Font subsetter and inspector: shrink TTF, OTF and WOFF web fonts

Subset a web font in the browser, nothing uploaded

A full font ships thousands of glyphs while a page uses a fraction of them: font subsetting keeps only the characters you need, then compresses the result as WOFF. The serious tools are command-line programs (pyftsubset from fontTools, glyphhanger) or require uploading the file to a server, which is a problem with a licensed font.

This tool runs entirely in your browser: it reads TTF, OTF, WOFF 1.0 and TTC files, shows a full inspector (names, licence, fsType permissions, variation axes, Unicode coverage per block, languages covered, OpenType features, size of every table), then builds a TrueType subset with component and ligature closure, writes TTF and WOFF, checks that the produced font loads and generates the @font-face CSS with unicode-range.

1. Choose a font

   Drop a TTF, OTF, WOFF or TTC file, or click “Choose a font”. For a TTC collection, pick the font to extract. A WOFF2 file (wOF2 signature) is detected and refused with an explanation.

2. Read the inspector

   Check the family, version, licence (name ID 13) and fsType embedding permissions, the fvar axes, the Unicode coverage per block, the languages likely supported and the GSUB/GPOS features. The glyph map and the search box answer “does this font have ẞ or €?”.

3. Choose the characters to keep

   Paste your text into “Text to keep”, tick presets (Basic Latin, Latin-1, Greek, Cyrillic…) or type Unicode ranges. The counter shows which characters are present, which are missing and how many glyphs are kept after closure.

4. Set the output mode

   “Keep OpenType features” preserves kerning, ligatures and variable axes; “Smallest file” renumbers glyphs and drops the layout tables. Decide about hinting removal and the minimal name table.

5. Build the subset

   Click “Build the subset”: the tool shows the size before and after, the per-table detail, the rendering self-check, the “Download .ttf” and “Download .woff” buttons and the CSS snippet to copy.

How the subset is built

The tool reads the sfnt directory described by the OpenType specification, then the cmap table (formats 4 and 12 first) to map your characters to glyphs. It then computes the closure: the components of composite glyphs (an “é” that reuses “e” and the accent), then, when features are kept, the GSUB substitutions of types 1 to 4 (single, multiple, alternate and ligature forms such as fi, fl, ffi) iterated to a fixpoint, the sub-lookups of contextual substitutions (types 5, 6 and 8, type 7 extensions resolved) conservatively, the variants of the MATH table and the colour layers of the COLR table.

• “Keep features” mode: glyph IDs do not move; unused outlines are emptied (zero-length loca entries), the cmap is rebuilt, GSUB, GPOS, GDEF, kern and fvar stay untouched and therefore valid, and gvar is only pruned of the removed glyphs.
• “Smallest file” mode: glyphs are renumbered compactly, components are remapped, hmtx, loca, glyf, cmap, maxp, hhea and post are rebuilt, layout and variation tables are dropped.
• Hinting removal: fpgm, prep, cvt, hdmx, VDMX and LTSH go away, instructions are stripped from simple and composite glyphs, maxp and gasp are adjusted the way pyftsubset does.
• Writing: sorted directory, tables aligned on 4 bytes, table checksums and head checkSumAdjustment recomputed, short or long loca depending on size, then WOFF 1.0 with zlib per table (a table stays raw when compression does not help).
• Self-check: the produced font is loaded with the FontFace API and your preview text is rendered with the original and the subset side by side, pixels compared.

What each table weighs and what the tool does with it

When to subset a font

• Marketing site or blog: ship only Basic Latin, Latin-1 and typographic punctuation to speed up the first paint of text.
• Headline or logo set in a display font: keep only the letters of the slogan, a few kilobytes instead of hundreds.
• CJK fonts (TrueType): paste the real text of the page to go from several megabytes to a few dozen kilobytes (IPA Gothic: 6.2 MB down to about 55 KB as WOFF for a paragraph of 154 distinct characters, 22 KB in “Smallest file” mode); CJK fonts shipped as OTF/CFF are not subset.
• Audit before buying or embedding: check language coverage, OpenType features (tnum, smcp, ss01…) and fsType permissions without installing the font.
• TTF to WOFF conversion: get a valid WOFF 1.0 and the matching @font-face snippet without sending a licensed file to a server.

Common problems and fixes

• Some characters show in another font: they are not in the subset; the counter and the self-check flag the missing ones, add them to the text to keep.
• Ligatures or kerning disappeared: you chose “Smallest file”; switch back to “Keep OpenType features”.
• The WOFF is not as small as expected: in keep mode, hmtx and loca keep their original size; WOFF compresses those areas very well, and the smallest mode really shrinks them.
• WOFF2 font refused: convert it to TTF or WOFF with another tool first (the tool ships no Brotli/WOFF2 decoder).
• OTF font not subset: its outlines are CFF; convert it to TTF before subsetting, or use only the inspection and WOFF conversion.

Privacy: the font is read, analysed and rewritten by your browser; no file is uploaded or stored, which matters for a licensed font. Limits: CFF/OTF outlines are not subset, WOFF2 is neither read nor written, contextual substitutions are kept conservatively (the subset may contain a few more glyphs than strictly needed), variation sequences (cmap format 14) are dropped, bitmap and colour tables are dropped in smallest mode, files over 200 MB are refused. Check your licence before publishing a subset font.