How to use an image to font converter
An image to font converter does one job well and everything else badly. It takes shapes that already exist in a picture and rebuilds them as vector outlines inside a font file. It does not invent letters you never drew. Knowing where that line falls is most of what separates a clean conversion from a frustrating one.
What the converter is actually doing
Conversion runs in four passes. First the image is separated into ink and background, which is a thresholding problem — the tool estimates a cut-off between dark and light and applies it across the picture, correcting for uneven lighting where it can. Second, the remaining ink is grouped into connected regions, so a stroke that touches itself becomes one shape rather than several. Third, those regions are grouped into characters and laid out in reading order. Fourth, each character is traced into Bézier outlines and written into an OpenType font.
Every problem you are likely to hit belongs to one of those four passes, which is why the fix is usually a different photograph rather than a different setting. If pass one cannot find your ink, no amount of smoothing later will recover it.
Preparing a source image that traces cleanly
The tracer reads contrast, not intent. A picture that looks perfectly legible to you can still be ambiguous to a threshold, usually because the paper is unevenly lit or the ink is thin and grey rather than solid.
- Use flat, even light. Window light on an overcast day beats a desk lamp, which throws a bright patch on one side and a shadow on the other.
- Fill the frame with the letters. Detail you did not capture cannot be traced back in.
- Shoot square to the page. Perspective skew turns vertical stems into slanted ones and confuses the row detection.
- Prefer solid black ink on white paper. A fine-liner, brush pen or marker all work; pencil is the hardest case because the strokes are grey and grainy.
- Leave clear space between characters. Letters that touch are read as one shape, and separating them again is guesswork.
- Avoid ruled or gridded paper. Printed lines are ink as far as the threshold is concerned.
The two input modes
A picture containing several distinct letterforms is treated as a specimen sheet: the characters are separated, put in reading order, and mapped onto the code points you assign. A picture containing a single piece of artwork is treated as exactly that — one glyph, mapped into the Private Use Area at U+E001, where it will not collide with a real character.
The mode is chosen from the picture, so if a sheet of letters comes back as one icon, the characters were probably touching or the background was not fully removed.
Which controls to reach for, and when
The adjustment panel exposes five controls. They are worth understanding individually, because turning all of them up is not the same as turning the quality up.
Detail
Controls how small a speck survives the trace and how closely the curve follows the pixel edge. Raise it when fine serifs, hairlines or brush-tips are being rounded away. Lower it when dust, paper texture or JPEG noise is producing tiny stray shapes.
Smoothing
Controls how aggressively corners are rounded into curves. This is the control most likely to change the character of your type, so it is measured from the image by default rather than fixed at a value. A geometric, hard-cornered face needs almost none; a brush script needs a lot. If your square letters come back with soft corners, smoothing is too high.
Background removal
Moves the ink-versus-paper threshold. Raise it when parts of the paper are being kept as ink; lower it when thin strokes are dropping out. Watch the preview rather than the number.
Spacing
Sets how much side bearing each glyph gets. It does not change the outlines, only how close the letters sit when you type with them.
Baseline
Sets where the baseline falls relative to the traced shapes, which is what decides whether descenders hang below the line properly instead of sitting on it.
Checking the font before you install it
A file that opens is not the same as a font that is right. Two checks catch most problems in under a minute.
First, look at the character map and confirm every shape is mapped to the character you meant. A misassigned glyph is invisible until you type with it. Second, type a pangram in the preview and read it at text size, not display size — spacing errors and baseline drift are obvious in a line of running text and nearly invisible in a single large letter.
If the segmentation is uncertain, the review step opens instead of reporting a clean result. That is deliberate. It is better to be asked than to install a font with two letters silently merged.
What you get out
The export contains the installable OTF, a full-colour SVG of the traced artwork, one SVG per glyph, a font.css with an @font-face rule, a mapping.json recording which shape went to which code point, a specimen.html you can open in a browser, and a README with install instructions for macOS, Windows and Linux.
Questions
- Can it add letters that are not in my image?
- No. The converter only produces glyphs for shapes it finds in the picture. If you photograph A to M, you get A to M. Anything else would be an invention rather than a conversion.
- What image formats can I use?
- PNG, JPG and WebP. PNG is best because it has no compression artefacts around the edges of strokes.
- Does my image get uploaded anywhere?
- No. Tracing and font building run in your browser, and the source image is never sent to a server.
