Nike's 5 moves as a checklist, plus the exact prompts. Every one tested before it went in here.
You are a commercial film director who stages ruined takes on purpose. MY PRODUCT: [what you sell or shoot]MY WORLD: [where your brand lives] Context: I want an ad built on a staged accident. The shoot visibly goes wrong, a knocked camera, a collapsing set, a ruined take, and that wrong moment becomes the ad. It reads real because nobody is posing. 1. Invent one believable on-set accident around my product. Physical, mid-action, nobody posing. 2. Write 3 image prompts of that moment, one camera flaw each: knocked framing, missed focus, hard flash. 3. Give every prompt the craft layer: one motivated light source, a named film stock with heavy grain, a muted three-colour palette, clothing described garment by garment. No logos. Output only the 3 prompts, numbered, under 80 words each.
We filled it in with a black hoodie drop, London five-a-side cage. Claude sent back three staged accidents. Prompt 1: "A clothing rail of black hoodies crashes sideways across wet tarmac inside a caged five-a-side pitch at night, garments spilling mid-air, a stylist in an oversized washed-black puffer lunging after them. Hard on-camera flash against sodium floodlight haze behind chain-link. Knocked framing, horizon tilted. Cinestill 800T pushed one stop, heavy grain, palette of black, sodium amber and wet grey. No logos." That alone gets you a decent frame. We then mixed it into our own prompt system, full camera forensics, a locked colour transfer, and fired it in Higgsfield:
You are a creative director doing visual research. MY WORLD: [your niche + 2 visual refs you love]CULTURE TO RAID: [an adjacent culture, e.g. terrace football] 1. Decode the 10 visual codes of that culture: styling, colour, light, setting, attitude. Keep every code literal, no genre labels. 2. Write 3 image prompts that collide those codes with my world. One collision per prompt, name the code it uses. 3. Give every prompt the craft layer: one motivated light source, a named film stock with heavy grain, a muted three-colour palette. Styled like a campaign, never a snapshot. Output: the 10 codes numbered, then the 3 prompts.
We raided terrace football for a black-and-cream streetwear world. First codes back:
01 Club scarf held taut overhead between two fists
02 Hooded technical jackets, badge on the left arm
03 One cold floodlight tower cutting through night fog
04 Flare smoke drifting over concrete terrace steps
...six more, then the collisions.
COLLISION 1 / the scarf: lone figure on empty terrace steps, striped club scarf taut overhead, flare smoke rolling low, one cold floodlight through fog, Cinestill 800T.
Same process: Claude's collision, mixed into our prompt system with a colour transfer, fired in Higgsfield Soul:
You are my overnight production assistant. MY BRAND: [what you make]MY LOOK: [3 words + one hex colour] 1. Build a world kit: one location line, one cast description, one locked palette, one light source, one film stock. 2. Write a script that generates 5 new image prompts, each a new scene inside this world, and saves them to a dated folder with captions. 3. Make it repeatable: I run it every night, same world, new scenes every morning. Nothing changes between runs except the scene.
First run, one paste. It built the kit and the folder appeared: world_kit/day_01/ scene_1_the_scarf.json scene_2_the_badge.json scene_3_the_banner.json scene_4_the_whistle.json scene_5_the_exit.json Inside scene 1: "a lone figure in a drawstring anorak, hood up, holds a club scarf taut overhead, flare smoke drifting sideways, an empty concrete terrace under one cold floodlight tower... 35mm lens, hard flash, pronounced organic grain" plus a ready caption. Five scenes, every morning, same world.