Aligning a full-length novel takes hours on a Mac running Storyteller in Docker/OrbStack

Self-hosting Storyteller on an Apple Silicon MacBook via OrbStack. Short clips align fine, but a full novel takes so long that I avoid using it for real books. I assumed OrbStack was already using the Mac's hardware since it is not "Docker".

Resolution: Containers on macOS cannot reach the Mac's GPU — this is true of Docker Desktop, OrbStack, and any other Docker runtime on macOS, since they all run the container inside a Linux VM — so Storyteller silently falls back to CPU transcription. The gap barely shows on a short clip but is enormous on a whole novel. The fix is the offloading-transcription setup: run a whisper.cpp server natively on the Mac (outside the container) so it can use Metal, then in Storyteller's settings switch transcription to remote and set the base URL to http://host.docker.internal:<port> — use the port whisper.cpp reports listening on, typically 8080. Keep the whisper.cpp terminal window open (minimized is fine) for as long as you want transcription available. For English, tiny.en is enough; otherwise large-v3-turbo. Sensible thread settings are Threads: 4, Processors: 1 — do not go higher. Everything else, including kicking off jobs, still happens through Storyteller's web interface.