VectorTrace vs vectorizer.ai
Both trace raster images into vector paths. The structural difference is where the tracing happens — their server, or your tab — and everything else follows from that.
- Product
- vectorizer.ai
- Checked
- 2026-08-26
- Rule
- Sourced and dated
- Price
- Free preview
Side by side
| Side by side | vectortrace | vectorizer.ai |
|---|---|---|
| Where the image is processed | The image is traced by WebAssembly in a Web Worker in your own tab. It is never uploaded, so there is nothing to store, delete or leak. | A hosted service: the image is sent to their servers, traced there, and the result returned. Their site describes the API and the web app in those terms.Source · checked 2026-08-26 |
| Works offline | Yes, once the page has loaded. The engine is a WebAssembly module in the tab, not a request. | No — a server-side tracer needs the network by definition.Source · checked 2026-08-26 |
| Output formats | SVG and PDF today; EPS and DXF as those exporters land. All written by the same Rust exporters, from one document. | Multiple vector formats, listed on their site. Check the current list there rather than here — ours would go stale and theirs is authoritative.Source · checked 2026-08-26 |
| Pricing model | Previews are free and unlimited. Downloads need Pro at €9.00 per month or €69.00 per year, tax inclusive. | Credit-based, per their pricing page. We do not restate a competitor price: it changes without telling us, and a stale number is worse than a link.Source · checked 2026-08-26 |
| Public API | POST /api/v1/vectorize, documented with a generated OpenAPI 3.1 document at /api/openapi.json. | An HTTP API is documented on their site.Source · checked 2026-08-26 |
Worth knowing
vectorizer.ai is a hosted service: you upload an image, their machines trace it, and you download the result. That design buys them a straightforward path to heavy models and a lot of server-side compute, and it costs you the upload.
VectorTrace runs the engine as WebAssembly in your browser. That costs us the ability to throw a large GPU model at the problem and buys you a product that works offline, has no queue, and never holds your artwork. Which trade is right depends on whether the file is yours to send.
We have not published a quality comparison, because we have not run one that meets our own standard (D13). Our own figures on the committed corpus are published — boundary error, region count and node ratio, with the date and the engine version attached — and the row below points at them. Theirs are absent because we have not measured them, not because they lost.
What we have not measured
- Which produces fewer nodes for the same boundary accuracy on YOUR artwork? Trace the same file in both and compare the node count your editor reports.
- How does each handle the anti-aliased edge of a small logo? Zoom both results to 800% and look at where the boundary sits relative to the original pixels.
- Does their licence permit the use you have in mind for the output? Read their terms; ours are at /legal/terms.
Benchmark
Measured, not marketed. The corpus, the method and every figure the run produced are published in full.
Methodology → /benchmarks- boundary error, p95
- 0.39 px
- region-count error
- 0.113
- node-count ratio
- 6.35×
Why vectortrace
- Ticket 1 / 3
In-browser and private
The engine is WebAssembly running in a worker on your own machine. Your image is never uploaded — so there is nothing to leak, nothing to delete, and nothing to take on trust.
- Ticket 2 / 3
Editable paths, not a wrapper
The engine groups the image into colour regions, takes the boundary between them and fits curves to it. What comes out is real geometry — one fill per region, every node yours to move — produced in the tab and never on a server.
- Ticket 3 / 3
Formats print and production use
SVG and PDF today; EPS and DXF for cutters, engravers and embroidery software as the exporters land. Real files, opened in Illustrator, Inkscape and LibreCAD — not an SVG with a JPEG inside it.
Questions
- 01Is VectorTrace better than vectorizer.ai?
- We have not measured their output, so we are not going to claim it. Our own numbers on the committed corpus are on /benchmarks, dated and versioned; theirs are not, and a comparison needs both. What we can state is structural: our tracing runs in your browser and theirs runs on their servers. Trace the same file in both and judge the output yourself.
- 02Why does the in-browser difference matter?
- Because an unreleased brand mark, a client file under NDA or a medical scan should not travel to a third party to be traced. If your file is not sensitive, the difference is convenience: no upload wait, no queue, and it works on a plane.
- 03When will there be actual benchmark numbers on this page?
- Ours are there now: boundary error, region count and node ratio on the committed smoke set, each with the date and the engine version. A head-to-head needs their output under terms that clearly permit publishing it, which is a decision about their licence rather than about our harness.