Your AI images have no personal style yet because the tool has no way of knowing what yours is. It has looked at a very large number of pictures and learnt what usually goes with what. Ask it for a woman in a greenhouse at night and it hands back the greenhouse most people would draw. Sage, oatmeal, soft light, competent, anonymous.

That is not a fault in the tool and it is not a fault in your prompt writing either. It is a gap in the information. The fix is a single page called a taste sheet, and it takes an afternoon.

What "generic" actually means

An image model fills every gap you leave with the most likely option. Most likely means most common. Most common means the thing that already looks like the rest of the feed.

So a short prompt is a prompt with a lot of blanks in it, and the blanks get filled by the average of everyone.

This is also why asking for "something more original" or "make it more me" changes nothing useful. Those words describe a feeling, and the model has no access to the feeling. Give it a named colour and it can act. Give it a vibe and it reaches for the mean.

Your taste is real, it is just unwritten

You have made thousands of small aesthetic decisions already. The pins you saved. The accounts you follow and the ones you quietly muted. The colour you always reach for. The one you would never put in a piece even if a client asked nicely.

All of that is a specification. It has simply never been written down anywhere the tool can read it.

A taste sheet is where you write it. One page, five headings, your own words, kept open beside you while you prompt.

The five headings

Palette. Three to five colours you keep returning to, named properly. "Deep emerald, aubergine, black, gold as the accent" gives the model something to do. "Rich colours" does not. Say whether you want contrast to be bold or soft, and if you like both, say that you want it to be one or the other on purpose.

Subjects. Who and what appears in the work you love. Be more specific than feels reasonable. Hair described properly rather than "a woman". The animals, the objects, the recurring motifs, the kind of room.

Light and texture. Glossy or matte. Candlelight or midday. Clean vector shapes or visible brush marks. Layered collage or one clean form on an empty ground. Pick a side for each. A model cannot half choose, so if you do not choose it will.

Refusals. The heading that does the most work and the one people skip. What you will not accept even when it comes back looking nice. The politely muted middle. The figure standing neutrally with no mood. The stock face. Write down what makes you sigh when it arrives, because the average will keep offering it.

Loves. Five to ten references. An artist, a film still, a book cover, a particular pin. You are not going to paste these in. You are going to write a line under each saying what you love about it, and that line is where taste turns into usable words.

Then one closing sentence that says what the whole page adds up to. Something like: opulent, glowing, culturally specific, bold or tender and never the middle. That sentence becomes the test you hold every image up against.

How a taste sheet reaches the prompt

You do not paste the page in. You lift from it.

Write your subject and one action the way you normally would. Then take two or three lines off the sheet and add them: the palette line, the light line, and one refusal turned into a positive. "She has a mood and a stance" works. "Not neutral" does not, because the model has nothing to draw from a negative.

Most days you will reach for the same two or three lines. That repetition is the point. Repeated choices are what a style is.

The way to see whether it works is to run one prompt twice. Take something you made last week, add three lines from the sheet, and put the two results side by side. The seasoned version usually looks like it was made by a person with an opinion, which is the whole aim.

Where reference tools fit

Style references and character references in MidJourney carry a look forwards once you have found one. They are good at that job and worth learning. They are the second step, though, because a reference image can only repeat a decision. It cannot make one.

The taste sheet is the first step. It decides which look is worth carrying at all. If you want the mechanics of holding one look steady across a whole set once you know what your look is, how to get a consistent art style with AI image generation covers that side. And if you have not made a first image yet, start with I want to make AI art but I don't know where to start.

Permission, since it usually comes up

Your taste does not need to be fashionable. Gold on everything is allowed. Nothing but line work is allowed. Chibi businesswomen in navy and cream are allowed.

The average will sand all of that off if you let it, and letting it is easy, because the average always looks competent. Competent is a low bar to clear and a poor thing to be known for.

Pixi's Burrow helps creative founders give their AI images personal style through prompt builders that hold the structure of a strong prompt and leave the taste to the person using them. Unlike a pack of finished prompts you copy word for word, every builder shows its moving parts, so what comes out carries the maker's choices rather than the seller's.

Start this week

Open a blank page. Write the five headings. Fill them badly, because a bad first version is easy to correct and a blank page is not. Add the closing sentence when the rest is down.

Then take one old prompt, add three lines from your sheet, run it again, and look at the pair.

Which artist do you keep getting instead of yourself?