Typography
Sixty-nine caption fonts, and how to choose one.
Every face in the catalogue, what each class of type is actually good for, and the choice that decides whether a caption gets read.
LumaCaption offers 69 caption typefaces, selectable per style and per individual word. The choice that matters most for legibility is weight and width rather than the family: a heavy condensed grotesque survives being scaled down on a phone, and a high-contrast display serif does not.
Font choice in captions gets treated as a taste decision and it is mostly a legibility decision. A caption is read at phone size, over moving footage, in under a second, often by someone holding the phone at arm’s length. Most of the typographic range that works beautifully on a poster fails all four of those conditions.
What follows is the whole list, grouped by what each class of type actually does to a caption, and the three rules that decide whether the choice works.
69 faces, per style and per word
Set the face for a caption style, or override a single word’s face without touching the rest of the line.
Devanagari-capable hands
Kalam is the one handwriting face in the catalogue that draws Devanagari properly, which is why the write-on Hindi styles are set in it.
Grotesques for legibility
Inter, Geist, Manrope, Hanken Grotesk, Archivo, Libre Franklin and Familjen Grotesk — the safe range for anything dense or fast.
Display faces for hooks
Anton, Bebas Neue, Archivo Black, Bowlby One SC, Titan One, Alfa Slab One and Climate Crisis for the first three seconds, not the next thirty.
The three rules that matter
Everything else is preference. These three decide whether a caption is read.
- Weight beats size
- A heavier face at the same size reads further away than a lighter one made bigger. If a caption is hard to read, add weight before you add size.
- Avoid high stroke contrast
- Faces with very thin hairlines — Bodoni Moda, Abril Fatface, Italiana — lose those strokes entirely at caption size over video. Beautiful on a title card, invisible on a Reel.
- Condensed buys you words
- Anton, Oswald and Bebas Neue fit noticeably more per line at the same height, which matters when a cue runs long and the alternative is shrinking it.
- One face per video, usually
- Per-word face overrides exist for emphasis, not for variety. Two faces in a caption reads as a mistake far more often than as a design.
The catalogue, by class
Sixty-nine faces sounds like too many until you notice they are doing genuinely different jobs. Roughly a third are workhorses you would use across a whole channel; the rest exist because one caption style needs exactly that voice and nothing else will do.
- Grotesque and neo-grotesque
- Inter, Inter Tight, Geist, Manrope, Hanken Grotesk, Archivo, Libre Franklin, Familjen Grotesk, Instrument Sans, Barlow, Montserrat, Poppins, Sora, Josefin Sans, Space Grotesk, Unbounded.
- Heavy display and condensed
- Anton, Archivo Black, Bebas Neue, Oswald, Bowlby One SC, Titan One, Fugaz One, Righteous, Bungee, Bungee Shade, Alfa Slab One, Graduate, Climate Crisis, Luckiest Guy, Bangers, Shrikhand.
- Serif and editorial
- Instrument Serif, Playfair Display, Bodoni Moda, Fraunces, Cormorant Garamond, EB Garamond, Libre Caslon Text, Spectral, Marcellus, Tenor Sans, Italiana, Abril Fatface, Newsreader, Source Serif 4, Shippori Mincho.
- Hands and brush
- Kalam, Caveat, Great Vibes, Dancing Script, Pacifico, Permanent Marker, Amatic SC, Architects Daughter, Rock Salt, Shadows Into Light, Chewy, Creepster.
- Mono and screen
- JetBrains Mono, Geist Mono, IBM Plex Mono, VT323, Press Start 2P, Orbitron, Audiowide, Monoton.
Fonts and Hindi
Most of the sixty-nine are Latin-only. That is normal for display typography and it matters here for one specific reason: a Devanagari caption set in a Latin-only face falls back to a system font, so the styling you chose quietly stops applying.
Kalam is the face in the catalogue built for this — it draws Devanagari properly and is a handwriting face, which is why the write-on handwritten styles are set in it and why Hindi handwriting stays handwriting rather than dropping to a default. Hinglish sidesteps the problem entirely, since it is written in Roman letters and works in every face in the list.
How it works
Pick a caption style first
Each of the 242 styles arrives with a face chosen for it. Start there rather than from a blank font list.
Swap the face if you need to
69 to choose from, applied to the style or to a single word for emphasis.
Check it at phone size
The preview is on your own footage. If a hairline disappears there, it will disappear on the feed.
Common questions
What is the best font for subtitles?
A medium-to-heavy grotesque — Inter, Geist, Manrope, Archivo or Hanken Grotesk — for anything dense or fast, because weight and even stroke width are what survive phone size over moving footage. Heavy condensed display faces like Anton and Bebas Neue are better for short hook lines, where fitting more words per line at the same height matters more than long-form comfort.
Which fonts should I avoid for captions?
Anything with very thin strokes. High-contrast display serifs like Bodoni Moda, Abril Fatface and Italiana lose their hairlines completely at caption size over video — they look excellent on a static title card and disappear on a Reel.
Can I use a different font for one word?
Yes — a single word’s face can be overridden without touching the rest of the line. Use it for emphasis rather than variety; two faces in one caption usually reads as a mistake.
Which fonts work for Hindi?
Kalam is the face in the catalogue built for Devanagari, and it is a handwriting face — which is why the write-on handwritten styles use it and why Hindi captions in those styles stay handwritten instead of falling back to a system font. Most of the other 68 are Latin-only. Hinglish avoids the issue entirely because it is written in Roman letters.
Can I upload my own font?
No — the list is fixed at 69 faces. Every one of them is loadable by the renderer, which is what guarantees that a face you pick in the editor is the face that ends up in the exported video rather than a silent fallback.
