Image Vectorizer

Notes

• The picture is first split into ink and paper at a threshold (chosen automatically with Otsu's method, or by the slider), after an optional light blur and a despeckle pass that drops blobs and pinholes below a size you set. Invert it for light-on-dark artwork.
• Outline traces the boundary of every ink region. The default engine, AI Vision, is built on VTracer by visioncortex (running as WebAssembly) and fits smooth Bézier curves to the boundaries and can also work in colour, clustering the picture into a stack of filled shapes with one path per region and one DXF layer per colour.Pixel-exact is the built-in alternative: it follows the exact boundary of every ink region along pixel edges, outer boundaries and holes alike, then simplifies the result (Ramer–Douglas–Peucker) to the tolerance you choose. Use it for logos, stencils, silhouettes and anything you will cut out or fill.
• Centerline thins each stroke to a single pixel down its middle (Zhang–Suen) and traces those into open and closed paths. Use it for line drawings, handwriting and sketches you will plot, engrave or laser-score in one pass — you get one line per stroke instead of two edges.
• SVG exports keep AI Vision's curves; DXF gets them flattened to fine polylines, which every DXF reader understands. Exports are in millimetres — set the output width and the DXF header says so.
• Images larger than 2400 px on their long side are scaled down before tracing. For the cleanest result start from the largest, sharpest source you have (a PNG rather than a JPEG screenshot), then raise Simplify until the point count is sensible.
• Everything runs in your browser; the image never leaves it.