RipGuide
How do I add captions to a video automatically?
Updated
In short: run the video through speech-to-text that gives every word a start and end time, read the result once with the sound off, fix the names and slang it got wrong, then burn the captions into the picture for short vertical video, or export an SRT file where viewers can switch captions on and off. YouTube's own help page says automatic captions can be wrong because of mispronunciations, accents, dialects and background noise, so the review pass is part of the job (YouTube help, read 18 October 2026).
How automatic captions work
A speech-recognition model listens to the audio and writes out the words. Good tools also return a time for the start and end of each word. That timing is what lets a caption appear exactly when the word is spoken, and it is what makes word-by-word highlighting possible (see word-by-word captions). The words are then grouped into short lines and drawn over the video, or saved in a caption file.
Everything after that is layout. The hard part is the first step, because a wrong word on screen is worse than no caption.
Why they come out wrong, and how to fix it
YouTube lists background noise, accents, dialects, overlapping speakers and mixed languages as causes of errors in its automatic captions (YouTube help). Stream audio hits all of them at once.
| What goes wrong | Why | What helps |
|---|---|---|
| Music or game audio under the voice | The model hears the mix and may transcribe lyrics, or miss quiet words. | Record the mic on its own track, or turn the game and music down before you transcribe. |
| Two people talking at once | Most engines output one stream of words, so overlap gets merged or dropped. | Fix those lines by hand. Separate mic tracks per person make it easier. |
| Slang, gamer terms, usernames, names | The model has not seen the word, so it picks a common one that sounds similar. | Search the transcript for the words you know are in it and correct them first. |
| Switching languages mid-sentence | The model commits to one language per stretch. | Set the language yourself if the tool lets you, and check every switch. |
| Crowd noise, a loud room, a bad mic | Speech is buried under noise the model cannot separate. | Clean the audio first (see removing background noise). |
Whatever the tool, read the transcript before you post. Do it with the sound off, the way many viewers will watch.
Burned-in captions or a caption file?
A burned-in caption (also called an open caption) is drawn into the video pixels. A caption file, such as SRT or WebVTT, travels next to the video, and the player draws the text. YouTube accepts many caption file formats, SRT among them, and says SRT and SBV need only basic timing; WebVTT can position text but its styling is limited to bold, italic and underline (YouTube help, read 18 October 2026).
| Burned-in | Caption file (SRT, WebVTT) | |
|---|---|---|
| Where the text lives | In the picture. | In a separate file the player reads. |
| Viewer can switch it off | No. | Yes, if the player offers captions. |
| Style and animation | Anything you can design: fonts, colours, word highlights. | Basic. The player decides how it looks. |
| Same on every network | Yes. | Only where the platform takes the file. |
| Fix a typo later | Re-export the video. | Edit the file and re-upload. |
| Translations | One language per export. | One file per language. |
| Right for | TikTok, Reels, Shorts and any clip you upload as a plain video. | Longer YouTube videos, courses, and anything that needs captions the viewer controls. |
For short vertical clips, burned-in is the usual choice because the style is part of the clip. If you also post the long version on YouTube, upload an SRT there as well. The two do not conflict. For the glossary version of this, see burned-in vs closed captions.
What TikTok, YouTube and Instagram offer themselves
| Platform | What the official page says |
|---|---|
| TikTok | Creators tap the auto captions icon while making a video and TikTok transcribes the speech. Creators can edit the text after it is generated, and viewers can turn captions off from the share panel. This comes from TikTok's newsroom post of 30 September 2021 (TikTok newsroom, read 18 October 2026), so the menus may have moved since. Check the app. |
| YouTube | Automatic captions are made for uploaded videos in dozens of languages. They may not be ready right after upload, and you edit or remove wrong lines in YouTube Studio under Subtitles. Live-stream captions are English only and are not kept after the stream ends (YouTube help, read 18 October 2026). |
| Reels have caption options in the app. Instagram changes its menus often, so check the Reels editor for what your account shows today. |
These are fine when you are posting one video to one app and you will fix the lines anyway. They are tied to that app: the style is the app's, and the text does not carry over to the next platform.
Making captions easy to read
- Few words per line. Short lines, one or two at a time, so the eye does not hunt across the screen.
- Large, heavy type with an outline or a box. It has to hold up over bright gameplay and over a face.
- Keep them off the buttons. TikTok's ad guidance says not to put text or key messages where buttons, usernames and captions may appear, and that the safe zone is more restrictive in the feed (TikTok ads help, read 18 October 2026). That page is for ads. TikTok doesn't publish a safe-zone size for ordinary posts in the pages I read, so keep captions in the middle of the frame and preview before you post.
- One highlight colour. Colour the word being said; leave the rest plain.
Accessibility
The W3C describes captions as a text version of the speech and the non-speech audio needed to understand the content, synchronised with the audio. Its guidance says automatic captions do not meet accessibility needs unless someone has confirmed they are fully accurate; they are a starting point that usually needs significant editing. It gives an example where automatic captions turned "4 to 5 minutes" into "45 minutes" and dropped a "not" (W3C WAI, read 18 October 2026). The web accessibility guidelines (WCAG 2.2, success criterion 1.2.2, level A) require captions for prerecorded video with sound, and describe captions as including who is speaking and meaningful sounds, not only dialogue (W3C Understanding 1.2.2, read 18 October 2026).
For stream clips this means: a caption file or burned-in text you have proofread, and a note for sounds that matter to the joke, such as a crowd laughing off-screen.
Steps
- Start from the cleanest audio you have. Use the mic track if you have one, or turn music and game audio down first.
- Transcribe with word timing. Pick a tool that returns a time for every word, not only for each line.
- Proofread with the sound off. Fix names, slang, numbers and any line where two people overlapped.
- Style for the small screen. Few words per line, big type, strong contrast, clear of the buttons.
- Choose the output. Burn captions in for short vertical video; also export an SRT or WebVTT file for platforms that take one.
- Watch the export once. Check timing and that nothing sits under the app's buttons.
How to do it in RipClip
RipClip transcribes with speech-to-text that keeps word timing, and every clip comes out with captions timed word by word. You can edit any word in the transcript, pick one of 373 one-tap caption templates, choose from 19 caption animations and more than 140 fonts, and save the look as a preset. Caption translation is a Premium feature. It works on videos from a Kick or Twitch link, a cloud-drive share link, a direct video link or an upload; it does not take YouTube links, so upload the file instead.
RipClip chose its speech-to-text engine with a blind listening test on real stream audio in October 2026: 20 hard 25-second stretches (a stadium crowd, rap videos, slots, German commentary, Spanish and English switching), each transcribed by three engines, with the RipClip owner choosing the most accurate transcript without knowing which engine wrote it. ElevenLabs Scribe v2 won 15 of the 20; two versions of Whisper won the other five. It is one listener and 20 stretches, so read it as how RipClip decided, not as a ranking of the engines for every kind of audio. RipClip keeps Whisper as a fallback if the main engine is unavailable.
If you only need captions on a single video you will post straight from your phone, the app's own caption button is quicker. RipClip is for when you want the same look on many clips from a long stream.
Related: keeping your facecam in a vertical clip, editing stream highlights faster.