Convert an image into vector shapes you can scale, edit and cut
One trace of your picture, written out as SVG, PDF, EPS or DXF.
Convert and preview without an account. Downloading is free too — it just needs a Google sign-in. No card, no subscription.
01 · 5 colours · 193 nodes
How it works
- Choose one PNG, JPEG, WEBP, BMP, GIF or AVIF file, up to 25 MB and 40 megapixels.
- Set the colour count between 2 and 64: low for flat marks and cut lines, higher for shaded artwork.
- Tick the outputs you need (SVG, PDF, EPS, DXF, PNG) and add a width and height in millimetres if the piece has a physical size.
- Compare the preview against your original, and change the colour count if shapes have merged or splintered.
- Sign in with Google (free) to download; pick more than one format and a ZIP arrives with them.
"Vector" is less a file type than a claim about what a picture is made of. A photo is a grid of coloured squares measured at a fixed density, so enlarging it forces the software to invent the measurements nobody took. A vector holds the same picture as a list of shapes: each region of colour described by its outline, and that outline a handful of coordinates with curve handles between them.
Three things follow, and together they explain why you keep being told to "send it as a vector". It re-renders at whatever dimension the device asks for. Every corner is a node you can drag, recolour or delete. And a cutter or plotter can follow the outline as a path rather than interpret a picture of one.
This page does that to a PNG, JPEG, WEBP, BMP, GIF or AVIF and hands the result over in whichever container the next tool expects: SVG, PDF, EPS, DXF, or a re-rasterised PNG. The geometry is built once and every writer serialises the same shapes, so a different output is not a different trace; it is one set of outlines spelled in another language.
Doing the tracing is colour segmentation, an explicit anti-aliasing model and curve fitting, combined with machine learning (AI) that runs on our own servers. The same image at the same settings gives the same result: predictable rather than inventive. Flat artwork (logos, badges, stickers, stencils, line art) comes through cleanly. Photographic shading has no outlines in it, so it arrives as many small flat regions.
What actually happens
- Raster 6×6 · 8 px · PNG 77 B
- 6 nodes · 1 path
- SVG 91 B · vector, not pixels
What survives the trip from image to vector, and what does not
Hard edges survive tracing; gradients turn into bands
The tracer puts a curve wherever one flat colour meets another. A logo, a badge or a line drawing is made of such meetings, so nearly all of it survives. A gradient has no meeting point, so the segmentation has to invent some: the ramp becomes a staircase of flat bands, as many as the colour count allows. The preview shows that before you download anything.
The vector is never sharper than the image
Every curve is fitted to the boundary the pixels describe. The anti-aliasing model reads the two-pixel ramp along an edge and places the outline where that ramp crosses halfway, which recovers position finer than a pixel, but it cannot recover a corner the image never resolved. A 200 px icon traces to a faceted outline; a 1,000 px export traces to a clean one.
Enlarging the image first buys little
Images above 1.5 megapixels are downscaled before tracing, so enlarging a small picture in an editor first mostly hands the tracer the same soft edges with more pixels in them; the detail it lacked is still missing. What steers the result is the colour count, which decides how many regions the picture is cut into. Spend the effort there, and on finding the largest clean original you have.
A traced image is a starting point for cleanup
Expect to open the result in an editor before it goes anywhere. Typical work: deleting the background shape, merging a sliver that a compression halo produced, pulling one node to square a corner. Cut-outs stacking makes that safe, since no region overlaps another, so removing a shape leaves a clean hole rather than revealing a whole shape that stacked ordering would have hidden underneath.
Choosing the colour count
Most visitors get here because somebody else refused a file. The print shop asks for a vector because their workflow will blow the mark up to the width of a shop window and a 600 px PNG will not survive the trip. The laser or vinyl operator asks because the machine drives a head along a path and cannot follow a picture of a line. A colleague asks because they need to recolour one element without repainting the pixels around it.
The other common case is salvage: the original design file is gone, the designer moved on, and the best surviving copy is a JPEG lifted off an old invoice. Tracing will not resurrect the nodes the designer drew, but it gives you closed, editable shapes to clean up by hand, which beats redrawing the mark from nothing.
Settings we suggest
Colours (2–64) decides most of the outcome. Two give one flat silhouette and a single clean contour, what a stencil or a cut line needs. Eight to sixteen holds a flat logo's palette; twenty to thirty suits badges and stickers with shading. Near 64 on a photograph you mostly buy more small shapes and a longer conversion.
Draw style sets what gets drawn: fill_shapes for solid regions (the usual choice), stroke_shapes to outline each shape, stroke_edges for edges only, which is what pen plotters and some cutting workflows want. Shape stacking: cut-outs writes each region so it does not overlap its neighbours, safest when the artwork will be cut or recoloured; stacked lays them over one another. Grouping (none, colour, parent, layer) only affects how paths are grouped in the SVG.
The output is really a question about the next person or machine. SVG for the web, editors and most cutting software. PDF for a print shop: one page, real cubic curves, not an image pasted onto a sheet. EPS for older print and sign workflows that still ask for it. DXF for CAD and CAM: R2010 (AC1024), outlines only, LWPOLYLINE runs and degree-3 SPLINEs, one layer per colour, nothing inside them shaded in. PNG is not a vector but a clean re-render of the finished shapes. A width and height in millimetres (1–5000) reaches all four vector formats: the SVG and the DXF are written at that scale, and the PDF and EPS page box is set to the same physical size, so a 150 mm badge opens as a 150 mm page rather than as something to rescale at the other end. The DXF also states millimetres in its own header, so CAD and CAM software opens it at size instead of asking which unit to assume. Leave the field empty and the SVG falls back to 96 px = 25.4 mm, with the PDF and EPS page taking its size from the pixel dimensions.
Questions
Is a vector always better than the PNG I already have?
Not always. Flat marks, type and line art gain from it. A photograph that will only ever appear at one size on a screen gains little, and the traced copy looks flatter than the original. The test: if a person would draw it with a pen, it traces well.
Can I edit the shapes afterwards?
That is the point of the exercise. Each region arrives as a closed path with its own nodes, so you can pull a curve straight, drop a shape you do not want, or change one colour without touching the rest. A DXF opens in AutoCAD 2010 and later, Fusion 360, LightBurn, Inkscape or FreeCAD.
Is the PNG output a vector?
No. It is redrawn from the finished vector shapes at 3× supersampling, on a transparent background at 1:1 scale: a tidied-up raster, useful when what you wanted was a cleaner PNG rather than editable outlines.
Does a photograph vectorise well?
It converts, but judge the preview by eye: a portrait becomes dozens of flat colour regions, sometimes a striking poster, rarely a faithful copy. Simple logos usually finish quickly; detailed drawings and photos can take several minutes. Each conversion has a 300-second limit; if it is not going to finish in time, the converter tells you and suggests a smaller image.
What can I put in, and what happens to big files?
One image per conversion: PNG, JPEG, WEBP, BMP, GIF or AVIF, up to 25 MB and 40 megapixels. Images above 1.5 megapixels are downscaled before tracing, so the output's pixel size is that smaller size; use the millimetre field when the physical measurement matters. Transparent areas stay empty in the vector; semi-transparent pixels are composited over white.
What does downloading need, and how long are files kept?
The preview is open to everyone and carries no watermark; a free Google sign-in releases the files. Anonymous visitors get 10 conversions per hour, signed-in accounts 60 per hour and 300 per day. Download links expire after 1 hour, and uploaded images and generated files are removed by an automatic 1-day deletion rule.
Related pages
- Convert raster images into real vector geometry
- Turn a PNG into the vector file your print shop asked for
- Turn a JPG into a real vector file
- Turn the image you already have into an SVG
- Convert an image to a DXF your CAD program will open
- Vectorising an image, from upload to a result you can trust
- An image vectorizer that stamps nothing on the result
- Turn a bitmap into real vector paths
- A free alternative to Vectorizer.AI, trade-offs included