Auto subtitles in After Effects with AI
No, After Effects has no auto captions, but with AI you can add them: ask ChatGPT, Claude or any MCP client connected to Prism to transcribe the voiceover, and timed subtitle text layers land in your open composition.
Every guide to captions in After Effects ends the same way: type them by hand, or leave for a captioning site and return with an SRT. Prism transcribes inside After Effects, every word becomes a marker, and the captions are built from those words.
What you say, and what lands
| You type | What appears in your comp |
|---|---|
| "Transcribe the selected interview layer, drop the filler words, and build one text layer per sentence at the bottom of the frame, timed to the words." | A [Transcript] guide layer with one marker per word, then one text layer per sentence with in and out points on the spoken words |
| "Now split those captions per word, karaoke style." | The same words as one timed layer each, rebuilt from the existing markers, no second transcription |
The captions are ordinary text layers: re-time, re-word, restyle or split them by hand.
Test it on your own project. A Free account is enough to start: Sign up, then connect your AI and ask for this.
How to do it
[Transcript] guide layer appears above the source with a marker per word. Your source layer is untouched.Languages and formats
| Languages | detected automatically, or set with a language code |
| Audio and video | MP3, WAV, M4A, FLAC, OGG, MP4, MOV, WEBM |
| Caption formats | one running line, one sentence per layer, or one word per layer |
Options when you transcribe
| Ask for | What changes |
|---|---|
| Separate speakers | each speaker's words are labelled, so an interview captions by voice |
| Remove fillers | "um" and "uh" are dropped before the captions are built |
| Mark sounds | laughter, applause and music are noted as events |
| Character timing | every letter carries a time, for write-ons |
What it costs
Transcription is a Prism AI capability, paid from your balance per minute of audio. Restyling the captions uses MCP actions, so a restyle never re-transcribes. Audio is sent only to produce the transcript, then discarded.
What it needs
| After Effects | 2024, 2025 or 2026, macOS or Windows |
| AI client | any MCP client: ChatGPT, Claude, Claude Code, Cursor, Codex and others |
| Plan | Pro with a Prism AI balance for the transcription; the caption layers use MCP actions |
What it does not do
It does not import or export an SRT file, and it does not work offline. There is no caption preset library: the look is what you describe.
Common questions
No, After Effects has no speech recognition of its own. With Prism connected to an AI client, the transcription runs from inside After Effects and the captions are built as text layers.
Select the layer and ask your AI client to transcribe it, or run Transcribe from the Prism AI tab. A guide layer with one marker per word appears above the source.
No. The timing comes from the transcript's word timestamps, not from a subtitle file, so there is nothing to export, upload or import.
Yes. Each caption is a native text layer with its own in and out point. Change the font, move it, split it or undo it as usual.
Yes. Ask for the speakers to be separated and each one's words are labelled, so the captions can be styled or positioned per voice.
Related
Number counter
Make a number counter in After Effects with AI: say the start, end and format, and get a text layer driven by slider keyframes with commas, percent or currency.
Word by word captions
Word by word subtitles in After Effects with AI: the Hormozi, TikTok and Shorts style, with the spoken word highlighted, built as text layers you can restyle.