[VaitaR]

What

brandFontFaces (captions.mjs in faceless-explainer, product-launch-video, pr-to-video) now declares each per-script subset file as its own @font-face with the matching unicode-range.

Why

Google Fonts / Fontsource ship a face as per-script files (source-serif-4-cyrillic.woff2, source-serif-4-latin.woff2). The dedup key was family+weight+style, so only the alphabetically first file survived. In a Russian explainer the captions got only the Cyrillic subset, and every Latin word and digit fell back to a system font.

Related work

None found.

How

The subset is read from the filename (latin, latin-ext, cyrillic, cyrillic-ext, greek, greek-ext, vietnamese) and added to the dedup key. Ranges are copied from the Google Fonts CSS. If the family name itself contains the subset word, the file is not treated as a subset.

Test plan

• Unit tests added: build-frame-subsets.test.mjs covers all three workflows. The new tests fail on main (3 fail) and pass with the fix; node --test on that file plus the three captions.test.mjs: 116 pass.
• Manual testing performed. The bug came from a real faceless-explainer project, but the fix is covered by the unit tests only.
• Comments follow CONTRIBUTING.md "Comments"

[jrusso1020]

Thanks for tracking this down. The diagnosis was right for the code it targeted, but that code no longer exists on main.

#5651 (688229314f, "publish seven knowledge skills aligned with Studio and retire the workflow skills") deleted the faceless-explainer, pr-to-video and product-launch-video skills, including all three captions.mjs copies and build-frame-subsets.test.mjs. brandFontFaces isn't anywhere in the tree any more, so there's nothing left for this fix to apply to. I'd close this one. If the same per-script subset dedup shows up in the new skills or in the CLI font handling, a fresh PR against that code would be welcome.

Need: no, the target was removed by #5651. Size: ok. Reuse: n/a. Simpler: n/a.

— Rames