screencastgen

Generate a LipSync Reader

The lip-sync pipeline combines narration, word alignment, and a reference face video. LatentSync generates presenter clips delivered through the hosted reader or a standalone offline reader ZIP. The offline reader ZIP is the recommended portable output because it preserves the synchronized document, narration, and presenter experience.

Basic command

screencastgen lipsync book.pdf \
  --ref-audio voice.wav \
  --ref-video face.mp4 \
  --format reader

Use a short, clear reference recording and a roughly ten-second face video with stable framing. Local LatentSync requires Linux or WSL2, an NVIDIA GPU, its sidecar environment, and downloaded checkpoints.

Output formats

Format Use case
reader Hosted reader assets plus a standalone offline reader ZIP with document, narration, and movable presenter; this is the default and recommended output
epub Text-and-narration accessibility export using Media Overlays; presenter omitted and reader support varies

For reader layouts, configure --face-position and --face-scale. Choose --latentsync-preset small for the fast 256px path or quality for the default 512px path.

To run inference elsewhere, combine the command with --backend remote and the remote GPU setup.