Word-by-word captions show a few words at a time and light up each one as it is spoken. Making them takes four things: a timestamp for every word, a cap on how many words are on screen, a placement clear of the app's own buttons, and a burn-in — because TikTok and Reels take no subtitle file that can carry a per-word highlight, and SRT and VTT cannot carry one anywhere.
A caption line can be a little early or late and nobody notices. A highlight that lands a beat off reads as broken.
An ordinary subtitle file times lines: this sentence from 00:04.6 to 00:07.2. A word-by-word caption needs the start and end of every word, and the precision of those times is the whole effect. At 30 frames a second, one frame is 33 ms. A highlight 100 ms off is three frames off — enough that the bright word visibly arrives after the voice, or jumps ahead of it.
suubbs transcribes on your own machine and gives every word its own timestamp. How close those timestamps land depends on which build does the work, and the numbers are worth knowing:
Raw speech timings are also uneven in a way that has nothing to do with accuracy. Measured on real audio, a highlight sat on one word for 405 ms, flicked across two words in 200 ms, then sat for 515 ms. That is faithful to the speaker and tiring to watch. The Highlight smoothness setting evens the cadence out within each caption, without letting any word drop below 130 ms, so nothing flashes.
If the highlight looks consistently late while you edit, check your headphones before you touch the timing: Bluetooth audio commonly arrives 150–300 ms after the picture. suubbs has a highlight-lead control for exactly this, and deliberately leaves it out of the export — it describes your headphones, not the video.
Most of the "style" of word-by-word captions comes from two numbers, not from the font.
Words on screen. This is a hard cap on how many words are visible at once, from 1 to 12 or "as many as fit". One word at a time is the punchiest and the hardest to read for long; two to four keeps a phrase together, which is easier to follow when the sound is off. When you change it, the lines re-split around the cap — and any line you edited or locked by hand keeps its own text and timing.
Line layout. The widescreen layout allows 42 characters over two lines, the standard for horizontal video. The vertical layout allows 22 characters over three lines, because a 9:16 frame is narrow and large type at 22 characters already crosses most of it.
Follow the spoken word. Four modes for what happens to the current word:
The presets are starting points for the rest. Pop is heavy upper-case Inter Tight with a dark outline, the look people usually mean by "TikTok captions". Highlight puts dark text on a lime box. Boxed, the default, is white text on a near-black panel that reads on any footage. Any look you build can be saved as your own preset.
One thing gives way when the highlight is on: fades. A highlighted caption is written as one event per word, and fading each of those in and out would flicker rather than read as a soft entrance, so fades only apply with the highlight off.
TikTok, Reels and Shorts all draw their own interface over the bottom and right of the frame. A caption placed for YouTube widescreen ends up underneath it.
| Covered area, % of a 9:16 frame | TikTok | Reels | Shorts |
|---|---|---|---|
| Name and caption text, along the bottom | bottom 22%, left 76% of the width | bottom 24%, left 78% | bottom 20%, left 78% |
| Button rail, up the right edge | right 22%, from 40% to 85% down | right 20%, from 44% to 84% down | right 20%, from 42% to 86% down |
| Top of the frame | search, top-right 9% | top bar, top 10% | top bar, top-right 8% |
Those zones were measured off screenshots of each app on a phone and drawn a little generously, since a caption placed slightly further clear costs nothing. They move when the apps update, so treat them as a guide rather than a guarantee. Put together, they say three things:
The studio can overlay each platform's zones on the preview while you place the caption, and dragging snaps to the centre, the thirds and the safe margins, so a caption lands on the same line in every video rather than at 49.6% in one and 51% in the next. A dragged position is exported as an exact coordinate. The zones in pixels, and the one area clear on all three apps, are on caption safe zones.
Which is why word-by-word captions are almost always burned into the video.
suubbs exports SRT, WebVTT, ASS and a word-level JSON. SRT and WebVTT time whole lines. They have no way to say "this word, now, in yellow", so exporting either drops the highlight — suubbs tells you so at export rather than letting you find out after posting. Of the formats suubbs writes, Advanced SubStation (.ass) is the one a video tool can render with the highlight (the JSON keeps per-word timings as data, but no player draws it). suubbs writes each word interval as its own line with the spoken word recoloured, boxed or grown. ASS's own karaoke tags were not used on purpose: they colour words as already sung, filling left to right, rather than picking out the current word, so they cannot reproduce this look.
Short-form platforms do not take an .ass upload, so the last step is drawing it into the picture. One ffmpeg command does it on your Mac, keeping the font, highlight and position exactly as the preview showed — the full walkthrough, including the font and quality settings that trip people up, is in how to burn subtitles into a video. One difference to expect: the Box treatment behind a whole caption does not currently come through an ffmpeg burn-in on its own; that page shows the one-line override that adds it, and the box is square-cornered, because ASS boxes have no corner radius.
The free browser version transcribes the first 60 seconds of whatever you give it, and nothing else is held back — styling, the highlight, safe-zone placement and the ASS export all work. A clip under a minute can be made start to finish there, without an account, with the looser browser timing. The Mac app is $9.99 once, has no length limit, and brings the 3 ms timing. Neither uploads your video.
What suubbs does not do is animated templates — emoji, bouncing entrances, b-roll. Tools built for that, like Submagic, render your clip on their servers instead; suubbs vs Submagic sets out which job each one is for.
Including the parts that depend on the platform rather than the file.
Two to four for most talking-head clips: enough to hold a phrase together for viewers watching with the sound off. One word at a time is punchier and harder to read over a long clip.
No. SRT and VTT time whole lines and cannot mark a single word as the one being spoken, so the highlight is lost. Export ASS and burn it into the video.
Either the word timings are loose — the browser model is within roughly 60–140 ms, the Mac app about 3 ms — or your audio is late. Bluetooth headphones commonly delay sound by 150–300 ms, which makes a correct highlight look late.
Above roughly three-quarters of the way down the frame and clear of the right fifth between about 40% and 85% down. Below and to the right of that, the apps draw their own buttons and text over the video.
No. suubbs transcribes on your own machine, in the browser trial and in the Mac app, and the burn-in with ffmpeg runs locally too.
Transcription detects sixteen languages, including Spanish, French, German, Portuguese, Japanese and Hindi, and writes the captions in the language it heard.