Blog›Guides

AI Product Photography Prompts: What Actually Works, Model by Model

We mined nearly 3,000 curated product and advertising prompts: which image models actually deliver for product shots, and the four prompt patterns behind the top-rated results.

Product shots are the single biggest commercial use of AI image generation, and Wikiprompt's corpus proves it: nearly 3,000 curated prompts are about products, ads and packshots. Instead of another single-model guide, this one mines the whole corpus: which models people use for product work, which ones deliver, and the prompt recipes that survive contact with a real brief.

The data: who wins at product shots

Average "wow" rating for product and advertising prompts, per model (minimum 8 rated examples):

| Model | Prompts | Avg wow |

|---|---|---|

| GPT Image 1 | 130 | 4.03 |

| GPT Image (family) | 587 | 4.01 |

| GPT Image 1.5 | 19 | 3.79 |

| Nano Banana | 354 | 3.75 |

| GPT Image 2 | 765 | 3.71 |

| Grok Imagine | 10 | 3.60 |

| Z-Image Turbo | 20 | 3.60 |

| Nano Banana Pro | 140 | 3.58 |

Two takeaways. The GPT Image family dominates both volume and quality for product work: its typography and layout control matter more in advertising than raw photorealism. And volume flows to the newest models faster than quality does - GPT Image 2 has 6x the prompts of GPT Image 1 but a lower average, because everyone is still learning its dials.

The recipes that keep working

Across the top-rated product prompts, four patterns repeat:

Name the campaign, not the picture. "Vertical advertising poster for [BRAND]" outperforms "beautiful photo of a bottle". The format word (poster, packshot, campaign grid, editorial) sets composition, negative space and text zones in one move.

Physics sells. The highest-wow shots have one exaggerated physical event: an explosion of soda and ice, melting cheese, splashing matcha. Static product on white is table stakes; motion is what earns the scroll-stop.

Write the headline verbatim. Quote the exact text you want rendered. GPT Image models will place it; leaving it vague produces gibberish labels.

Anchor materials and light. "Condensation droplets, studio rim light, glossy reflection on black acrylic" - material words are what separate a render from a photo.

Steal these

Cinematic Soda Product Explosion - the physics pattern at its best; swap the can, keep the chaos.

Ultra-Realistic Lemon Soda Commercial Shot - Nano Banana Pro doing condensation and citrus splash.

Premium Advertising Poster for Any Product - the reusable template of the set: format + placeholder + text zone.

Nike Air Max Muse Campaign Grid - multi-frame campaign consistency in one generation.

Cinematic Luxury Skincare with Text Panel - product + rendered copy side panel, the classic e-commerce hero.

Product Photography from Raw Ingredients - Midjourney's take: build the product out of its ingredients.

Meltdown Burger Poster - food physics + bold typography.

1960s BROCANE Advertising Poster - era-styled advertising, great for brand-nostalgia campaigns.

YUKI Matcha Mousse Cake Poster - Japanese-styled food poster with clean text hierarchy.

Cultural Vending Machine Diorama - the creative outlier: product as world-building.

A starter recipe

[Format: vertical advertising poster / hero packshot / campaign grid] for [PRODUCT],

one physical event: [splash / explosion / melt / pour] frozen mid-motion.

Materials: [condensation / gloss / matte ceramic], studio [rim / soft box] lighting

on [black acrylic / seamless white / textured stone]. Render the headline

"[YOUR COPY]" in [style] at [position]. Photorealistic, high detail, 4K.

Explore more

Browse by model: GPT Image 2 · Nano Banana · GPT Image 1, or search the whole corpus for product photography. Today's image-model rankings live on the image leaderboard.