.env.example

# Optional: override where app-generated WebM recordings are stored.
# If omitted, recordings are saved under Electron's userData directory.
# VIDEO_TOOL_RECORDINGS_DIR=/absolute/path/to/recordings

# Optional: VieNeu-TTS v3 Turbo local runtime. OpenScene auto-discovers a
# sibling VieNeu-TTS checkout, starts its existing .venv on app launch, and
# stops only the process it owns. Run `npm run setup:local-ai` once after clone.
# Only loopback HTTP URLs are accepted.
# OPENSCENE_VIENEU_BASE_URL=http://127.0.0.1:8001
# OPENSCENE_VIENEU_PROJECT_DIR=D:\tool ai\VieNeu-TTS
# OPENSCENE_VIENEU_AUTOSTART=true

# Optional: user-managed ComfyUI worker for Wan 2.2 Animate Motion Control.
# Plain HTTP is accepted only on loopback; use HTTPS for a remote worker.
# Export each graph with ComfyUI "Save (API Format)" and give the input/output
# nodes the exact titles documented in docs/planning/2026-09-06-ai-video-studio-phase-5c.md.
# OPENSCENE_COMFYUI_BASE_URL=http://127.0.0.1:8188
# OPENSCENE_COMFYUI_WAN_MOVE_WORKFLOW_PATH=D:\ComfyUI\workflows\openscene-wan-move-api.json
# OPENSCENE_COMFYUI_WAN_MIX_WORKFLOW_PATH=D:\ComfyUI\workflows\openscene-wan-mix-api.json
# Optional bearer token when your HTTPS reverse proxy protects ComfyUI.
# OPENSCENE_COMFYUI_API_KEY=

# Optional: explicit whisper.cpp overrides. Without them OpenScene discovers
# the verified runtime installed by `npm run setup:local-ai` under
# .local-runtimes/. Both override paths must be absolute. The checksum is
# optional for development and recommended for every custom runtime.
# OPENSCENE_WHISPER_CPP_PATH=D:\whisper.cpp\build\bin\Release\whisper-cli.exe
# OPENSCENE_WHISPER_MODEL_PATH=D:\whisper.cpp\models\ggml-small.bin
# OPENSCENE_WHISPER_MODEL_SHA256=
# Optional CUDA package. Leave off on GTX 1650: the official generic CUDA build
# benchmarks slower than its CPU build on this GPU because it lacks Tensor Cores.
# OPENSCENE_WHISPER_CUDA=false