Caption fonts that survive a phone screen: heavy, open, never thin

Type designers who work on small text agree on the shapes that survive it, and the agreement is old and boring. A large lowercase x-height, so the letters fill the line you gave them. Open apertures, meaning the gap in a c, an e or an s stays open instead of curling shut. Roomy counters, which is the enclosed white inside an a or an o. And moderate stroke contrast, because a face with thick stems and hairline joins loses the hairlines first.

That guidance rules out two popular choices. Heavily condensed and heavily extended designs are harder to read at subtitle size, which is exactly where a lot of caption styles live. So are thin and light weights, which look elegant on a laptop in a dark room and vanish on a phone held in daylight.

One more constraint nobody writes down. A caption is read in two to four word bursts while something moves behind it. Anything that asks the eye to work costs time you do not have: a script face, a display face with quirks, letter spacing pulled tight to look expensive.

So the useful question is not which font is prettiest. It is which font still reads when it is small, bold, moving, and sitting on top of a shot that changes every two seconds.

TikTok's captions used to render in Proxima Nova, which the app offered under the name Classic. They now render in TikTok Sans, a bespoke geometric sans drawn for the interface, the video captions and the marketing, shipped as a variable font. TikTok released it under the SIL Open Font License 1.1 in July 2025, so it sits on Google Fonts and anybody can use it, as of August 2026.

Instagram Sans came out of a 2022 project with Colophon Foundry and Instagram's own brand team. It is not licensed for public download, so the honest answer for anyone chasing that look is a geometric substitute such as Poppins or DM Sans.

Netflix's Timed Text Style Guide asks for Arial as a generic placeholder for a proportional sans serif, in white, at a size relative to the video resolution that fits 42 characters across the screen.

The BBC went further and drew its own. Reith was tested with mixed-ability reader groups against Helvetica, Arial, Gill Sans and Tiresias, and the BBC's subtitle guidance tells authors to use a wide face, naming Reith Sans, Verdana and Tiresias, so lines stop wrapping where nobody wanted them to.

Broadcast offers almost no choice at all. CEA-708, the standard behind digital television captions in the United States, defines a short list of styles: monospaced and proportional serif, monospaced and proportional sans serif, casual, cursive and small caps, plus an undefined default.

Read those five together and a pattern falls out. Every organisation on that list with an accessibility obligation converged on a plain, wide, medium-to-bold sans serif. Nobody with something to lose picked a display face.

1. Inter. SIL Open Font License. A tall x-height, open apertures and a weight axis that goes properly heavy. This is the safe default when a caption has to carry a whole sentence rather than three shouted words.

2. Montserrat at Black. SIL Open Font License. Wide round counters, and set in capitals at weight 900 it is the caption look most people are copying when they run this search. The other face behind that look is The Bold Font, offered by its foundry at no cost for personal and commercial work, with a paid version that adds lowercase and extra glyphs. Verdict: the loudest legible option here, and it stops being legible the moment the line runs past four words.

3. TikTok Sans. SIL Open Font License 1.1 since July 2025, variable, and covering more than 460 languages across Latin, Greek and Cyrillic. Verdict: the face the feed itself reads in, which is a real advantage on a native-looking post and a real disadvantage if you are trying not to look like everyone else.

4. Anton. SIL Open Font License. Heavy and condensed, drawn for headlines. Verdict: three words in capitals and nothing longer. Condensed and small is the combination the legibility research keeps warning about.

5. Bebas Neue. SIL Open Font License. Capitals only, condensed, and everywhere. Verdict: a headline face doing caption work, which is fine for a hook card and wrong for a spoken line.

6. Roboto, or Roboto Flex. Apache License 2.0. Plain, even, and already installed on most Android phones. Verdict: nobody will compliment it and nobody will fail to read it.

7. Poppins or DM Sans. SIL Open Font License. Geometric, round, and the closest legal stand-ins for the Instagram look you cannot license. Verdict: buy the shape, not the association.

The licence question has a short answer. Every font hosted on Google Fonts ships under an open licence, and the overwhelming majority are under the SIL Open Font License, which permits commercial use including video. The one rule that bites is that you may not sell the font files themselves. Adobe Fonts is the other common route, and Adobe's own licensing pages state that fonts included with a Creative Cloud subscription may be used to produce film, video and broadcast content, including video distributed on YouTube. What you cannot do under either is hand the font file to a collaborator and treat that as licensing them.

Change the typeface and the caption looks different. Change the outline and the caption becomes readable. Most people get that the wrong way round.

The reason is what sits underneath. Body text has one background. A caption has whatever the footage is doing that second: a white wall, then a window, then a face. No single text colour has contrast against every frame you are going to cut to.

So the contrast has to travel with the text. White or off-white type with a dark outline of roughly three to four pixels is the standard answer and it holds on most footage. A drop shadow on its own is not the same thing. W3C guidance is explicit that a shadow does not satisfy contrast requirements by itself, and it fails first over pale or busy frames. Outline plus a light shadow covers more cases than either alone, and a semi-transparent box behind the line is the blunt instrument that always works and always looks like a subtitle.

Broadcast settled this a long time ago. CEA-708 decoders have to offer five character edge effects, raised, depressed, uniform outline, and a drop shadow on either side, and the viewer has to be able to override whatever the stream sent.

Then come the three settings that outrank the name of the font every time. Size, which usually needs to be larger than looks right on a laptop. Position, high enough to clear the caption bar and the button column on the right. And how many words are on screen at once, which is two to four for spoken content and never a full sentence.

Cutroom has no font picker. It ships six caption packs, and each one is a complete look: typeface, weight, size, colour and rhythm, chosen together rather than assembled at eleven at night. Inside a pack you set colour, weight, size and position. The six are clean-ugc, bold-impact, karaoke-flow, editorial, fast-native and minimal-lux.

The timing comes from the same transcription that drives the cut, with word-level timing taken from your audio rather than estimated from a script, so words land on the beat of your speech. Switching packs is one action and costs nothing, which is the argument for testing two and watching both on an actual phone.

The limit, stated flatly: you cannot upload a font file. If a brand contract names a specific licensed typeface and the export has to use it, an editor with a font picker is the right tool for that file, and CapCut, Kapwing and Descript all have one.

What it costs is two visible charges. Cutting one upload is 100 credits, which covers transcription, the directing pass and the b-roll search, and captioning adds nothing to that. Exporting is 20 credits per finished minute, so a 30 second video is 110 credits from upload to file. Lite is $19.99 a month for 1,250 credits, about 11 of those, nothing on the picture at all, and cancelling is one click.

TikTok, Instagram, Netflix, the BBC, Google Fonts, Adobe, CapCut, Kapwing and Descript are trademarks of their respective owners. We are not affiliated with any of them, and the rankings and verdicts here are our own.