“On Brand” to an AI Model
Written date
Category
Web Design
Exploring how creative direction can become something a generative system can actually use.
I kept coming back to a pretty simple question: when we tell a model to make something “on brand,” what does that actually mean?
As a creative, I can look at an image and immediately know when something feels off. Maybe the lighting is too glossy, the crop feels too tight, the styling is too busy, or everything is technically correct but it still doesn’t feel like the brand. A model doesn’t really have that shorthand.
So I started playing with what it might look like to break those decisions apart and turn them into something more structured.
Instead of one giant prompt, I mapped things like composition, lighting, palette, material, styling, environment, and tone into separate rules and variables. Some could stay fixed at the brand level, while others could change depending on the product, campaign, or type of asset being generated.
Brand references → visual rules → configurable inputs → generation → review
The part I find most interesting is actually what happens after generation.
Two images can follow the same instructions and only one feels right. That judgment is still very human. So I explored a feedback loop where outputs could be reviewed against specific things like composition, material, lighting, styling, and overall brand fit, then fed back into the next round.
Instead of “this doesn't feel right,” feedback becomes more useful:
Too polished. Too centered. Not enough negative space. Lighting feels too commercial. The palette works, but the styling doesn't.
I also started thinking about how the same system could work at a much larger scale, where product data, shot types, and campaign rules feed into generation automatically, predictable issues can be flagged, and a person only steps in where judgment is actually needed.
What I'm interested in isn't really automating taste. It's figuring out which parts of creative direction can become a system, which parts should stay human, and how the two can work together.
“On Brand” to an AI Model
Written date
Category
Web Design
Exploring how creative direction can become something a generative system can actually use.
I kept coming back to a pretty simple question: when we tell a model to make something “on brand,” what does that actually mean?
As a creative, I can look at an image and immediately know when something feels off. Maybe the lighting is too glossy, the crop feels too tight, the styling is too busy, or everything is technically correct but it still doesn’t feel like the brand. A model doesn’t really have that shorthand.
So I started playing with what it might look like to break those decisions apart and turn them into something more structured.
Instead of one giant prompt, I mapped things like composition, lighting, palette, material, styling, environment, and tone into separate rules and variables. Some could stay fixed at the brand level, while others could change depending on the product, campaign, or type of asset being generated.
Brand references → visual rules → configurable inputs → generation → review
The part I find most interesting is actually what happens after generation.
Two images can follow the same instructions and only one feels right. That judgment is still very human. So I explored a feedback loop where outputs could be reviewed against specific things like composition, material, lighting, styling, and overall brand fit, then fed back into the next round.
Instead of “this doesn't feel right,” feedback becomes more useful:
Too polished. Too centered. Not enough negative space. Lighting feels too commercial. The palette works, but the styling doesn't.
I also started thinking about how the same system could work at a much larger scale, where product data, shot types, and campaign rules feed into generation automatically, predictable issues can be flagged, and a person only steps in where judgment is actually needed.
What I'm interested in isn't really automating taste. It's figuring out which parts of creative direction can become a system, which parts should stay human, and how the two can work together.