screencastgen

Run with a remote GPU

Remote mode keeps document extraction and media composition on the client while the GPU server performs TTS, alignment, transcription, and lip-sync inference.

GPU host

python3 scripts/setup.py --profile local-gpu
source .venv/bin/activate
screencastgen-server \
  --backend qwen \
  --device cuda \
  --aligner whisperx \
  --lipsync-provider latentsync

The server listens on port 8100 by default. Put authentication and TLS in a reverse proxy before exposing it beyond a trusted network.

Client

python3 scripts/setup.py \
  --profile remote-client \
  --server-url http://gpu-vm:8100
source .venv/bin/activate
screencastgen doctor \
  --profile remote-client \
  --server-url http://gpu-vm:8100

Run a client pipeline with the remote backend:

screencastgen highlight book.pdf \
  --backend remote \
  --tts-server-url http://gpu-vm:8100

The remote lip-sync contract submits a job, polls its state, downloads the result, and removes temporary server output. See the inference server and remote GPU client references for the HTTP contract.