The short answer
- Image models are brilliant at variety and terrible at consistency. A brand needs the opposite.
- A prompt system turns your art direction into reusable building blocks.
- Generate wide, then curate hard. A designer always makes the final call.
Why one-off prompts fail
Type a clever prompt into an image model and you get a striking picture. Type it again next week and you get a different light, a different lens and a different mood. Across a campaign, that inconsistency reads as cheap.
Build the system
- A style block. Lens, light, palette, texture and composition, written once and reused in every prompt.
- Subject templates. Product, people, place and abstract, each with fixed framing rules.
- Negative rules. What the brand never shows: certain colours, cliches, stock poses.
- Reference images. A small, curated set the model is guided by, so the look holds across tools.
“Generate wide, curate hard.”
The production loop
We generate in batches, shortlist the strongest 10%, then retouch and colour-match them in Photoshop so every image sits in the same world. Tools change monthly, so the system lives in a shared document, not inside one app.
Where people stay in charge
AI makes exploring fast. Taste is still human. The art director chooses, refines and says no, and that is what keeps a feed looking like one brand instead of a hundred prompts.




