Tool · GPT Image 2 pixel art

GPT Image 2 Pixel Art Generator — 16-Bit Sprites & Scenes

Write one line and get a 16-bit frame — a sprite, a room, a harbour — with hard pixels, a fixed palette and a light source you can actually see.

Marine harbour gate in a storm — 16-bit pixel art, 16:9

Rainforest canopy at night with glowing spores — 16-bit pixel art, 16:9

High-school corridor after hours, long window shadows — 16-bit pixel art, 16:9

Rooftop duel in the rain, lightning between the two — 16-bit pixel art, 16:9

Luffy — One Piece. Drawn from the world's own character atom: straw hat, open vest, both fists loose. Every edge is one pixel step. — **Luffy — One Piece** — Drawn from the world's own character atom: straw hat, open vest, both fists loose. Every edge is one pixel step.

Raiden Shogun — Genshin Impact. Her existing cover went in as the reference, so the braid and shoulder armour survive the downscale. — **Raiden Shogun — Genshin Impact** — Her existing cover went in as the reference, so the braid and shoulder armour survive the downscale.

Hori Kyouko — Horimiya. Asked at full length, so hands, skirt hem and shoes all land on the same grid. — **Hori Kyouko — Horimiya** — Asked at full length, so hands, skirt hem and shoes all land on the same grid.

Zoro — One Piece. Three katana and one closed eye: the props that keep him readable at 32 pixels wide. — **Zoro — One Piece** — Three katana and one closed eye: the props that keep him readable at 32 pixels wide.

Nahida — Genshin Impact. A small figure the model could have smoothed into a generic child; the leaf dress and the floating sigil are still hers. — **Nahida — Genshin Impact** — A small figure the model could have smoothed into a generic child; the leaf dress and the floating sigil are still hers.

Miyamura — Horimiya. Long hair over the collar, tie pulled loose, hands in pockets — the rest is two-value shading. — **Miyamura — Horimiya** — Long hair over the collar, tie pulled loose, hands in pockets — the rest is two-value shading.

Marine dry dock — One Piece. An isometric cutaway: trestles, a crane chain, three value steps of harbour water. — **Marine dry dock — One Piece** — An isometric cutaway: trestles, a crane chain, three value steps of harbour water.

Sumeru camp — Genshin Impact. Night light laid in three steps, spore dots placed by hand, no glow filter anywhere. — **Sumeru camp — Genshin Impact** — Night light laid in three steps, spore dots placed by hand, no glow filter anywhere.

Rooftop at dusk — Horimiya. Two characters in one frame, both on-model, and the light source sits inside the picture. — **Rooftop at dusk — Horimiya** — Two characters in one frame, both on-model, and the light source sits inside the picture.

A world decides the pixels

Pixel art is a style; a world is a set of decisions. Neta Studio keeps the cast, the costumes and the palette in one place, so the tenth frame still matches the first. You describe the beat — who is where, at what hour — and GPT Image 2 draws it inside those rules. Every render lands as a work you can open, re-run or publish.

The cast is already on file — Characters exist as atoms with covers, so "Luffy at the dock" arrives with a face, an outfit and a colour set before you type anything.
One model per image — GPT Image 2 for the frame you keep, a faster model for drafts. Same prompt, nothing to rewrite.
Every frame is a work — Eight candidates stay in the world as works, so you can show the one you kept and forget the rest.

Questions, answered.

Is GPT Image 2 actually good at pixel art?

Yes — when the prompt states the pixel grid, the palette size and the shadow rule. Left alone it drifts toward smooth illustration with a retro filter over it.

Why do my results look blurred instead of pixelated?

The prompt described a mood, not a constraint. Name a low logical resolution, nearest-neighbour enlarging, a 24-colour palette and a ban on anti-aliasing; those lines do more than "pixel art style" ever will.

Can it make a true 32×32 sprite?

Generate at 1152×1536 and downscale with nearest-neighbour, then tidy the silhouette by hand. Native 32-pixel output is unreliable on every image model, GPT Image 2 included.

Can I keep the same character between frames?

Keep the cast block — hair, outfit, one signature prop — identical and pass that character's cover as a reference. In a Studio world both are already stored, which is why the frames here hold together.

Does it do transparent backgrounds?

GPT Image 2 accepts a transparent-background request. If the edges come back ragged, ask for one flat background colour instead and cut it out afterwards.

What size should I ask for?

1152×1536 for a 3:4 sprite, 1536×864 for 16:9 scenery. Pixel art is resolution-independent once the palette is fixed, so a bigger canvas only means more room to draw.

Can the pixel art carry lettering?

Not reliably. Lettering is the weakest thing at low resolution — leave it out of the image and set the title on the page instead.

How is this different from Aseprite or Piskel?

Those are editors: they assume you already know what to draw and every pixel is your hand. This makes the starting frame, on-model, in about two minutes — then you take it into an editor to animate or clean up.

Start with one frame.

A sprite, a room, a harbour — the palette is already decided, so the tenth frame still matches the first.

Make My First Frame