Convert JXL to SVG

Converting JXL to SVG is not a re-encoding, it is a redrawing. A JPEG XL holds a grid of pixels and an SVG holds shapes, so the shapes have to be guessed at by tracing the edges — which works beautifully on a logo and badly on a photograph.

  • Where it runs In your browser. The file is never uploaded.
  • Rebuilt SVG works differently from a JXL, so this is not the gradual degradation a lossy codec applies. What SVG can express is reproduced faithfully; what it has no equivalent for does not survive at all.
  • File size limit Up to 100 MB per file, free, without an account.
  • Worth knowing Tracing redraws the picture as flat shapes, which is what lets the result be scaled to any size without going soft. It works by finding the edges between areas of similar colour and drawing an outline around each one, so it is at its best where those areas genuinely exist: a logo, an icon, a flat illustration, a scanned line drawing. A photograph has no flat areas, so tracing invents them — the result is posterised and larger than what you started with. The traced file is measured against yours, and the page tells you which of the two happened before you download it.

Up to 100 files at once. Mixed formats are fine.

Tracing a JPEG XL is a redrawing, not a conversion

Rendering an SVG into pixels is deterministic: the file says what the shapes are and a renderer draws them. Going the other way there are no shapes in the file at all — a JPEG XL is a grid of colours, and every curve, corner and outline an SVG would need has to be inferred from where those colours change.

That inference is what tracing is, and it is why this pair is described differently from every other one on the site. Nothing about the output is the input. The result is a new drawing that resembles the original to a degree that depends entirely on what the original was, which is the subject of the next two sections and the reason this page exists in the form it does.

What a JPEG XL photograph becomes when you trace it to SVG

It becomes a bad photograph in a large file. The tracer works by finding regions of similar colour and drawing an outline around each one, so it needs flat areas to exist. A photograph has none — it has a continuous gradation of tone in every direction — so the algorithm invents the flat areas by quantising the colours, and the visible result is posterisation.

This was measured before the feature shipped rather than assumed afterwards: a photograph came out around four times larger than the source and visibly posterised, and a gradient banded for the same reason, because a gradient is precisely the absence of the flat regions the algorithm looks for. JPEG XL lists photography among its uses, so a large share of the people who land on this page have exactly that kind of file, and the honest advice is not to convert it.

What traces well out of a JPEG XL

Flat artwork, and it works better than most people expect. A logo round-trips faithfully — in testing a 6 KB source became a 16 KB SVG with its sharp edges, its correct colours and its rounded corners intact, and the result scales to a billboard without softening. Icons, single-colour marks, flat illustrations and scanned line drawings all fall into the same category.

The common thread is that these images genuinely contain the thing the tracer is looking for: areas of one colour with a definite boundary between them. The SVG that comes out is larger in bytes than the source image and infinitely smaller than a raster large enough to print, which is the trade the format exists to make. That is the case this conversion is for, and it is the only case where it should be the first thing you try.

Text is the worst thing to trace out of a JPEG XL

A screenshot containing text produced an SVG roughly forty times the size of the source. The reason is anti-aliasing: every letter is drawn with a fringe of intermediate colours, and each of those fringes becomes its own small pile of paths. Multiply that by a page of type and the file explodes while looking worse than the pixels it replaced.

If a JPEG XL is mostly words, the question is almost always what you want the words for. To read or paste them, text recognition is the right tool and gives you characters instead of shapes. To keep the look of the page, a raster format keeps it at a fraction of the size. Tracing type is only defensible for a wordmark — a handful of letters that are artwork rather than text — and there four colours and smooth detail is the setting to start from.

Choosing how many colours the SVG keeps

The colour count is how many flat colours the drawing is reduced to, and it is the setting that decides everything else. Four suits a logo or a single-colour mark and produces the cleanest shapes and the smallest file. Eight is the default and covers most flat illustrations. Sixteen follows shading more closely at a real cost in size, and thirty-two comes closest to the original while producing much the largest file.

The instinct is to raise it, and on the wrong image that instinct is what turns a poor result into an unusable one. More colours on a photograph means more invented regions and more paths describing them, not more realism. The rule that holds: raise the count only when you can see a specific flat area in the original that the trace has merged into its neighbour. Lowering it is the more useful direction. A two-colour mark traced at four colours often comes back with a third and fourth colour made entirely of the anti-aliased edge between the other two, and dropping to four — or redrawing the mark on a clean background before tracing — removes those outline shapes rather than tidying them.

Smooth, balanced or sharp when tracing the JPEG XL

The detail setting controls how tightly the paths follow the pixels. Smooth accepts a looser fit and discards shapes below sixteen pixels, which removes the specks anti-aliasing leaves along every edge. Balanced is the default and halves both thresholds. Sharp follows the pixel grid closely and keeps shapes down to two pixels, specks included.

Blurring the image before tracing would reduce the speck count further and is deliberately not done, because it softens exactly the sharp edges that are the only reason to trace a logo in the first place. Start at balanced. Move to smooth when the result is littered with stray fragments around the outlines, and to sharp only when a genuine small detail — a serif, a thin rule, a notch in a mark — has been swallowed.

The 1,600-pixel ceiling, and why it costs the SVG nothing

The image is scaled down to 1,600 pixels on its longest edge before tracing. Tracing cost grows with the number of edges rather than with the file size, and a large photograph has an edge almost everywhere, so left alone a 12-megapixel source can produce a multi-megabyte SVG over tens of seconds — a worse outcome than the one you were trying to avoid.

Unlike a raster conversion, this is not a loss of fidelity. The output is resolution-independent, so a logo traced at 1,600 pixels wide and the same logo traced at 4,000 render identically at any size; what changes is only how closely the paths hug the original pixel grid. The downscale is done with the browser’s filtered scaler rather than a nearest-neighbour loop, because an unfiltered downscale would alias every diagonal into a staircase the tracer would then faithfully reproduce.

Both file sizes are shown before you download the SVG

Nothing here inspects the image and predicts whether it will trace well. A colour-count heuristic would be a guess, and the real answer is available for free a moment later: trace the file, then compare the two byte counts. The interface shows both, so you find out which of the outcomes above you got before you commit to it.

That is the check worth making a habit. An SVG smaller than or close to the source almost always means the trace found real shapes. An SVG several times larger means it invented them, and the honest response is to discard the result rather than to tune the settings — no combination of colours and detail turns a photograph into flat artwork, because the flat artwork was never there.

The SVG is editable in a way the JPEG XL never was

This is the real prize when the trace works. A JPEG XL is not an editable format: you can filter it, crop it and re-encode it, but you cannot select the mark and change its colour, because there is no mark, only pixels that look like one. An SVG is a text file describing shapes, and Illustrator, Inkscape and Figma all open it as objects you can move, recolour and delete.

So the reason to trace a logo is rarely scale alone. It is to recover an asset that exists only as a picture — an old mark from a scanned letterhead, an icon whose source file nobody kept — into something that can be worked on again. The output is also a text file, which means it can be opened in an editor, diffed, and put under version control like any other source, and its fills can be changed by search and replace when a brand colour moves. Where the original vector file still exists somewhere, finding it always beats tracing, because a trace is an approximation and the original is not.

When to redraw instead of tracing the JPEG XL

A trace inherits every flaw in its source. If the JPEG XL is a low-resolution copy, a screenshot of a logo, or an image that has been through several lossy re-saves, the tracer faithfully reproduces the softened edges and the compression artefacts as shapes, and the result looks approximately right and wrong in detail at large sizes.

For anything that will carry a brand — a sign, a print run, a website header — a trace is the starting point and not the delivery. Somebody redrawing the mark properly in a vector editor, using the trace as a template, produces a file with clean curves and correct geometry. This conversion is excellent at getting you something to work from and should not be mistaken for something to ship.

How to trace a JXL image into an SVG

  1. Drop the JXL file here — a logo, an icon or a flat illustration, not a photograph.
  2. Set the colours (four for a logo, eight for an illustration) and leave detail on balanced.
  3. Compare the two file sizes shown before downloading. Much larger means the trace did not find real shapes.

JXL and SVG: a grid of pixels against a list of shapes

JXL compared with SVG
JXLSVG
Full nameJPEG XLScalable Vector Graphics
File extension.jxl.svg
Media typeimage/jxlimage/svg+xml
CompressionEither, depending on the settingUncompressed
First published20212001
Published byJoint Photographic Experts GroupW3C
SpecificationISO/IEC 18181SVG 1.1
LicensingOpen standardOpen standard
Standing todayNicheCurrent
Bit depth32
Colour it can describeRGB, greyscale, wide gamutRGB
Opens in a browserSome browsersEvery browser
Considered insteadAVIF, WebP, PNGPNG, PDF

What survives

Transparency survives. Both JXL and SVG store an alpha channel, so a cut-out stays cut out and nothing is filled in behind it.

Animation survives. JXL and SVG both hold multiple frames, so the result moves rather than freezing on the first one.

What the target format adds

SVG is a working format and JXL is a finished one. What comes back is editable text and objects rather than a picture of a page, which is usually the reason for the conversion and also where its limits are.

Opening the result

SVG opens in every current browser. JXL has narrower browser support than that. If the file is going onto a web page or into a form, that is usually the whole reason for the conversion.

The usual programs do not overlap: JXL opens in GIMP and ImageMagick, SVG in Inkscape, Adobe Illustrator and Figma — so whoever receives the result needs something from the second list.

What each format is for

The two are aimed at different work: JXL at archiving and photography, SVG at the web, line art and logos and editing. That is worth weighing before converting, because the reason one exists is usually the reason the other is awkward.

JXL is Joint Photographic Experts Group's format, published in 2021. It records 32 bits per channel.

SVG comes from W3C and dates from 2001, specified as SVG 1.1. Inkscape, Adobe Illustrator and Figma all read it.

SVG was published in 2001 and JXL in 2021. The older one is generally the safer file to hand to somebody; the newer one usually does the job in fewer bytes.

JXL to SVG: what traces well and what does not

Are my JXL files uploaded anywhere?

No. This conversion runs entirely inside your browser, so the file never leaves your device. You can confirm it yourself: open the network tab of your browser's developer tools and convert something. You will see the page load, plus the analytics and advertising the site is paid for with — and nothing carrying your file.

Will this work on a photograph?

No, and it is worth being blunt. Tracing works by finding edges between areas of flat colour, and a photograph has none. In testing a photograph came out roughly four times larger than the source and visibly posterised. If your JPEG XL is a photograph, this is the wrong conversion.

What does work?

Flat artwork. A logo round-trips faithfully — in testing a 6 KB source became a 16 KB SVG with sharp edges, correct colours and rounded corners intact, and it now scales to any size. Icons, single-colour marks, flat illustrations and scanned line drawings behave the same way.

How many colours should I choose?

Eight is the default and suits most flat illustrations. Four is right for a logo or a single-colour mark and gives the cleanest shapes and the smallest file. Sixteen and thirty-two follow shading more closely and produce much larger files, which on the wrong image is a worse outcome than not tracing at all.

Why did my screenshot produce an enormous SVG?

Because every anti-aliased letter becomes its own pile of paths. A screenshot containing text was measured at around forty times the size of the source image. Text is the worst possible input for tracing; if you want the words, use text recognition instead.

Is there a resolution limit?

The image is scaled down to 1,600 pixels on its longest edge before tracing. That costs nothing in the output, because an SVG is resolution-independent and renders identically at any size — it only stops a large photograph producing a multi-megabyte file over tens of seconds.

How do I know whether it worked?

Both file sizes are shown before you download. Nothing here tries to predict from the image whether it is traceable; the real answer is available a moment later by comparing the byte counts and looking at the result.

More about these formats

Where these figures come from

The claims this page makes about JXL and SVG are checkable, and these are the documents that settle them.