# Pixel art is a constraint, not a filter

> Real pixel art has a grid, a fixed palette, and every square placed on purpose. Generated versions usually fake all three, and the difference shows up the moment you look closely.

August 17, 2026 · https://justog.club/stories/retro-pixel-art-tee

Pixel art looks like the easiest style to ask for and it's one of the hardest to actually get.
Not because the drawings are complicated — because the whole discipline is a set of restrictions,
and restriction is the one thing generation is worst at.

## What makes pixel art pixel art?

Three rules, and breaking any one of them produces something that only looks right at a distance.

- **A real grid.** Every pixel is the same size and lands on the same lattice. No drifting
  squares, no rounded corners, no pixels that are one and a half pixels wide.
- **A fixed, small palette.** A dozen colours, chosen deliberately, reused everywhere. Not
  hundreds.
- **Every pixel is one flat colour.** No gradients, no anti-aliased edges, no in-between tones
  that appeared because something was resized.

That's the entire definition, and it's why the craft exists: when you only get a few hundred
squares and twelve colours, deciding *which* squares carry the shape is the whole job.

## Why does generated pixel art look wrong?

Because most models make a normal picture and shrink it, which is a different thing entirely.

Image models work in a smooth latent space rather than on a lattice, so what comes back is
typically a high-resolution illustration downscaled to look chunky. The artefacts are consistent
and easy to spot once you know them: muddy or anti-aliased edges, pixel sizes that vary across
the image, and palette bloat — hundreds of near-identical colours where there should be twelve.

It fakes the aesthetic and misses the discipline, because the discipline is *scarcity* and these
systems are trained on abundance. That's not a flaw you can prompt your way past with the word
"pixel."

## So how do you actually get it?

By asking for the constraints as facts, and by looking at the result at full size.

Concrete requests do land: **large uniform pixels, twelve colours maximum, no anti-aliasing, hard
square edges, no gradients, dithering instead of blends.** Naming a coarse grid explicitly — the
sprite is 48 squares across — gives the model something to be consistent about rather than a
style to imitate.

Then there's a judgement only you can make, and it's the reason this genre is different from the
others here: **look at the actual squares before you commit to it.** Not for quality — for whether
the grid is real. A design that only holds together as a thumbnail is a design that will be a
blur at chest size.

## Is a coarse grid better on a shirt?

Usually, and this is the useful bit of luck in the style.

The coarser the grid, the more it works as a garment print. Big squares mean big flat shapes,
hard edges and no fine detail to lose — the same properties that make any print legible across a
room. A 32-pixel-wide sprite blown up to chest scale reads instantly; a 256-pixel one becomes
texture.

It also solves the palette question for you, because a coarse grid can't support many colours
without turning to noise.

## Can I have a character from a game?

No — sprites are artwork, and a recognisable one is protected regardless of how few pixels it
has. Being small doesn't make it public.

The technique isn't owned by anyone. Grids, dithering, limited palettes and the hardware-era
conventions they came from are common property, and the constraint that produced them is
reproducible by anyone who accepts it.

The reason to make your own is more practical: a familiar sprite reads as merchandise, whereas an
unfamiliar creature drawn with the same discipline reads as a thing you found.

## What subjects work best?

Things with a strong silhouette and no reliance on faces.

Insects, tools, animals, machines, food, weapons, plants — anything whose identity survives being
reduced to a shape. The beetle above works because a beetle is silhouette-first: horn, body,
legs, done.

Human faces are the hard case, because at low resolution a face is about four pixels of
information and the difference between an expression and a smudge is one square. Traditional
sprite artists solved that by exaggerating wildly rather than by adding detail — and that's still
the answer.

## How do I get one made?

Describe the subject at [JustOG](/) — what it is, roughly how many pixels wide, and how few
colours. Pick a direction, drag the crop frame, see it composited on the real garment, and it's
made to order and shipped.

Designs other people have published are in the [shop](/shop).

The look came from not having enough. Asking for the shortage back is the only way to get it.