How Do You Add Captions to a Video Automatically With AI?
Add word-timed captions to a video: upload up to 3 minutes, fix the transcript, pick one of seven styles and render a captioned MP4.
You add captions automatically by uploading a talking clip to VdoBloom AI Captions, letting it transcribe every word with its timing, correcting any mis-heard words, choosing a caption style and rendering. The result is a captioned MP4 with the words burned into the picture and lit up as they are spoken. Uploads can be up to 3 minutes and 300 MB, and the job costs 2 credits per source minute with a 4-credit minimum.
Updated October 2026.
What AI Captions does
The tool has one job: put readable, word-timed text on a video you already have. It listens to the speech, writes a transcript, and stores a start and end time for each word. Those timings drive the on-screen highlight, so the current word changes colour or pops at the moment it is said rather than in a fixed rhythm.
That is different from a subtitle file. A subtitle file is a separate document the player loads alongside the video, and the viewer can turn it off. Burned-in captions are part of the picture, so they appear the same way in every feed, in every player, and in downloaded copies. For short-form social video, burned-in is usually what you want, because a large share of viewers watch with sound off.
AI Captions handles a single speaker clip best: it transcribes the upload, lets you correct the text, then renders.
Upload and pick the spoken language
Choose a video file from your device. The upload is checked before any charge: a clip longer than 180 seconds or larger than 300 MB is refused immediately.
Next, set the spoken language. The default is Auto-detect, which suits a single-language clip with clear audio. If you know the language, choosing it directly can improve the transcript on short or noisy recordings. The picker covers English, Spanish, Portuguese, French, German, Italian, Indonesian, Turkish, Hindi, Bengali, Tamil, Telugu, Marathi, Malayalam, Arabic, Russian and Japanese, alongside auto-detect. Non-Latin scripts are drawn with fallback fonts on the server, and a script the renderer cannot draw is refused before you are charged.
Record close to the speaker, avoid music under the voice, and trim long silent intros before uploading. The tool does not remove background music, so a clip with a loud track underneath the speech will produce weaker timings.
Fix any mis-heard word before rendering
The transcript arrives as editable lines. Read it before you touch the style. Every name, brand, product and number is a candidate for a wrong transcription, and those are exactly the words a viewer will notice. Correct them in the text editor; the render uses your edited words, not the raw transcription.
Lines break on pauses, sentence endings or roughly seven words, which keeps each caption short enough to read at speed. If a line is too long for the frame, split it. If two short lines belong together, join them. The timing stays anchored to the spoken audio, so editing the text changes what appears, not when the highlight moves.
Two things are worth checking in every clip: the first line and the last line. The first line is the one that decides whether a muted viewer keeps watching, and the last line is the one that carries your call to action. Fix both, then render.
Seven styles and which one suits TikTok, Reels and Shorts
The style picker has seven presets. They differ in font, weight, capitalisation, words per group and highlight colour. Each one is a fixed look; there is no font picker and no colour picker, which keeps a batch of clips consistent.
| Style | What it looks like | Good for |
|---|---|---|
| Hormozi pill | Three words or fewer, heavy caps, a coloured rounded block behind each word | Fast talking-head clips and direct sales points |
| MrBeast bold | One or two words, very large, tilted, bright active colour | High-energy Shorts and challenge or reaction clips |
| Word pop | One word at a time, large, with a strong pop on the spoken word | Punchy hooks where every word should land |
| Karaoke highlight | A full line where the spoken word lights up in turn | Podcast cuts and slower explanations |
| Emoji accents | Bold text with emoji marks on key words | Light, entertainment-led clips |
| Keyword colour | Mostly white with selected keywords in a second colour | Tutorials where one term matters |
| Clean minimal | Smaller, lighter text with a subtle outline | Professional or educational content |
For TikTok, Reels and Shorts the same rule applies: choose the largest style that still fits the safe area, and keep the caption block clear of the platform interface. Hormozi pill and MrBeast bold are the loudest and read best on a small phone screen. Word pop works when the script is short and deliberate. Karaoke highlight is the calmer option for a longer spoken segment. Clean minimal is the safest choice when the video is a tutorial and the text is supporting the picture rather than driving it.
Keep the shape or reframe to 9:16
The layout picker starts with Keep shape, which renders in the aspect ratio of your upload. If the clip is already vertical, that is the right choice. If it is horizontal, the 9:16 options reframe it: Smart picks between follow-face, blur fill and a two-speaker stack, Follow face crops to whoever is talking, Blur fill puts the whole frame on a blurred background, and Two speakers stacks two people top and bottom. Centre crop appears in the shared layout list but not in this picker.
Caption placement is handled by the style, not by a drag control. The presets sit in the lower part of a portrait frame and higher on a landscape frame, which generally keeps them clear of the platform’s own controls. Check the result against the platform’s current safe-zone guidance before posting; those measurements change, so use the platform’s page rather than a fixed number from an older article. For more on framing, see AI caption and subtitle tools and the Submagic alternatives round-up, and how to make a YouTube Short with AI.
Cost, renders included and refunds
AI Captions is billed by source length: 2 credits per source minute, with a minimum charge of 4 credits. A 2-minute clip is 4 credits; a 3-minute clip is 6 credits. The price is shown before the job starts.
One charge includes three renders for the same upload. That is the point of the render allowance: fix a word, try a second style, or switch from Keep shape to 9:16 without paying again. The render button shows how many are left. A failed render is refunded, so a technical failure does not consume your allowance.
If the clip is longer than three minutes, use Clip Studio, which accepts uploads up to 90 minutes and cuts them into short vertical clips with captions already burned in. It is the right tool for a podcast source; AI Captions is for a clip that is already short. Credit packs are one-time purchases that never expire, and the options are on the pricing page.
What it does not do
- It does not export SRT or VTT. The output is a captioned MP4 with the text burned into the picture.
- It does not translate captions. The transcript is in the language that was spoken.
- It does not remove filler words. If the speaker says “um”, that word is transcribed and, if you leave it in, captioned. You can delete it from the transcript, but the audio still contains it.
- It does not style per word. There is no font or colour picker; the seven presets are the set of looks.
- It does not accept a link. You upload a file from your device.
Want the bold, word-by-word look from big creator channels? The Hormozi and MrBeast style captions guide maps each look to a preset and the settings that get you there.
Frequently asked questions
How long can the video be?
Up to 3 minutes (180 seconds) and 300 MB. Longer sources belong in Clip Studio, which accepts up to 90 minutes.
Can I download the captions as an SRT file?
No. AI Captions produces a captioned MP4 with the words burned in. There is no SRT or VTT export.
Can I edit a word after the transcript appears?
Yes. Every line is editable before you render, and the render uses your corrected text. Check names, brands and numbers, which are the most common errors.
How many renders do I get?
Three renders are included in the one charge for that upload. The button shows how many are left, and a failed render is refunded.
Does it work with Hindi or Arabic audio?
Yes. The language picker includes Hindi, Bengali, Tamil, Telugu, Marathi, Malayalam, Arabic, Russian and Japanese, and non-Latin scripts are drawn with fallback fonts. A script the renderer cannot draw is refused before charging.
Open AI Captions to caption your first clip: upload a file under three minutes, check the transcript, pick a style and render. New accounts start with 10 free credits and no card required, which covers a 2-minute clip with a render to spare, so you can see how a style reads on your own footage before you commit to a longer batch.
Ready to try it?
Create your first AI video in minutes — no credit card required.
Start Creating Free →