Captions Generator

Your video never leaves this tab

This is the part worth understanding, because it decides everything else about the tool. There is no server in the path. The file is opened by the page, your browser decodes its audio, and a speech recognition model runs in a background thread on your own hardware. The only thing that crosses the network is the model itself, coming down to you the first time you use the page.

You can check that rather than take our word for it. Open your browser network tab, drop a clip in, and watch: the model arrives, and nothing goes back the other way. Once it has arrived, the tool keeps working with your connection switched off.

Two things follow from it. Nothing is metered, because serving you costs almost nothing, so there is no reason to cap your minutes or hold the clean export behind a payment. And footage you would rather not hand to a company you have never met stays on the machine it is already on: an unreleased edit, a client project under embargo, a medical or legal recording, or simply your own face.

How a word gets its own moment on screen

A speech model returns text, not timings. Getting the animation to land on the right syllable takes a per-word start and end, and that is a separate thing the model has to be asked for. Whisper produces it by aligning the attention it paid to the audio against the audio itself, which places each word to within a fraction of a second. That list of words and times is the whole input to everything you see afterwards.

Those words are then grouped into the pages you actually read. A page ends when it is full, when the speaker stops, or when the sentence does, and it stays on screen until the next one begins so captions never blink off mid-thought. Set how many words share a page and the grouping is recomputed instantly, because the timings never change.

The preview and the export are the same renderer, called twice. What you approve on screen is what gets encoded, frame by frame, into the MP4. That is why there is no render queue to wait in and no surprise when the file lands.

Which file to take into which editor

If you are posting straight from here, take the MP4: the captions are already in the picture and no platform can strip them. If you are cutting the video somewhere else first, take a subtitle file, and which one depends on where it is going.

• SRT is the one that works everywhere. CapCut, Premiere Pro, DaVinci Resolve, YouTube, Final Cut Pro, VLC. Text and timings, nothing else.
• ASSis the only one that keeps the per-word timing, in karaoke tags, so the word-by-word animation can be rebuilt rather than flattened into lines. Anything built on libass reads it: VLC, mpv, and ffmpeg's subtitles filter. The editors are stricter, and none of the four below lists it.
• iTTis Apple's own format and the right answer for Final Cut Pro, which accepts SRT, iTT and CEA-608 and nothing else. It carries positioning and styling as well as the words.
• VTT is the web standard, for a subtitle track on your own site or a YouTube upload.
• Text is the transcript on its own, which is what you want for a description, a blog post or show notes.

Getting better captions on the first pass

• Name the language instead of leaving detection to guess. It is one dropdown and it removes a whole class of error, especially on a clip that opens with music before anyone speaks.
• Read the transcript before you export. Names, brands and technical words are what a speech model gets wrong, and fixing one is a click. The timing it worked out is kept.
• Fewer words per page for fast delivery, more for a calm voice. Three or four suits most short-form talking; one word at a time is the aggressive look, and it needs speech to match.
• Keep captions clear of the interface. Every app puts its own buttons and text over the bottom of a vertical video, so drag the position up until the preview looks safe. You can check the exact covered area for each platform in the post previewer.
• Caption on a computer if the clip is long. Everything is held in memory while it works, so length is bounded by your device rather than by any limit we set.

Where the captions are going

Which file to take depends on what is at the other end, and the editors disagree with each other about what they will accept. These pages say what each one takes and why.

• Captions for CapCut
• Captions for Final Cut Pro
• Captions for Premiere Pro
• Captions for DaVinci Resolve
• Captions for TikTok
• Captions for Instagram Reels
• Captions for YouTube Shorts