Caption styles that hold retention

When the caption is the reason a muted scroller stays, pick a timed family. When the caption is only how they read the speech, pick a static one. Retention is the job. Brand match is a different page.

Caption Studio ships 45 CSS presets in nine families: Classic, Paper, Ink, Halo, Hype, Signature, Headline, Snap, Reveal. Seven of those families land the whole line at once (kind: subtitle). Two of them are word-timed (kind: timed): Snap pops the current token; Reveal types the line in. That split is a retention instrument, not a taste swatch. A skincare brand can still need Snap on a 12-second claim. A prank channel can still need Paper on a 90-second story. Fit-your-brand is which family looks like you. This page is which family holds the watch.

The bill is a second axis, and it is not the nine families. The live transcribe-and-burn engine, add captions to a video, runs 21 basic looks at 1× credits and 9 dynamic looks at 2×. Preview a few seconds before you commit a 2× render to a clip whose retention job is "be readable." Style that does not hold is a double charge for a strobe.

When the caption is the retention

Ask one question before you open the catalog: if the sound is off and the face is ordinary, do the words on screen have to perform?

Yes, when the cut is a list, a punchline, a price, a "wait for it," or a claim the mouth is about to finish. The viewer is reading, not watching a person. The caption is the A-roll. Timed families exist for that. Snap and Reveal give the eye a moving target so the next word is a reason not to swipe.

No, when the cut is a face, a product in hand, or a demo the viewer already wants. The caption is a transcript of speech they would hear if they unmuted. A block that arrives and sits is the correct clock. Karaoke on that file is a second performance competing with the first. Retention software will not name the fight. You will see a mid-clip drop that looks like "the hook died" when the actual event was the type jumping on the mouth.

The Verizon Media / Publicis survey of 5,616 US adults (2019) found 80% of consumers said they were more likely to watch a video to the end when captions were available, and 50% said captions matter because they watch with the sound off. That is an argument for words on screen. It is not an argument for word-pop on every talking-head. XR Extreme Reach's 2026 five-country study put always-or-often caption use around half of US viewers. Availability is the lift. Drawing is the retention craft on top.

Silent autoplay still needs burned-in captions. The family is which drawing you burn.

Snap and Reveal: two timed clocks

Snap is Anton, bottom, effect: pop. The current word scales and takes an accent (pink on the base, then gold, cyan, lime, or mono). Use it when each token is a beat: a fast ad read, a joke that lands on one word, a price. Against a slow voiceover it looks frantic. There is no top variant. If the mouth sits in the lower third, the punch hits the mouth. Reframe or pick a static family.

Reveal is TikTok Sans, bottom, effect: typewriter. Words appear and stay; the line builds. Use it when the sentence is a construction: a three-item list, a recipe step, a "here is the rule." The hold is the point. Reveal on a one-word punchline is a delay. Snap on a twelve-word rule is a strobe.

Both need word-highlight captions: a timestamp per token, which means actual speech. A silent product plate has nothing to time. An authored hook you typed is not karaoke; it is an overlay. Running Snap on a mute B-roll hoping the CTA "feels native" will transcribe nothing and perform nothing.

Timed families hold early. They fatigue. A 12-second UGC ad can live on Snap for the whole file. A 3-minute Short that stays on Reveal for every sentence becomes a lyric video the viewer did not ask for. If the caption is the retention only in the open, time the first two lines in Snap or Reveal and let the rest of the cut fall back to Classic. That is two renders, not a moral failure. One preset across a three-minute file is how a good hook becomes a strobe at 0:40.

Classic, Paper, Ink: hold without performing

Classic is TikTok Sans, white fill, black stroke, bottom (or Classic Top). It is the everyday social caption. The stroke is the retention feature: the line survives a busy plate without a card eating the face. Use Classic when the person is the hold and the words are how a mute viewer keeps up. Classic Yellow is a highlighter, not a second personality.

Paper is Montserrat, black on a white card. Maximum readability, maximum occlusion. Use Paper when the plate is already noisy — street, shop, screen recording — and missing a word would lose the claim. Do not use Paper on a tight talking-head; the card is a second chin.

Ink is Inter, white on a black card. Same job as Paper with the luminance flipped. Dark plates (night, product on black, cinema-ish B-roll) want Ink. Light kitchens want Paper. Sheer variants trade a little contrast for a little less cover. If the claim is a number, do not sheer it.

These three hold because they do not ask the eye to chase. Reading speed becomes the limiter, not the effect. A Classic line that flashes for 400 ms still fails; the family did not fail, the cue did. Split the clause. Hold the last word a beat after the mouth stops. Retention here is "the sentence was finishable," not "the word popped."

Halo, Hype, Signature, Headline: when they help and when they steal

Halo is Poppins Bold with a heavy ring. It is Classic with more mass, for plates Classic's 1px stroke cannot save: foliage, sparkle, white-on-white kitchens. Halo is still a subtitle family. It does not perform. It survives. If Snap is too much and Classic disappears, Halo is the middle.

Hype is Oswald, condensed caps, coloured stroke. Energy is the point. Hype holds on a challenge or a drop. It steals on a founder talking about a refund policy. Caps plus a red ring will not make a weak open hold; they will make a serious sentence look like a trailer.

Signature is Pacifico, often middle. Cursive at phone size is a personality, not a transcript. A five-word toast can hold. A paragraph cannot. Middle placement also sits on the face. Signature is a closing card or a name, not a retention engine for speech.

Headline is Bebas Neue, tall caps, default top. It is a title row. If you use it as the speech caption, you have turned every sentence into a poster. That can hold a cold open for two seconds and then exhaust the viewer. Pair Headline as the authored hook (or pick the overlay tool) and put the transcript in Classic.

The catalog lives at caption styles. Pick the family for the job the first three seconds have to do, then stop decorating. The AI caption generator burns the look into the file. Preview on this clip, not on the grey chip in the picker. A Snap Gold that held on last week's black hoodie will vanish on this week's white podiums.

If the caption is the retention, Snap or Reveal. If the caption is the transcript, Classic, Paper, or Ink. Everything else is a special case you can name in a sentence. If you cannot name it, you are styling.

Hold the next cut in Versely.

FAQ

Is Snap always better for retention than Classic?

No. Snap holds when the words are the show. Classic holds when the person is the show. Karaoke on a face is a second performance. Measure the swipe, not the preset name.

Do timed families cost more?

On the Caption Studio CSS catalog, Snap and Reveal are looks, not a second bill. On the VEED transcribe-and-burn path, dynamic presets bill at 2× the basic rate. Preview before you render a 2× look on a clip that only needed Classic.

Can I mix Snap on the hook and Classic on the rest?

Yes, as two burns or as a hook overlay plus a Classic transcript. One timed preset for a three-minute file is how the open holds and the middle strobes.