Animated captions are subtitles that move with the voice: the word being spoken lights up, grows, or appears on its own. Most short-form video has them now, and every caption tool offers a handful of styles with names like karaoke, pop and box.
The choice usually gets made on looks. It deserves a little more thought, because the styles differ in how hard they make the viewer's eyes work and in how well they cope with fast speech, long lines and imperfect timing. Editif's subtitle generator renders five of them. Below is how each one behaves on screen, where it works, and the settings that matter more than the style itself.
#What makes a caption animated
An ordinary subtitle has two timestamps for a whole line: when it appears and when it goes. An animated caption needs a start and an end for every word. That is the whole difference. Timing each word by hand is slow work, so in practice the timings come from speech recognition. Ours come from Whisper, which returns the moment each word begins and ends along with the text.
Everything after that is drawing, which means no style can rescue bad timing. If recognition places a word a quarter of a second late, all five styles will highlight it a quarter of a second late.
#The five styles
In the editor they sit in a row under the style presets: Karaoke fill, Active word, Box, Pop and One word. Presets set the font, size and colors. The animation is a separate choice, so any preset works with any of the five.
#Karaoke fill
The whole line is on screen in the base color, and the highlight color sweeps through it from left to right, filling each word over the time it takes to say. A word that takes half a second to say takes half a second to fill. It's the smoothest of the five and the closest to an actual karaoke screen, and it suits slower, continuous speech: tutorials, interviews, anything sung. It's also the style the editor opens with.
It is the most sensitive to timing, because it uses both the start and the end of every word. When word timings are estimates, the sweep visibly runs ahead of the voice or lags behind it.
#Active word
The line is on screen and the word being spoken switches to the highlight color all at once. The words before and after it stay in the base color. This is the look most people mean when they say word-by-word captions. Because it only reacts to the moment each word starts, it hides small timing errors that the karaoke sweep would expose. If you're unsure which style to use, start here.
One detail of how we draw it matters more than you'd expect: the highlight stays on a word until the next word begins. Speakers pause mid-sentence all the time, and a highlight that switches off in every gap makes the caption flicker.
#Box
The text stays one color and a rounded box in the highlight color sits behind the current word, jumping from word to word. A filled box is a much bigger change than a color swap, so it stays visible over busy footage: gameplay, street scenes, anything with bright color behind the text. The price is weight. A box moving across a long line draws attention to itself, so it works best with large type and few words per line.
#Pop
The current word changes color and grows by 16% over about an eighth of a second, easing out so that it settles and doesn't snap. The rest of the line holds still. It adds energy to fast, punchy delivery and gets tiring in a long explanation where every single word bounces.
Scaling a word is trickier than it sounds. Enlarging text spreads its letters apart, so if you draw the bigger word on top of the original, the original shows through the gaps. Our renderer takes the word out of the line and draws the scaled copy in its place.
#One word
One word on screen at a time, large, popping in from three-quarters size as it's spoken. Nothing else competes with it, which is why it gets used for hooks and punchlines. It can also be set much bigger than a full line, since it only ever has to fit a single word.
The cost is that the viewer loses the sentence. They can't read ahead, can't glance back, and have to read at exactly the speaker's pace. Look away for a second and those words are gone. That's fine for a fifteen-second clip and hard work for ninety seconds of explanation, and it's the least forgiving style for anyone reading in a second language.
| Style | What moves | Works well for | Think twice when |
|---|---|---|---|
| Karaoke fill | Color sweeps through each word | Slower speech, music, longer lines | Word timings are estimated |
| Active word | The current word changes color | Almost everything | You want more punch |
| Box | A box jumps behind the current word | Busy or bright footage | Lines are long or type is small |
| Pop | The current word changes color and grows | Fast, energetic delivery | The video is long or calm |
| One word | One large word at a time | Hooks, punchlines, very short clips | Viewers need to follow an argument |
#Three timing rules that matter more than the style
The animation gets the attention, but these three rules decide whether a caption is comfortable to read. They apply to every style.
- Lines appear early and leave late. Each line comes on 0.2 seconds before its first word and stays for 0.3 seconds after its last. The head start gives your eyes time to find the line before the highlight starts moving.
- A pause starts a new line. If the speaker stops for more than a second, the caption breaks there, so a line never hangs on screen through a silence with half its words still unlit.
- Lines are capped at seven words and at the text width you set, whichever is reached first, with at most two lines on screen.
On a landscape video the seven-word cap is usually what ends a line. On vertical video the width runs out much sooner. We ran a sample of ordinary speech through the same line-breaking code using the Hormozi preset (Montserrat bold, size 72, text width 85%) on a 1080 by 1920 frame. Lines averaged two words and never ran past four. So at those sizes on a phone-shaped video, you are effectively choosing a two-word caption whichever of the first four styles you pick, and One word is less of a jump from the others than it looks in the menu.
#Size and placement on vertical video
Font size in Editif is set against a frame 1080 pixels tall and scaled to your video. Size 72 is 72 pixels on a 1080p landscape video and 128 pixels on a 1080 by 1920 vertical one. The useful consequence is that a preset looks the same on a 720p export as it does on 4K.
The margin works the same way, and this is where vertical video trips people up. Margin is the distance from the bottom of the frame to the caption, also measured on a 1080-tall frame, so dividing it by 1080 gives the fraction of the frame height. The default of 60 puts captions about 6% up from the bottom edge. That is right for a landscape YouTube video. On Shorts, Reels and TikTok that part of the frame is covered by the app's own interface: the description, the account name, the audio label, and a column of buttons down the right-hand side.
For vertical video, raise the margin to 220 or more, which lifts the captions about a fifth of the way up the frame, or set the position to center. The safe area differs between apps and moves when they redesign, so upload one clip as a private or draft post and look at it on a phone before you render a batch.
If your source is landscape, crop it to 9:16 before adding captions, not after. Line breaks are calculated for the frame you upload, and cropping afterward can cut words off at the sides.
Two other settings are worth a moment. Uppercase is a toggle next to the styles. Capitals read fine in two-word bursts and get tiring in long lines, so it suits vertical clips better than landscape ones. And on footage where a color highlight gets lost, switch on the background strip or use the Box style before reaching for a thicker outline.
#Where the words come from
There are three ways to get text into the generator. They differ in how the word timings are produced, and that sets a ceiling on how good the animation can look.
| You provide | Text comes from | Word timing comes from | Credits per minute |
|---|---|---|---|
| Only the video (Auto-generate) | Speech recognition | Recognition, word by word | 15 |
| The video and a plain .txt script | Your file, word for word | Recognition, matched to your words | 10 |
| The video and an .srt or .vtt file | Your file | Your cue times, split evenly between the words | 10 |
Auto-generate is the quickest and uses the larger of our two recognition models. Its weakness is that there's no review step between recognition and rendering, and the captions are burned into the picture. If the model mishears a name, a product or a technical term, the mistake is in the video.
So for anything with names or jargon, do it in two passes. Get a transcript first (our transcription tool is free), correct it, save it as plain text, and upload that with the video. Your wording is kept exactly and each word is timed against the audio. Uploading your own text also gives you per-word styling, where you can pick out individual words in a different color, weight or size.
Uploading an .srt is the route to avoid for animated styles if you can. A subtitle file only has a start and an end for each cue, so that time is divided equally among the words in it. With Active word or Box that's usually close enough. With Karaoke fill, the sweep moves at a constant speed while the speaker doesn't. Sound descriptions in brackets, such as [music] or (applause), are dropped before rendering. And if the file's timing is off, the captions will be too. We wrote up how to diagnose and fix out-of-sync subtitles separately.
#When not to burn captions in
All of this produces open captions: text that is part of the picture. That suits short-form platforms, where the feed autoplays and the styling is part of the appeal. It has costs elsewhere. Viewers can't switch burned-in text off, resize it or have it translated, and a typo can only be fixed by rendering again.
For long-form video on YouTube, or anything on your own site, upload a caption file as well as, or instead of, burning the text in. A caption track can be turned off, restyled by the viewer and auto-translated. We have a walkthrough for adding an SRT file to a YouTube video, and if you need to pick a file format, SRT, VTT and TXT are compared here.
#Limits, cost and what happens to your video
Most of Editif's tools run in your browser and never upload anything. This one is different, and you should know that before you use it: recognition and frame-by-frame rendering happen on our servers, so the video is uploaded. Files are deleted automatically after three days.
Uploads can be up to 200 MB in MP4, MOV, MKV, AVI or WebM, and the result is an MP4. The picture is re-encoded at about the source bitrate and the audio track is copied across untouched. If your file is over the limit, compressing it first is usually enough.
The generator runs on credits. A new account gets 100 free ones, which at the time of writing covers six minutes of auto-generated captions or ten minutes with your own text. They expire after 30 days. Current prices are on the pricing page.
#A sensible starting point
If you want one configuration to begin from for a vertical talking-head clip: Active word, a bold sans-serif preset at size 66 to 72, text width 85%, margin 220, two lines. Check the first render on a phone, at arm's length, with the sound off. If you can follow it like that, the style is doing its job.