Free Vectorizer Benchmark

What we found

• White dots are an engine problem, and two engines avoid them. VTracer left under 0.01% of the artwork see-through and ours 0.02%. ImageTracer.js left 9.9% on average, up to 100% on the doodle, and AutoTrace 2.1%.
• Color accuracy depends on how an engine picks its colors. Vectorizely had the lowest color error overall (2.09). By group, Vectorizely was closest on flat logos and icons, on JPG and small logos, on illustrations and on gradients and photos; ImageTracer.js on line art. AutoTrace's files were a median 28 times the size of ours. VTracer's defaults merge close shades into one color, and its color error was 5.82, the highest of the four.
• File sizes differ by more than ten times. The median SVG was 9.9 KB from ours, 37.2 KB from VTracer, 132.5 KB from ImageTracer.js and 278.2 KB from AutoTrace. Anchor points, the points you'd edit, follow the same order.
• For black-and-white line art, our B&W mode is the most faithful. It reproduced 96.4% of the ink, against 96.3% for Potrace and 91.3% for VTracer's B&W preset, with the fewest anchor points (573 against Potrace's 688) and the smallest files.
• No engine makes a good vector of a photo. The closest on our photo was VTracer, at a color error of 3.8 and 1,061 KB. Every result is flat patches of color.

What changed since the first run

We first ran this benchmark on 2026-10-04. Since then we changed two things in our engine and one in the scoring, and reran every engine on the same images. The other engines ran exactly as before.

• Our B&W mode. The first run had Potrace ahead on line art, with 96.3% ink overlap against 92.8% for our B&W mode, whose strokes came out slightly heavy. We rebuilt it to trace contours straight from the grayscale image, so edges land inside pixels instead of on pixel boundaries. It now reproduces 96.4% of the ink.
• Soft shadows in our color mode. A soft drop shadow or glow used to come out as a hard dark edge on one side and nothing on the other. Our color mode now traces it on its own, as a stack of see-through shapes under the rest of the image, one per level of opacity. Its mean color error went from 2.18 to 2.09.
• How color error is averaged. The first run averaged it over the pixels that either the source or the SVG covers. AutoTrace can't read transparency, so it ran on images flattened onto white and traced the white background too, which spread its error over every pixel while the other engines' error was spread over the artwork alone. Color error is now averaged over every pixel, for every engine. The first-run figure above is scored the same way.

By kind of image

Mean color error per group. The lowest in each row is in bold.

Up close

Each row zooms into the part of one image where the engines disagree most, picked by the scoring script, not by hand. The original is shown as raw pixels, and each SVG is drawn sharp at the same zoom.

Black-and-white line art

On the 4 line art images we also ran the black-and-white engines: Potrace, VTracer's B&W preset and our B&W mode. They leave the paper transparent, so instead of see-through pixels this table scores ink overlap: of the pixels that are dark in the source or in the SVG, the share that are dark in both. Higher is better.

Which free tools run these engines

These free tools name the engine they run on their own site, in their docs or in their code (checked 2026-10-04). How close a tool comes to its engine's results here depends on its settings and on anything it does to the image first. Several popular converters, among them Convertio, PicSVG and pngtosvg.com, don't say which engine they run, so they aren't listed.