Auto subtitles and captions in After Effects

Flexile runs whisper.cpp, an open-source (MIT) C/C++ implementation of the Whisper speech model, as a separate process on your computer. There is no Python and no separate installer, and you need no API key or speech service account. Your audio is never uploaded.

The speech engine and the models download from Flexile’s servers only when you click Download, and update only when you click Update. The panel’s automatic updates never install them. Downloads resume after an interruption and are checked against their published size and SHA-256 hash.

Our guide to local Whisper subtitles explains the engine and the models in more depth.

Five models, from 32 MB to 574 MB

• Tiny, 32 MB: fastest, rough
• Base, 60 MB: fast
• Small, 190 MB: good for speech, and the one offered first
• Medium, 539 MB: better with accents
• Large v3 Turbo, 574 MB: best for songs

The first download adds the speech engine (about 3.5 MB). After that, the largest model you have installed is the default. Settings > Speech models lists your models with their size and lets you delete them.

Flexile transcribes what it hears, speech or singing. To catch Whisper’s known mistakes on music, Flexile drops words timed after the end of the audio and cuts a phrase of one to three words repeated more than four times down to four.

When you already have the words, switch on I have the lyrics or script, then paste the text or load a text file. Flexile only times your words: they stay exactly as written, and each line becomes one subtitle.

Flexile matches your words to the words it hears and gives any word it misses a time between its neighbours. On our demo song, Large v3 Turbo aligned 51 of the 53 lyric words.

32 languages plus auto-detect

Auto-detect listens to the first 30 seconds. For songs, pick the language yourself. The picker lists English, Polish, German, Spanish, French, Italian, Portuguese, Dutch, Ukrainian, Russian, Czech, Slovak, Swedish, Danish, Norwegian, Finnish, Turkish, Greek, Hungarian, Romanian, Bulgarian, Croatian, Serbian, Japanese, Korean, Chinese, Arabic, Hebrew, Hindi, Indonesian, Vietnamese and Thai.

Select a text layer before you create the subtitles, and every subtitle copies its font, size, colour, position, effects and Flexile IN/OUT animation. Flexile hides that base layer and keeps it as the template. A layer with keyframed Source Text can’t be the base.

No base layer? Pick one of four presets: Clean, Outline (the default), Bold yellow or Lower third. Flexile adds it as a hidden Subtitle style layer in the comp.

Karaoke highlight and word colours

Switch on Highlight the word being said and each word takes the highlight colour while it is spoken or sung, yellow (#ffd60a) by default. Flexile builds it from a native text animator with a Fill Color property and an Expression selector based on words.

Word colours give every occurrence of a word or phrase you choose its own colour. The full walkthrough is in our guide to karaoke text.

With an AI app connected to Flexile over MCP, you can ask your AI to transcribe the audio, style the subtitles and create the layers:

Add subtitles to the voice-over in the Outline style.

Your AI can’t download speech models, so click Download once in Audio > Subtitles first.