Subtitle a Video Automatically With AI, in French

You shot a decent video, it's clear, and it still goes unnoticed because nobody turns the sound on. Subtitling by hand easily takes 30 to 45 minutes of work for 60 seconds of video: writing, splitting, timing each line to the speech. With a transcription AI, the same video comes out subtitled in 2 to 3 minutes, and you're left with just two or three words to fix.

This guide shows you two ways to do it in French: with the built-in features of your phone and social platforms, or with a specialized third-party tool. You'll also see the settings that really matter (text size, position, number of words per line) and the typical mistakes automatic French produces: badly split liaisons, homophones, proper nouns. No technical skills required: if you can drop a file into a window, you can do this. For the text part, starting with proofreading the transcript, you can get help from the Skilzy chat before going back to the subtitling tool of your choice.

How an AI actually subtitles a video

A subtitling AI does three things: it listens to the audio track, it writes the text, then it syncs each sentence with the exact moment it's spoken.

The first step is called automatic speech transcription. The model analyzes the sound and produces text. The best known in this field is Whisper, OpenAI's open source model, trained on 680,000 hours of multilingual audio, a large share of it French. It's Whisper, or a derivative, running under the hood of most of the tools you'll come across.

The second step is timestamping: the AI attaches a precise moment to each word (for example 00:00:04,120). That's what keeps subtitles from lagging half a second behind, the most irritating flaw for the eye.

The third step is styling: font, size, color, word-by-word animation. At the end, you get one of the following two formats.

• A separate subtitle file (.srt or .vtt): a plain text file you upload alongside your video on YouTube, LinkedIn or a website. Viewers can turn them on or off, and search engines read their content.
• Burned-in subtitles (also called burn-in): the text is baked into the image. You can't turn it off, but it shows up everywhere, including on TikTok, Instagram and Shorts where playback often starts muted.

For vertical formats, burning in is almost always the right call. For a long video on YouTube or a website, keep the .srt file: it also feeds the page's search ranking.

Method 1: the built-in feature on your phone and social platforms

It's free, it's already installed, and for a story or a Reel published within the minute, it's often the best effort-to-result ratio.

No reason to hide this option: in plenty of cases, it gets the job done.

On TikTok. In the editor, open the right-hand menu, then Captions. The app transcribes automatically in French and lets you fix each line before publishing. You can also generate translations. Styling stays limited to a few presets.

On Instagram. Two routes: the Captions sticker you add on top of the video in the Reels editor, or the automatic captions option in your account settings, which turns them on by default. Editing is possible but less comfortable than on TikTok.

On YouTube. After uploading, YouTube generates automatic French subtitles in YouTube Studio, under the Subtitles tab. You can edit them line by line or upload your own .srt file, which is preferable for a video you intend to keep alive. These subtitles are indexed — a real lever for an AI-powered faceless YouTube channel.

On your phone directly. On iPhone, Apple's Clips app displays live captions as you speak and burns them into the video. On Android, several manufacturers offer transcription built into the gallery editor, with availability varying by model and system version.

The limits of this route are always the same: styling is locked, you don't get a file you can reuse elsewhere, and transcription happens inside the app you're publishing in. If you're distributing the same video across four platforms, you do the job four times.

Method 2: specialized third-party tools

They make sense when you want a highly polished look, batch processing, or a clean subtitle file to reuse.

Here are the families of tools you'll come across, in no particular order of preference.

Pricing for these services changes often: check the official page before committing, especially on annual plans.

A word on running Whisper on your own machine: it's the cheapest solution to use and the most respectful of your files, since nothing leaves for a server. In exchange, you have to go through the terminal and wait longer if your computer doesn't have a dedicated graphics card. It's a good option for a large volume of sensitive content, not for subtitling a Reel between two meetings.

The settings that separate subtitles people read from subtitles people ignore

A subtitle is readable when it fits on two lines max, stays on screen long enough to be read, and doesn't hide behind the app's interface.

Four settings do most of the work.

Reading pace. Professional subtitling standards hover around 17 characters per second and 42 characters per line for a horizontal video. Beyond that, the eye gives up. AI tools often split too finely or too coarsely: take a look at your longest lines before exporting.

The number of words on screen. In a vertical format, 3 to 5 words per screen with the spoken word highlighted works better than a full sentence. On a long 16:9 video, do the opposite: complete sentences on two lines, which is easier on the viewer.

The safe zone. TikTok, Instagram and YouTube Shorts overlay buttons, a handle and a caption along the edges. Keep your subtitles between 20% and 70% of the image height, leaning toward the lower middle. A subtitle hidden behind the Subscribe button is useless.

Contrast. White text with a black outline, or text on a semi-opaque band. Avoid bright colors on a light background: it looks nice on your calibrated screen, and it's unreadable in full sun on a phone at 30% brightness.

These rules apply to any short-form content, including when you're producing Reels and TikToks with AI in series.

The typical mistakes of automatic French, and how to fix them fast

French trips models up on four specific points: homophones, liaisons, proper nouns and figures.

Proofreading 60 seconds of subtitles takes about two minutes. It's the best investment in the whole chain, because a glaring mistake displayed in large type in the middle of the screen costs more in credibility than a sloppy edit.

The checklist:

• Grammatical homophones: a/à, ces/ses/c'est/s'est, et/est, ou/où, la/là. Models rely on context and get it wrong mostly at the start of a sentence.
• Badly split liaisons: „les enfants ont attendu" can come out as „les enfants nont attendu". Watch for words that start with n' or t' for no reason.
• Proper nouns, brands and acronyms: your company name rarely gets through on the first pass. Many tools accept a custom vocabulary list: fill it in once and it serves all your videos.
• Figures and units: „vingt-neuf euros quatre-vingt-dix" is correct but unreadable as a subtitle. Rewrite it as €29.90. Same logic for percentages and dates.
• Punctuation and capitals: some models return a block with no periods or commas. Add at least the periods — they guide how lines get split.
• Anglicisms: „engagement", „feed", „prompt" are often transcribed phonetically. Correct them to the spelling your audience expects.

One time-saving trick: proofread the text before styling it. Fixing a word in a text editor takes three seconds; fixing it once the animation is applied takes thirty.

What to remember

If you publish now and then, the subtitles built into TikTok, Instagram or YouTube Studio are enough and cost nothing. If you produce several videos a week, go through a specialized tool where transcription, styling and export all live in one place: what you mainly get back is time on proofreading and cross-platform reposting. In every case, stick to the rule of two minutes of proofreading per minute of video, keep your subtitles out of interface zones, and limit yourself to two lines. The rest is comfort.

The simplest way to form your own opinion: subtitle a video you've already published, using two different methods, and compare the time spent.