
16:9, 2x2 grid, do this for an unusual but real classic motorcycles INPUT = [ TOPIC / HERO_OBJECT / BRAND_TEMPERAMENT / VIEWPORT / GRAVITY_LAW ] Run HERO_PRODUCT_GRAVITY: TOPIC_DNA := infer the category from TOPIC alone — what's sold, who buys it, what its shelf looks like HERO_DNA := ONE photographic hero object, cut out, oversized, 30–40% of frame; real material response — specular on foil, matte on paper, subsurface scatter on skin or fruit BRAND_DNA := derive the palette FROM the object — 1 dominant, 1 accent pulled from a hue the object actually contains, 1 near-neutral ground LIGHT_ANCHOR := decide key light position ONCE (upper-left or upper-right); every shadow — object, cards, pills, floating confetti — falls the same direction at the same angle and softness GRAVITY_LAW := the hero is either GROUNDED (contact shadow touching a visible surface plane) or SUSPENDED (ambient occlusion beneath, blur growing with height) — never both, never floating with a hard cast shadow and no plane to cast it onto DEPTH_ORDER := ground wash < decorative rays/confetti < hero object < UI chrome. Nav, cart and buttons are ALWAYS in front, never occluded TYPE_LOGIC := one display face for the headline (heavy — pick condensed OR wide, not both), one neutral for everything else; headline ≤ 6 words, body renders as legible micro-type, never enlarged gibberish IMPERFECTION := exactly one element breaks its bounding box — a badge rotated 4–7°, a sticker overlapping the product silhouette, a stray mark in the gutter FRAME := desktop 16:10, 64–80px outer gutter; the bottom edge reveals the top 15% of the next section so the page reads as scrollable MOOD := appetite, immediacy, one unmistakable action Render: one screen where the product's own light governs every pixel of interface around it. Forbidden — objects: stock-photo humans, lorem blocks at display size, a second competing hero Forbidden — look: conflicting shadow directions, cut-out halo or chewed edges, cast shadows on flat 2D illustration, palette colors absent from the object, three type families, nav bar sitting behind the product
This image was generated with GPT Image 2.5 by OpenAI. The prompt on this page is the exact text used to create it.
The example was generated with GPT Image 2.5 using: 1672:941 · Poster · Product Photography · Advertising. Reusing these settings gives the closest match.
Copy the prompt as-is first, then replace only the subject while keeping the style, lighting, and composition keywords. Generate a few variations — outputs differ between runs.
Yes. Prompts work best as starting points — adjust descriptors, aspect ratio, or style tags to make the output your own.