Taste is not a mystery

“Taste is the differentiator” has become one of those lines people repeat until it sounds self-evident. I agree that taste matters. I’m less convinced by the way it gets described: as an instinct some people were lucky enough to be born with. That explanation gives you nowhere to go. I’d rather pull the word apart and look at the skills hiding inside it.

I felt this clearly while making Omaru’s launch site. The agent could generate options at speed. My job was to set a direction, reject most of what came back, and explain how to improve the parts worth keeping. It made my own judgement unusually visible.

Taste comes apart

The first layer is visual craft: colour, composition, balance, contrast, hierarchy. Painters, photographers, and cinematographers use the same vocabulary. You can study these principles, practise them, and point to where they succeed or fail in a piece of work.

Then there’s coherence. A single screen might look good while the product as a whole feels stitched together from five different references. Coherence is what makes a photographic series, a film, or an app feel like it came from one point of view. In product work it applies at every scale, from the relationship between two buttons to the relationship between the app and its website.

Models are already capable at both layers when the direction is clear. They know what contrast is. They can follow an established system. The uneven results often come from asking for something vague, then expecting the model to infer the visual standard you had in mind.

The layer everyone gets wrong

The next layer is relevance, and this is where attractive work can still fail. Does the design communicate the idea? Can someone understand it? Does its visual language make sense for the people it’s meant for? A beautifully made answer to the wrong question is still the wrong answer.

I used to think relevance was beyond a model. I’m not so sure now. A model can recognise the difference between a 1970s Braun manual and a Y2K rave flyer. What it can’t know by default is which reference is useful for this product, for this audience, at this moment. That choice still needs direction.

The useful advantage, then, is having a broad map of visual culture and enough vocabulary to navigate it. Tech didn’t invent this language. Poster design, album art, film, photography, comics, and architecture had been developing it long before we started arranging rectangles on screens. The more of that history you know, the more precisely you can choose a direction instead of reaching for whatever currently looks “modern.”

You can get there

Design has always been iterative: make something, look at it, adjust it, and try again. When producing each version took hours or days, experienced designers learnt to discard weak ideas in their heads before drawing them. That speed of judgement looked like a mysterious talent. Some of it was simply an adaptation to expensive iteration.

Agents make versions cheap enough that you don’t have to do all that filtering upfront. You can put several real options on screen and react to what’s actually there. Taste becomes less abstract in that setting. It’s the repeated decision to keep this, remove that, and push one direction further. It’s also the ability to connect those decisions back to craft, coherence, and relevance.

The explanation matters. “It feels off” may be an honest reaction, but it gives the next version nothing to work with. Naming the problem, whether it’s weak hierarchy, mixed references, or the wrong tone, is how judgement turns into useful direction.

That’s why taste feels newly important. Faster production puts more weight on selection and direction. Both improve as you learn the craft, look at more work, and get more precise about your reactions. None of that requires a gift bestowed at birth.

It starts with looking

A practical way to build taste is to pay closer attention to work you already love. Don’t stop at saving the reference. Ask what gives it its character, what holds it together, and why it suits its audience. Over time, those observations become vocabulary. That vocabulary helps you make better work, with an agent or without one.