2026-08-07

Cleaning Up a Pixel Sprite's Background Before It Goes in the Game

Once AI has generated your pixel character's animation, one stretch of work is left: strip the background, clean the edges, and export at the size your engine needs. It is duller than generating, but cutting corners here shows up in the game as a halo around your character or as pixels that look smeared. This walks through that last stretch in order.

Why background removal is harder on pixel art

The same background removal looks different on an illustration and on pixel art. The difference is resolution.

AI video generation animates your character over a green screen. When that green is stripped, ambiguous pixels always survive along the boundary — pixels that are part character colour, part green. Green fringe. On a 1024px illustration that band is under 0.1% of the image and nobody notices. On a 64px pixel character, the same single row is 1.5% of the character's width. Zoom in and you see a line of green-tinted pixels tracing the outline.

In-game it gets clearer. Dark backgrounds make the character's edge glow; light backgrounds give it a dirty outline. In pixel art a single pixel is a unit of design, so that one row simply reads as sloppy.

Step 1: let automatic removal take the big areas

Start with automatic background removal. Wherever character and background separate cleanly by colour, this handles it.

There is one value worth tuning: how aggressively to cut the edges. Cut hard and the green disappears cleanly, but so do the character's thin parts — hair tips, blade edges, the trailing corner of a cape. Cut softly and detail survives, but green remains. On pixel characters thin details are often only two or three pixels wide, which makes cutting hard the riskier mistake. Cut softly and clean the rest by hand.

Step 2: magic wand for the trapped areas

Automatic removal always leaves background trapped inside the character: between arm and torso, between legs, inside the grip of a weapon. One click of the magic wand clears each of those.

The tolerance value matters. Low tolerance erases only pixels almost identical to the one you clicked; high tolerance spreads into similar colours. Pixel art uses few colours, so a low value (around 16) usually catches the whole region. Push it up and at some point it starts eating the character's own dark areas — start low and raise it gradually.

Step 3: turn on the pixel grid and use a square brush

This is the part that actually matters for pixel art.

Zoom in far enough and one image pixel becomes a large square on screen. Turn on the grid and each cell is exactly one image pixel, so you can see precisely which pixel you are about to erase. Without the grid, magnification alone blurs the colours together and the edge you aim at is not the edge you erase.

Brush shape matters just as much. A round soft brush is the wrong tool for pixel art. It erases progressively softer toward its rim, which leaves half-erased pixels everywhere it passes — pixels sitting at 40% or 70% alpha. They are hard to see on screen and show up in the engine as a translucent smear around your character.

A square brush snaps to the pixel grid. Set it to 1 pixel and exactly one cell is erased. When you are cleaning an outline one pixel at a time, this is the only way to be precise.

Check by changing the background colour

Even when you think it is clean, leftover pixels are visible against some backgrounds and invisible against others.

Inspect a dark character against a dark background and the leftovers hide in it. Switch to white and they appear. Light characters are the reverse. Magenta — a colour that overlaps with almost no character palette — reveals translucent pixels best of all; pixel tooling has used it to mean "this is not part of the sprite" for decades.

Viewing the transparency as a flat colour rather than a checker pattern helps too. The checker is itself a grid, and two grids of different sizes over the same area stop the eye from telling art from background.

The last trap: changing the export size smears the pixels

This one is easy to miss even for the people building the tools, so it gets its own section.

When you export a sprite sheet and specify a frame size, the images get resized — and *how* they get resized decides everything. Most image libraries default to smooth interpolation (lanczos, bicubic and friends). Excellent for photographs, destructive for pixel art.

A simple test shows it. Take an 8×8 checkerboard using exactly two colours and scale it to 32×32:

- Smooth interpolation: 66 colours — 64 in-between shades that were never in the original - Nearest neighbour: 2 colours — the original palette, untouched

Smooth interpolation invents colours between pixels. Every hard edge becomes a gradient, and all the pixel-level cleanup you just did disappears in a single step. Resizing pixel art requires nearest neighbour, always.

To be straight about it: GenioPlus defaulted to smooth interpolation here for a while, and we fixed it to nearest neighbour recently. If you use another tool, check whether its resize options include "nearest neighbor" or a "pixel" mode. If they don't, don't change the size — export at original size and scale in your engine instead.

In short

Four steps for the last stretch of a pixel sprite: automatic removal for the big areas, magic wand for the trapped ones, square brush over a pixel grid for the edges, and a background-colour switch to verify. Then confirm your export resizes with nearest neighbour.

None of it takes long. It only requires knowing what you are looking at in each step — and you cannot know that unless the tool shows you the pixels.