Optical kerning

Every letter pair gets its space from somewhere. Usually it comes from the font: the type designer walks the pairs that need help, an open A beside an open V, a T over a round o, and writes a correction for each into a kern table the shaping engine applies whenever those two letters meet. Optical kerning throws that table away and computes the space instead, by looking at the two outlines and measuring the white trapped between them, then closing or opening the pair until every gap along a line holds roughly the same amount of white. Adobe’s applications offer it as a setting beside the font’s own values, which is where most designers meet the word.

The distinction to hold is with kerning itself. Kerning is the adjustment: the space added or removed between one particular pair, whoever decided it and however it was decided. Optical kerning is one method of deciding it, from the shapes rather than from the table, so it is a kind of kerning and not an alternative to it. Neither is tracking, which opens or closes every gap in a run by the same amount and knows nothing about which two letters it is standing between.

It earns its keep where the table is thin or absent. A face rushed to release, a free webfont, or a subset that shipped without its kern table will leave a display line visibly lumpy, and computing the space from the shapes is better than trusting values that were never written. It also handles pairs no table could cover: a glyph pulled from a fallback face has no pair value with the letter beside it, because the two were never in the same font. The cost is symmetrical. Run it over a face whose designer spaced it carefully and it overrides months of judgement with an average, flattening the deliberate looseness after a capital or the tight fit of a designed ligature pair. The rule of thumb is to reach for it at display sizes, on faces you do not trust, and to leave a good text face alone.

On the web the choice does not exist. The font-kerning property only switches the font’s own table on and off, and no browser computes spacing from outlines, so web typography is metric kerning or none at all. The specimen on this page shows all three states anyway: the unkerned pair and the metric one are what the browser really does, while the optical values were computed ahead of time by rasterizing the two letters, measuring the white between them, and solving for the offset that matches a reference pair. That is why the three bars land level only in the optical setting, which is the whole claim the method makes.