Turn a photo into pixel art

Pick a photo, frame the part you want, and get a real piece of pixel art at 32, 48 or 64 pixels — with an exact palette, including NES hardware colours and fuse-bead colours. An image model redraws the photo as flat deliberate shapes; the grid and the palette are then fixed here, precisely.

This tool sends your photo to an external image model to be redrawn. It is the one tool here that does that — everything on www.nemawashitools.com stays in your browser.

JPEG, PNG, WebP, GIF or BMP. Up to 20MB.

Takes about 20–40 seconds.

When this helps

Someone making a fuse-bead portrait of their cat

Photo-to-bead converters posterise the photograph, so the result is a blurry cat made of 600 beads in 40 colours you do not own, with no readable face.

Choose your bead range — Hama midi or Perler, and the 24 or 48 colours a starter box holds if that is what you own — and 48 pixels. The model simplifies the cat to flat shapes first, so the eyes and the white chest patch survive, and every colour in the result is a bead you can buy.

A developer who needs a placeholder sprite

Commissioning pixel art for a prototype costs more than the prototype, and a downscaled photograph does not read as a sprite at any size.

Crop to the subject, pick 32 pixels and the NES palette, and you get a sprite-shaped image with a readable silhouette — enough to see whether the idea works before paying an artist.

Someone making an avatar from a holiday photo

Phone photos are mostly background. Shrinking one to 48 pixels spends the whole grid on sky and leaves a face two pixels wide.

Zoom and drag until the face fills the square before uploading. Only the cropped part is sent, and the face gets the whole grid.

FAQ

What is the free preview, and why does the result look different?
The preview is your photograph reduced straight to the grid, in your browser. Nothing is uploaded to make it and it costs nothing, so use it to settle the things it does answer: whether the crop is right, whether the size is big enough, whether your subject survives at 29x29. Finalizing does something else — an image model redraws the picture as flat shapes first, and then the same reduction runs on that. We compared the two blind on forty pairs and preferred the redraw every time, so expect the finished result to be bolder and more deliberate than the preview, not merely cleaner.
Can I just keep the preview?
Yes. "Download this" saves it as a PNG at the grid you chose, for free, without uploading anything. It is a real answer for a simple subject or a quick test; the paid redraw is what you want when the picture has to read as pixel art rather than as a small photograph.
Does my photo leave my device?
Yes — and this is the only tool here that does that. The cropped square is sent to an external image model to be redrawn. Everything on www.nemawashitools.com works the other way and never uploads your file. The crop happens in your browser first, so only the part you framed is sent, not the whole photograph.
Which model redraws it?
"Faithful" keeps the subject, the pose and the framing as close to your photo as possible and is the default. "Stylised" reinterprets more freely — colours, light and setting may change. "Fast" returns in about five seconds. All three send the image to a third-party provider.
Why did one style fail when the others worked?
Each style is a different provider, and each has its own safety filter that checks the picture it just drew. Recognisable characters, logos and brand marks are the usual trigger, and the most faithful style trips it most often precisely because it stays closest to your photo. When that happens, trying Stylised or Fast usually works. Nothing is charged for a style that was declined in a comparison run — the other rows still come back normally.
Why does the same photo give a different result each time?
Image models are not deterministic, and none of the ones used here reports a seed. The same photo at the same settings can come back noticeably better or worse. Running it again is a normal part of using this — the pixel-art step afterwards is exactly reproducible, but the redrawing is not.
Are the NES and fuse-bead palettes accurate?
The NES palette is the standard sRGB approximation of the console's 54 colours; PICO-8's sixteen are exact. The bead palettes are Hama midi's 75 solid colours and Perler's 103, from the open beadcolors database, which samples real beads rather than copying a brand's printed chart. Translucent, glitter and glow-in-the-dark beads are left out, because a colour measured through a transparent bead is not a colour you can lay on a pegboard. Every colour in the output is taken from the palette you chose — that part is guaranteed, not requested.
Do the 24, 48 and 72-colour bead sets match my starter box?
Not exactly, and nothing can: which beads a given box contains varies by product and none of the brands publish it as data. Those sets are the 24, 48 or 72 colours out of that brand's own range that best cover the rest of it, with black and white always included. Use them when you own a small box and want a picture that does not assume 100 colours; check the result against the beads in front of you before you start.
What size is the file I download?
A PNG at its true pixel size — 48 pixels on the long edge means a 48-pixel-wide file, not a 1024-pixel one pretending. Scale it up in any editor with nearest-neighbour to keep the pixels sharp.