How it works

From a dropped file to an SVG you can edit, in six steps. Each child page goes into one part in depth.

1. Load and prepare the image: decode it, downscale big ones, upscale small ones. → The pipeline
2. Find the colours. Flat art’s real colours are sharp peaks in its histogram; those become the palette. → Palettes
3. Trace. Every preset uses Inlay, our own shared-edge tracer, which traces every boundary between two colours once, so the pieces fit together like inlaid wood. → Inlay
4. Fit curves. Each boundary becomes the fewest straight lines and curves that follow it, with sharp corners kept sharp; Pixel art keeps its exact pixel staircase instead. → Curve fitting
5. Optimise with SVGO in the background, so the numbers shown are the numbers you download.

Throughout, quality is measured rather than judged by eye. → Measuring quality

Why classical tracing, not AI

The research before any code was written (Research) found that no downloadable model beats a good classical tracer at the tracing itself. The learned models (StarVector, OmniSVG) need a 16 GB GPU and 10–80 seconds per image, and on illustrations they lost to VTracer outright. AI earns its place around the tracer, for removing backgrounds and upscaling low-resolution logos, and that is on the roadmap.