technical words map to real patterns in the training data; adjectives about quality do not.
Say what you don't want
Generated imagery loves clutter, extra hands and invented text on packaging. Add explicit exclusions: no text, no logos, no props, no reflections, one item only. Text especially — if a label matters, add it afterwards in a design tool rather than hoping the model spells your brand correctly.
Work in sets, not singles
A shop looks amateur when every image has a different light and crop. Write one base prompt, then vary only the angle line for each shot. Keep the base prompt in a file so next season's photos match this season's.
Know when to photograph instead
Generation is excellent for backgrounds, lifestyle context, mood and social crops. It is still poor at rendering your exact product with the right stitching, label and proportions. For the hero shot of a physical item, a phone on a windowsill with a white card reflector beats any prompt.
A reusable template
"[Product, material, colour] on [surface], [lighting], [angle], soft natural shadow, muted [palette], no text, no logos, single item, product photography." Fill the brackets, keep the tail, and you'll get a consistent set in three or four attempts.