How To Vectorize A Blurry Or Low-Resolution Logo
We traced 12 logos at seven sizes and scored each against its original vector. How small is too small, and why enlarging first makes it worse.
Measured in this post
| What | Before | After |
|---|---|---|
| Mean error vs the original vector, 1024 px sourcecorpus/harness/blog/measure-live.mjs lowres — 12 hand-authored reference logos rendered at each size, traced on the live site, scored against the original vector at 512 px. | — | 0.12% |
| Same, 256 px sourceSame run. | — | 0.22% |
| Same, 96 px sourceSame run. | — | 0.75% |
| 96 px source enlarged to 1024 px before tracingBicubic resampling to 1024 px, then traced; worse than tracing the raw 96 px file for 9 of 12 logos. | 0.75% | 1.29% |
| SVG size, same comparisonMedian SVG size produced by the live site. | 3.1 KB | 42.4 KB |
The only copy of the logo is a 120-pixel PNG from an old email signature, and the sign shop needs a vector by Friday. It is the most common reason anyone goes looking for a tracer, and it comes with a lot of confident advice — most of it never tested.
So we tested it. This post reports how much a logo's source size actually costs you, and one piece of standard advice that turned out to make things worse.
How we measured it
To grade a trace you need to know what the right answer was. So instead of starting from blurry logos of unknown origin, we started from twelve hand-authored vector logos, where the correct shapes are known exactly.
We rendered each one to a PNG at seven sizes, from 48 to 1024 pixels on the long edge, and traced every file through our live site the way a visitor would. Then we scored each trace against the original vector it came from, rendered at the same size: the average colour error per pixel, and the share of pixels that are visibly different.
Twelve logos times eight inputs is 96 traces. The script and the raw results are in our repository.
What a smaller source costs you
| Source size | Mean error | Pixels visibly off | Paths (median) |
|---|---|---|---|
| 1024 px | 0.12% | 1.0% | 5 |
| 512 px | 0.14% | 1.1% | 5 |
| 256 px | 0.22% | 1.3% | 5 |
| 128 px | 0.48% | 2.0% | 6 |
| 96 px | 0.75% | 2.6% | 6 |
| 64 px | 1.23% | 3.4% | 8 |
| 48 px | 1.08% | 3.5% | 7 |
Three things stand out.
Down to 256 pixels, you lose very little. At 512 px the trace is almost indistinguishable from the full-size result, and at 256 px the error is still only 0.22%. If your logo is at least that big, the source is not your problem.
Below 128 pixels, it falls away quickly. By 96 px the error is six times the full-size figure. Thin strokes and small lettering are the first things to go, because at that size a stroke is only one or two pixels wide, and there is no edge left for any tracer to find.
Below 64 pixels, the numbers stop improving in any orderly way. The 48 px results were slightly better than the 64 px ones. At that size each logo is really a handful of coloured squares, and how well it traces depends on where its edges happen to fall on the pixel grid more than on its size.
The advice that made it worse
The most common tip for a small logo is to enlarge it first: open it in an image editor, scale it up to 1000 pixels or so, then trace the bigger file. It sounds reasonable. A tracer has more pixels to work with, so it should do better.
We tested it. Each logo at 96 px, enlarged to 1024 px with standard bicubic resampling, then traced:
| Traced as-is at 96 px | Enlarged to 1024 px first | |
|---|---|---|
| Mean error | 0.75% | 1.29% |
| Pixels visibly off | 2.6% | 7.2% |
| Paths (median) | 6 | 14 |
| SVG size (median) | 3.1 KB | 42.4 KB |
Enlarging first made the trace worse for 9 of the 12 logos. On the median logo it more than doubled the paths and made the file more than ten times bigger.
The reason is that enlarging does not add detail. It invents a soft blur between every pair of neighbouring pixels. A crisp two-pixel edge becomes a smooth ramp twenty pixels wide, and the tracer then has to decide where in that ramp the edge really is, and whether the in-between shades are colours of their own. It gets that wrong in both directions: edges come back rounder, and the in-between shades come back as extra slivers of colour.
A good tracer already reads a small image more carefully than a blurred copy of it. Ours treats each edge pixel as partial coverage of the shape on either side of it, which recovers a smooth edge from the pixel staircase without inventing anything. Give it the original pixels.
What to do with a small or blurry logo
- Look for a bigger copy first. The website header at full size, social media profile uploads, old print PDFs, and the files the original designer sent. A 400-pixel copy is worth a lot more than any amount of cleanup on a 120-pixel one.
- Check whether a PDF or EPS is already a vector. Open it in Illustrator, Inkscape or Affinity. If the logo selects as shapes, there is nothing to trace.
- Trace the original pixels. Do not enlarge, sharpen or re-save it first. Every re-save of a JPG adds compression damage; enlarging adds blur.
- Trace in a flat-art mode and set the exact number of brand colours, so anti-aliased edge shades merge back into the colours they belong to.
- Zoom in on the lettering before you download. Small text is where a small source shows first. A few letters may need tidying by hand in a vector editor.
When to redraw instead
If the logo is under about 100 pixels and full of small lettering, the tracer is recovering shapes that are barely in the file. It will still give you a clean, scalable version of what is there, and that is often all a sign or a T-shirt needs. But for a brand identity that will be printed large, the trace is best used as an accurate guide to redraw over, not as the final artwork.
Everything above is reproducible: the twelve reference logos, the resampling settings, and the scoring are in the measurement script in our repository.
Common Questions
- What is the smallest logo that can be vectorized well?
- In our measurements, quality held up well down to about 256 pixels on the long edge and fell away below 128. Under about 100 pixels, thin strokes and small lettering are barely in the file for any tracer to recover.
- Should I upscale a small logo before tracing it?
- No. Enlarging a 96 pixel logo to 1024 pixels first made the trace worse for 9 of 12 logos we tested, with more than double the paths and a file ten times bigger. Trace the original pixels.
- Can AI upscaling fix a blurry logo?
- It can make a logo look sharper, but it guesses at detail rather than recovering it, and those guesses get traced as if they were real. Finding a larger original copy is always the better fix.
- When should I redraw a logo instead of tracing it?
- When the only copy is tiny and full of small lettering, and the logo will be printed large. The trace is then best used as an accurate guide to redraw over rather than as final artwork.