Over the past few months I've been building an AI image workspace, and part of that work meant reviewing a lot of product photography prompts — over 1,500 of them, across GPT Image, Nano Banana (Gemini), and Midjourney. Most were mediocre. A few patterns consistently produced commercial-grade results.
I open-sourced the best templates and 122 image+prompt pairs here: awesome-ai-product-photography-prompts (CC0, free to use commercially).
This post is the distilled version: the five patterns that separate usable product shots from AI-looking junk.
1. Product truth first, style words last
The single biggest quality difference. Weak prompts front-load atmosphere ("cinematic, 8k, award-winning...") and mention the product as an afterthought. Strong prompts pin the product's physical facts before anything else:
Ultra-realistic minimalist product photography of a small transparent
glass facial oil dropper bottle with a black rubber pipette top...
Material, container type, closure, color — stated explicitly. The model can't hallucinate what you've pinned down.
2. Fictional branding beats no branding
Counterintuitive: prompts that invent a brand name ("NOIR BREW coffee", "VOLT RUSH energy drink") produce dramatically more convincing packaging than prompts that avoid labels entirely. The model composes a full label design around the name — typography, hierarchy, the works.
Two rules:
- Invent the name. Never use a real brand (trademark trouble, and the model will just clone the real trade dress).
- Quote it in the prompt:
labeled "HUNSET"— quotation marks strongly improve text rendering accuracy.
3. One image, one job
The prompts that fail hardest try to do everything: hero shot + lifestyle context + ingredients + text overlay space. Every strong prompt in my collection has exactly one job:
- Packshot: product centered, seamless background, even light
- Detail: macro on one material or feature
- Lifestyle: one believable moment, product stays the focal point
- Campaign: dramatic single concept with negative space for copy
Same taxonomy works for image-to-video: one main motion, one supporting motion, explicit camera behavior.
4. Describe light like a photographer, not an adjective machine
"Beautiful lighting" does nothing. What works is naming the setup:
...warm sunlight filtering through wheat stalks from behind, creating
glowing translucency inside the honey and soft halo edges around the jar,
shallow depth of field, shot on 90mm macro lens, f/2.2
Direction, quality, and what the light does to the product. Lens and aperture references reliably shift the output toward commercial photography conventions.
5. Never generate your claims
The one rule with real-world consequences: AI lettering is a mood, not a source of truth. Prices, product names, dates, legal text — add them in your design tool after generation. Every "AI product photo fail" screenshot you've seen is someone violating this.
Browse the full collection
All 122 examples with their complete prompts are in the GitHub repo — organized into product photography, food & drink, and poster design. Each image links to a live page where you can run the prompt against your own product.
The underlying gallery (1,500+ prompts with generated results) is free to browse at Pixonara — no signup needed to view prompts.
If you've found prompt structures that consistently work for product content, I'd genuinely like to see them — PRs to the repo are welcome.
Top comments (0)