Upload a video or audio file and get timed captions in minutes. The Whisper speech model runs on your own device, so nothing is uploaded. Pick from 24 animated styles, including the popular one or two words at a time in the centre, with your own colours and fonts. Edit any line, preview the captions over your video, then export SRT, VTT or plain text, or burn the styled captions into a new video.
How to use
- Add your video or audio. Drop an MP4, MOV, WebM, MP3 or WAV file into the tool. Files up to 10 minutes work best.
- Choose a language and model. Leave language on auto-detect or pick yours. Tiny is fastest, Base is more accurate, Small is the most accurate and needs a decent device.
- Generate captions. The first run downloads the model once and your browser keeps it for next time. Then the audio is transcribed with timestamps.
- Edit, style and export. Fix any words, pick a caption style, then download SRT, VTT or TXT, or burn the styled captions into the video.
Features
Runs on your device
The Whisper model executes in your browser using WebGPU when available, with a WebAssembly fallback. Your video and audio are never uploaded.
Three accuracy levels
Pick Tiny for speed, Base for balance, or Small for the best accuracy on clear speech.
Multilingual
Transcribes many languages and can auto-detect the spoken language. Hindi, Spanish, Arabic, French and more are supported by the model.
Editable captions
Every caption is a line you can edit before exporting, with a live preview over your video.
SRT, VTT and TXT
Export the formats that YouTube, Premiere, CapCut, DaVinci Resolve and most editors accept.
24 animated caption styles
One word or two words at a time in the centre, karaoke highlights, colour boxes, glow and gradient designs, with your own colours, fonts and animation.
Burn-in option
Create a new video with the styled captions drawn onto it, exactly as in the preview.
What people use it for
- Captions for Reels, Shorts and TikTok clips that people watch with the sound off
- A searchable transcript of a lecture, meeting or interview recording
- Subtitle files (SRT or VTT) to upload to YouTube, Vimeo or a course platform
- Show notes and quotes pulled from a podcast video
- Accessibility captions for training and onboarding videos
- Reading along while learning a language from native speech
Why captions matter for short video
A large share of people watch social video with the sound off, on public transport, in bed, or at work. Captions let your message land without audio, and they make your video usable for people who are deaf or hard of hearing. They also help viewers follow fast speech, accents and technical terms.
Captions also give search and recommendation systems more text to understand your video. Uploading an SRT file to platforms that support it, such as YouTube, gives you accurate subtitles without relying on the platform's own guess.
How the on-device AI works
Whisper is an open speech recognition model released by OpenAI. This tool uses a compact version converted to run in the browser through Transformers.js. Your video's audio is extracted, converted to 16 kHz mono, and fed to the model in 30-second windows. The model outputs text with start and end times, which become your caption lines.
The first time you run a model size, your browser downloads it once, roughly 40 MB for Tiny and more for the larger sizes, and stores it for later visits. After that, transcription works even with a slow connection.
Getting the most accurate captions
Clear audio matters more than anything. Record close to the microphone, reduce background music, and avoid people talking over each other. If the model mishears a name or a technical word, fix it in the editor before exporting. Choosing the language yourself, rather than relying on auto-detect, also helps when the audio is short or mixed.
Animated caption styles: one word at a time
The popular short-video look shows one or two words at a time in the centre of the frame, often with a different colour on the word being spoken. This tool has 24 designs in that spirit, from bold single-word pops to karaoke lines, colour boxes, glow, gradients and a plain classic subtitle. Every design can be adjusted: words at a time, position, size, animation, font and three colours.
The designs are drawn on a canvas, so the preview over your video is exactly what gets burned into the exported file. Fonts are bundled with the site and the Poppins font includes Devanagari, so Hindi captions show correctly.
SRT, VTT and TXT: which export do you need?
SRT is the most widely accepted subtitle format. Almost every editor and video platform can import it, so choose SRT when you are unsure. VTT is the web version, used by HTML5 players and some course platforms. TXT is just the words without timing, which suits blog posts, show notes and transcripts.
If you plan to post on a platform that cannot take a subtitle file, such as Instagram or TikTok, use the burn-in option instead. That draws the captions into the picture so they always show.
What leaves your device and what does not
Your video and its audio never leave your device. The tool extracts the sound locally, converts it, and runs the speech model inside your browser tab. What is downloaded is the model itself, which comes from a content delivery network the first time you use a size and is then stored by your browser.
That is why the first run takes longer. After that, captioning works even on a slow connection, because only the model cache is needed.
Making captions that people actually read
Short captions of a few words keep pace with fast video. Break lines where a speaker would naturally pause, keep each caption on screen long enough to read, and avoid covering faces. The Short setting splits sentences into roughly 28 characters per line, which suits vertical video. Use Standard for horizontal video such as tutorials.
Troubleshooting
The words appear slightly early or late
Word timing comes from the speech model and can drift on fast speech, music or overlapping voices. Try the Base or Small accuracy, and check the preview. If a line looks wrong, edit it.
My Hindi or other captions look different from the style preview
The bundled fonts cover Latin and Devanagari. Other scripts use a font from your device, so the look can differ.
The first run seems stuck on downloading the model
The model is a one-time download of tens of megabytes or more, depending on the size you picked. Keep the tab open, try the Tiny model on a slow connection, and try again if your network dropped.
Captions came out in the wrong language
Auto-detect can guess wrong on short or noisy clips. Choose the spoken language yourself before generating.
Names and technical terms are misspelled
Speech models guess unusual words. Fix them in the editor before you export. The text is fully editable.
The tab crashed or the phone got very hot
Speech recognition is heavy work. Use the Tiny model, close other tabs, or caption a shorter clip. Desktop browsers with WebGPU are fastest.
Limits and good to know
- Best for videos up to about 10 minutes. Longer files can run out of memory on phones.
- Accuracy depends on audio quality, accents and background noise. Always review the text before publishing.
- Burning captions into a video records the video in real time, so it takes about as long as the clip itself.
Related searches
- animated captions generator
- word by word captions
- captions one word at a time
- auto captions generator
- video to subtitles
- add subtitles to video free
- generate srt file
- video to srt converter
- auto captions for reels
- speech to text for video
- subtitle generator online
- caption video without uploading
- free whisper captions
FAQ
Is my video uploaded anywhere?
No. Audio is extracted and transcribed inside your browser. The model files are downloaded from a content delivery network, but your video and audio never leave your device.
Is the caption generator really free?
Yes. There is no sign-up, no watermark and no limit on how many videos you caption. The site is supported by ads.
Which languages are supported?
The Whisper model supports roughly 100 languages with varying accuracy. Common languages such as English, Spanish, Hindi, French, German, Portuguese and Arabic generally work well on clear audio.
Why is the first run slow?
The first run downloads the speech model. Your browser stores it, so later runs start much faster.
What is the difference between SRT and VTT?
Both are plain-text subtitle formats with timestamps. SRT is the most widely supported by video editors and platforms. VTT is the web standard used by HTML5 video players.
Can I use the captions on YouTube, Instagram or TikTok?
YouTube accepts SRT and VTT uploads. Instagram and TikTok generally need captions burned into the video or added by their own editors, so use the burn-in option for those.
Why did my phone slow down or crash?
Speech recognition is demanding. Try the Tiny model, close other tabs, or caption a shorter clip. Desktop browsers with WebGPU are fastest.
Can I get captions that show one word at a time?
Yes. Choose a design, then set Words at a time to One word, Two words, Three words or Whole line. The preview updates immediately.
Do caption styles appear in the SRT file?
No. SRT and VTT are plain text with timing. The styles are drawn into the video when you burn the captions in.
Do I need to create an account?
No. There is no sign-up and no limit on how many videos you caption.
Can I caption an audio file instead of a video?
Yes. MP3, WAV and M4A files work the same way as video files. Burn-in is only available for video.
Can it translate captions into other languages?
The tool transcribes the language spoken. It does not translate. You can paste your own translation into each caption line before exporting.
Why are timings approximate?
The model returns start and end times for each sentence. When a long sentence is split into shorter captions, the time is shared in proportion to length, so individual lines can be slightly early or late. Check a few lines in the preview.