For TikTok, Reels and Shorts
Animated captions that light up word by word, generated in your browser.
Drop in a video, the AI transcribes it, you fix any mistakes, you export in 1080×1920. Your video is never uploaded to a server: everything happens on your computer.
- No upload
- No sign-up
- Pro: one-time payment, no subscription
The tool
1 Pick a video
2 Fix the text and pick a style
Click a timestamp to jump there; edit the text directly (timings are kept).
3 Export
Free version: a small "Captioned with Repères" watermark appears at the bottom right. Pro removes it.
Export runs in real time: a 60 s video takes about 60 s. Keep this tab visible while exporting.
Pricing
Free
€0
- Unlimited transcription, in your browser
- 3 animated styles, adjustable colors and size
- 1080×1920 video export
- .srt and .ass export
- Small, discreet watermark on the video
Pro
€19 one-time payment
- Everything in Free
- No watermark
- All future updates of the tool
- No subscription, no account
Secure checkout by Gumroad, which emails you your license key. The product description on Gumroad is in French.
Got a license key?
How it works
Local transcription
The Whisper speech recognition model (OpenAI, MIT license) runs in your browser with WebGPU, or WebAssembly as a fallback. It is downloaded once from Hugging Face, then cached.
Local rendering
Every frame is redrawn with your captions and recorded by your browser. The original audio is kept.
Nothing is uploaded
The video, the audio and the text never leave your device. The server only receives anonymous counters (a visit, an export) and, if you have one, your license key so it can be checked.
Limits, honestly
- A recent desktop browser is required. Tested on desktop Chrome and Edge (WebGPU). Firefox and Safari have not been fully tested yet; without WebGPU the tool falls back to WebAssembly, which is slower. On phones it may fail for lack of memory.
- First run. The model is a one-time download of ~90 MB (Fast) or ~250 MB (Accurate).
- Processing time depends on your hardware. WebGPU uses your graphics card; the WebAssembly fallback runs on the CPU and is noticeably slower. The tool shows how long the transcription took. Export takes as long as the video.
- The AI sometimes gets it wrong (names, numbers, acronyms). Review and fix the text before exporting.
- Output format. MP4 on recent Chrome/Edge and Safari, WebM on Firefox. WebM imports into CapCut or converts easily; TikTok and YouTube also accept .srt files.
- Short videos. Built for short-form content (up to 3 to 5 minutes). Beyond that, the browser may run out of memory.
- 8 spoken languages can be selected: English, French, Spanish, German, Italian, Portuguese, Dutch and Arabic. No translation: captions are in the spoken language.
FAQ
Does my video really stay on my computer?
Yes. The file is read by your browser; no request sends it anywhere. You can check it in the "Network" tab of the developer tools: only the model (from Hugging Face), the library (from jsDelivr) and our anonymous counters go over the network.
What does Pro include?
It removes the watermark from exported videos, for €19 paid once (any applicable tax is calculated by Gumroad at checkout). The key is checked with Gumroad, then remembered in this browser.
I switched computer or browser.
Enter your license key (emailed by Gumroad) again on the new device.
Can I use it for commercial videos?
Yes, the videos you make are yours.
Why are some pages in French?
Repères is a small French project. The tool and these guides are in English; the legal notice and terms of sale are only available in French for now.
It doesn't work for me.
Try an up-to-date Chrome or Edge on a computer, and the "Fast" model. If it still fails, email us (address in the legal notice) with your browser and operating system.