Turn any image into a clean, editable vector.

Before / after

Zoom in until one of them breaks.

Same cat, twice. On the left, the source PNG — real pixels, and at 16× they are all you get. On the right, the SVG GetVect traced from it: curve-fitted outlines that stay sharp at any magnification. Drag the divider, change the zoom, then drag the picture itself to go looking somewhere else.

SOURCE PNG TRACED SVG

1× 4× 16×

Drag the divider · drag the picture to pan when zoomed · arrow keys do both

This trace is the local engine alone. AI Enhance was off, nothing was uploaded, and no model touched the image before or after it.

• 17 KBSVG, from 568 KB
• 42shapes, 7 layers
• 8colours, Smart AA
• 0network calls

That is Frankie at 8 colours with Smart anti-aliasing: a 568 KB raster in, a 17 KB SVG out, in 7 editable colour layers and 42 shapes, transparent background intact.

Flat sticker art is the easy case, and every vectorizer looks good on it — hard edges, few colours, nothing to guess at. So the numbers above are not measured on Frankie. They come from noisy logos, photographs and deliberately awkward synthetic art, on every build. If your input is messier than a sticker, that is the case being tested.

From a script, or an agent

It is a command, not just a window.

The same engine, without the UI. Point it at a file and it writes the vector beside it.

getvect logo.png                 # -> logo.svg
getvect shot.jpg -f dxf -c 16 -p photo
getvect logo.png out.svg --stats # JSON on stdout

Every setting the app exposes is a flag. It writes nothing to stdout unless you ask, exits non-zero with one line on stderr when something is wrong, and never prompts — because the thing calling it usually cannot answer a question.

It will not overwrite an existing file. Writing to a path that is already there exits 73 and writes nothing, so a caller that guessed at a filename cannot destroy one it never saw. Pass --force when replacing is what you meant.

SKILL.md is that contract written for an agent rather than a person: when to reach for it, when not to, how to pick settings for a logo versus a photo versus line art, and what to do on each exit code. The full CLI reference has every flag, range, default and failure mode.

What it does

Every control the job needs, on your own hardware.

Not a one-button converter. Every setting a real job needs is exposed, and every one of them runs on your machine.

AI Enhance — optional, bring your own key

For shaded artwork that no tracer handles well.

Soft shading and gradients are the hard case: a tracer has to guess where one colour ends, and it guesses badly. The fix is not a better filter — it is redrawing the picture flat before tracing it. AI Enhance does exactly that: background removed, shading flattened into bands, outlines redrawn at even weight. The tracer then works on art that is already clean, and the result is dramatically better than any amount of noise reduction.

GetVect can do the same step by asking an image model for it. Paste your own Google Gemini API key, switch AI Enhance on, and the enhanced image becomes the working image the engine traces. One call, about eight seconds on a 1024 px source.

The fully-local version — background removal, then an on-device generative flatten, with nothing leaving the machine — is the plan of record in issue #1. The provider layer was written for it. Setup, quality tiers and the debug switch are in the docs.

Download

One file. No account, no installer that asks questions.

macOS on Apple Silicon (M1 and later) and Windows on x64. Intel Macs and Linux build from source (npm run dist) but are untested, so they are not offered as downloads.