Convert WOFF to WOFF2

A file with this extension rarely turns up alone. It sits in a `fonts` directory beside an `.eot`, an `.svg`, a `.ttf` and sometimes a second `.woff` for the italic, with a stylesheet nearby containing a five-entry `src` list and a comment about Internet Explorer. That folder was generated by a web font tool at a point when serving four containers was the only way to reach every browser.

None of that is true any more, and the useful frame for this conversion is not "make this file smaller" but "reduce this folder to one file per weight". The rewrap is the easy part. The parts worth spending five minutes on are the rule around it, the character set the kit was built with, and whether you are licensed to be doing any of this — each of which has a section below.

Both containers hold the same sfnt tables that a TTF or an OTF holds: the glyph outlines, the character map, the horizontal metrics, the kerning and the OpenType feature definitions. WOFF wraps them and deflates each with zlib. WOFF2 first transforms the tables into a representation designed to compress well — the glyph and loca tables in particular are restructured — and then applies Brotli to the result.

Two compressors and one extra preprocessing pass is where the difference comes from, and for an ordinary text face it lands at roughly thirty per cent. The typeface is untouched: identical outlines, identical spacing, identical features. A page that swaps one for the other renders the same and downloads less, which is as clean a change as web performance work ever offers.

It is tempting to reach for the desktop original on principle, and here it makes no difference. WOFF is a lossless container, so unwrapping it recovers exactly the tables that went in, and re-wrapping those as WOFF2 produces the same result you would get from the TTF in the same folder. There is no generational loss of the kind image formats suffer and no setting that can be got wrong.

That is worth stating because the advice is the opposite for other conversions in this family. Going from a WOFF to an installable TTF or OTF, provenance matters a great deal — the web copy is usually subset and renamed, and the desktop original is a materially better file. Going to WOFF2 you are staying inside the same character set either way, so use whichever file you have.

A generated rule from that era lists five sources: EOT with a query-string hack, WOFF2 if it was generated at all, WOFF, TTF, and SVG with a fragment identifier. Two of those are for browsers nobody supports. The EOT entry exists for Internet Explorer 8 and the SVG entry for a version of Safari on iOS that predates the format everything now uses.

The registry records WOFF2 as reaching every current browser, which is what makes the reduction safe: one `src: url("brand.woff2") format("woff2")` line is the modern rule. Look at your own analytics before deleting the WOFF entry rather than after, because "every current browser" and "every browser that visits this particular site" are not the same statement — a corporate intranet is the case that breaks it. Everything below the WOFF can go without checking anything.

This is the part of the page nobody comes for and it is the part most likely to matter. Commercial web font licences routinely forbid modifying, converting or re-encoding the files supplied, and a conversion is exactly that even when nothing about the typeface changes. Some foundries supply WOFF2 themselves and expect you to ask for it rather than make it.

A font that came with an inherited project has a licence somewhere — in the purchase records, the brand handover, the repository history, or a text file the generator dropped into the same folder. Find it first. If the font came out of a downloaded template, treat it as unlicensed until proven otherwise: a template author’s right to include a font says nothing about yours to serve it. Open licences are the easy case; the SIL Open Font License permits conversion, subsetting and redistribution provided the font is not sold on its own and reserved names are not reused.

Almost all of the weight saving in a web font comes from removing glyphs, so kits were routinely cut down to basic Latin and a handful of punctuation. That was reasonable then and it is where the gaps in a modern page come from: a euro sign, a proper em dash, curly quotation marks, an ellipsis, or a name containing ł, ğ or ș.

Re-wrapping restores none of it. If the site shows a blank box or an obviously mismatched fallback glyph in a heading, the character was never in the file, and the fix is to re-cut from the desktop original with a subset that covers what the site actually sets. That is a different job from this one, and it is the job worth doing if the kit is more than a few years old.

A kit of that vintage has one file per weight and style: regular, italic, medium, semibold, bold, bold italic. Six downloads, six `@font-face` blocks, and a fixed set of weights that a designer will eventually want to sit between.

If the typeface has since been released as a variable font, one WOFF2 can replace the lot with a continuous weight axis, and the total is usually smaller than the static set it replaces. That is a decision about the typeface rather than about the container, and it belongs in the same session as this conversion because it changes what you convert. Check whether the foundry ships a variable cut before you spend an afternoon re-wrapping six static files you are about to delete.

Every font carries a name table, and kit generators freely rewrote it — a file whose typeface is called something quite specific is frequently named `brandsans-webfont` inside. Because the `@font-face` rule declares whatever family name you like, the mismatch usually goes unnoticed for years.

It stops being harmless in two situations: when the internal name collides with a font installed on the visitor’s machine, and when two files prepared at different times both claim one family name and the browser has to pick. If type is switching unpredictably between weights on some machines and not others, opening the fonts and reading their internal names is the fastest route to the cause. The conversion carries the name through unchanged, so whatever the WOFF said, the WOFF2 says.

Three things are worth setting at the same time as the file changes, because they are the rest of the download cost. A long cache lifetime with a hashed filename, since a font never changes in place. A `preload` link for the one or two faces above the fold, which removes the round trip spent discovering the font after the stylesheet parses. And `unicode-range` if you have separate subsets, so a page that sets no Cyrillic never fetches the Cyrillic file.

Fonts are also one of the few asset types where compression at the server is counterproductive: WOFF2 is already Brotli-compressed internally, and gzipping it again adds work for no saving. Most servers exclude font MIME types by default; if yours does not, excluding them is a one-line change.

This conversion runs on our server rather than in your browser, because fontTools has no browser build. The file travels over an encrypted connection, is converted in an isolated environment, and the working directory is deleted when the job finishes. That is a genuine difference from most conversions here and it is stated plainly rather than buried.

The free tier allows 25 MB per file, which is far more than any web font, and a batch comes back as a single ZIP. A batch is usually what this job actually is — the kit has six files in it and the rule is being rewritten once, so converting them one at a time is a waste of an afternoon.