Convert photos to retro pixel art instantly. Control block size for 8-bit to 64-bit looks. Great for game sprites, thumbnails, and retro aesthetics.
Pixel art is not a low resolution photograph. It is a drawing where every pixel was placed on purpose, on a grid small enough that individual pixels are visible design decisions. Converting a photograph into something that reads as pixel art is therefore a compression problem with an aesthetic answer: you have to decide which few hundred squares and which sixteen colours carry the image, and throw away everything else.
The first step is reducing the grid, and how you reduce it decides whether the result reads as pixel art or as a blurry thumbnail.
Bilinear and bicubic resampling, the defaults in most software, average neighbouring pixels together. That is correct for photographs and completely wrong here, because averaging produces intermediate colours at every edge and the grid stops being crisp.
Nearest neighbour sampling picks one source pixel per destination cell and keeps hard edges, which is what you want, but it also drops thin features entirely: a one pixel wide highlight has a good chance of landing between samples and vanishing.
Area averaging followed by a hard quantisation step usually gives the best of both. Average to get a stable representative colour for each cell, then snap that colour to the palette so no soft intermediates survive.
Whatever the method, the enlargement back to viewing size must be nearest neighbour, or the pixels get smoothed back into mush by the browser.
A photograph has millions of colours. Pixel art usually has between four and sixty four. That reduction is the single most defining step.
Hardware palettes carry a specific cultural weight because they were constraints somebody actually worked within. The Game Boy's four shades of green, the Commodore 64's sixteen colours, CGA's magenta and cyan, the NES palette, Teletext's eight: each of them makes an image read as belonging to a particular machine and era. Bitgrain includes twenty four of these.
When you snap a photograph to a small palette, banding appears immediately in any smooth gradient, most obviously in skies and skin. This is where dithering matters: a Bayer or Floyd-Steinberg dither applied during quantisation scatters the error so the eye reconstructs the gradient. Ordered dithering suits pixel art better than error diffusion because its repeating pattern reads as intentional texture, where error diffusion reads as noise.
At thirty two by thirty two pixels there is no room for subtlety. What survives is silhouette and value contrast, so the preparation is the same discipline as designing a logo.
Push contrast before downsampling, not after. Once the image is reduced, every adjustment is quantised to a handful of steps and fine control is gone.
Simplify the subject. A face against a busy background loses the face, because the background noise occupies the same few pixels the features need. Crop tighter than feels comfortable.
Watch the darkest and lightest points. Small palettes tend to have very few near-black and near-white entries, so anything that depends on subtle shadow detail will flatten out completely.
The most common way to ruin finished pixel art is to export it wrong.
Any scaling other than an exact integer multiple introduces resampling artefacts and destroys the grid. Export at 1x, or at 2x, 3x, 4x and so on, never at 1.5x.
Use a lossless format. JPEG compression works by discarding high frequency detail, and pixel art is nothing but high frequency detail, so JPEG turns clean edges into coloured fringes. PNG is the right choice.
SVG export is also available and is useful when the artwork needs to go on a poster or a garment, because each pixel becomes a real vector rectangle that scales to any size without a resampling step.
Everything runs in your browser, nothing is uploaded, and the quantisation is deterministic, so reopening the project a year later gives you the same grid and the same palette.
Raise contrast, crop tight on the subject, downsample to a small grid with nearest neighbour or area averaging, quantise to a small palette with an ordered dither, then enlarge back with nearest neighbour only.
Four to sixteen for a strongly stylised result, up to sixty four if you need the photograph to stay recognisable. Hardware palettes like Game Boy, C64 and NES give the image a specific era rather than just fewer colours.
Almost always because it was scaled with smooth interpolation somewhere in the chain, or exported at a non-integer multiple. Enlarge only by whole number factors and only with nearest neighbour.
PNG, always. JPEG discards exactly the high frequency detail that pixel art is made of, which produces coloured fringing around every hard edge.