WHISPER CAPTIONS

your audio, subtitled — before the coffee gets cold

Local AI captions for After Effects.
Transcribes the selected layer with faster-whisper — no cloud, no account, no fees — and builds a clean, timed caption layer in one click. Speaks Hebrew like a native (עברית ✓).

Free & open source · Windows · After Effects 2022+

Install in 3 steps

One installer does everything — registry, panel, and its own Python. No admin rights.

1

Download & extract

Hit the download button and extract the ZIP anywhere you like.

2

Double-click install.bat

Registers the panel and sets up a self-contained Python environment. Got an NVIDIA GPU? It offers CUDA acceleration too.

3

Restart After Effects

Open it from Window > Extensions > Whisper Captions. The first transcription downloads your chosen model (once).

💡

No Python on your machine? Run winget install astral-sh.uv once, then install.bat again — it handles the rest.

How it works

🎙️

Select & transcribe

Select the audio/video layer in your comp, pick a model & language (or auto-detect) and hit Transcribe. Everything runs on your machine.

→
✂️

Smart splitting

Set max characters per line and 1–3 lines. Long sentences split at natural boundaries, and a live preview shows exactly how captions will sit in your comp.

→
✨

One keyframed layer

A single "Captions" text layer with Source Text keyframes — bottom-centered, drop-shadowed, and trivially easy to restyle.

Why you'll like it

🧠

100% local AI

Powered by faster-whisper on your machine. No cloud, no account, no per-minute fees — your audio never leaves your PC.

🇮🇱

Hebrew done right

Ships with the ivrit.ai Hebrew-tuned Whisper model in the picker, sets true right-to-left paragraph direction on the layer, and even previews in Hebrew.

✂️

Smart caption splitting

Chars-per-line + max-lines controls with a live comp preview. Splits at sentences and clauses first, so captions read naturally.

🎛️

One layer, clean timeline

All captions live on a single text layer as Source Text keyframes — restyle once, applies everywhere.

⚡

GPU accelerated

NVIDIA GPU? Transcription runs on CUDA, several times faster than realtime. No GPU? Falls back to CPU automatically.

🌍

99 languages

Whisper's full multilingual lineup with auto-detect — from English promos to Hebrew voice-overs to Japanese interviews.

Requirements

  • ✓ Windows 10 / 11
  • ✓ After Effects 2022 or newer — scripted RTL direction needs AE 2024+; for Hebrew also enable Preferences → Type → South Asian and Middle Eastern text engine
  • ✓ Python or uv — the installer sets up its own environment
  • ✓ 150 MB – 3 GB disk per model — downloaded once, on first use