what is pixel-art-fixer fr?

Pixel Art Fixer takes images that look like pixel art but are not actually built from a clean pixel grid, and converts them into real, grid-aligned pixel art. This is pure image processing, no AI model is used to do the conversion itself. The problem it solves is common with AI image generators, upscalers, and lossy image pipelines: they often produce pictures that look pixelated, but the pixels do not line up to a clean grid, cell sizes are not whole numbers, edges are blurred together, or the image is saved at a much higher resolution than the actual art it contains. None of these fake versions can be properly tiled, recolored, animated, or edited pixel by pixel until they are converted back into true pixels. Most existing tools try to detect a single repeating pattern in the image and snap everything to it, which often picks up the wrong scale, such as a texture or a face, instead of the actual pixel grid. Pixel Art Fixer instead uses three separate detection methods that each look for the grid in a different way. When they agree, that answer is trusted. When they disagree, a more detailed scoring system steps in to resolve the harder cases, including things like non-whole-number cell sizes, warped grids, and heavy blur or JPEG compression damage. Once the grid is figured out, the tool separates two decisions: where each pixel boundary goes, and what color each pixel should be. This keeps edges crisp while still preserving accurate colors, including small details that a simpler tool might lose. According to the project's own benchmark testing across thousands of damaged sample images, Pixel Art Fixer recovers the original image size correctly far more often than other tools it was compared against. The project includes two versions, one written in Python for readability, and one written in Rust that runs significantly faster, meant for servers and processing many images at once. For images too damaged for this method to fully recover, the team also offers a separate AI based version.