Support.

That is expected. The cloning engine and the isolation model are not bundled, to keep the download small. Open Settings (Command-comma) → Models and press Download All. That fetches about 3.3 GB once. Generation and Isolate Vocals stay unavailable until it finishes.

By default inside the app's own container. Settings → Audio → Download location shows the folder and free space, and Change… lets you point at any folder you can reach, including an external volume. The choice is remembered across launches. Models already downloaded are not moved for you.

Separation is genuinely compute-heavy: roughly 45 to 60 seconds per eight seconds of audio on Apple silicon. It runs on the GPU. A long import takes a while, and the progress bar reports where it is.

Yes. Settings → Audio → Import processing. Isolation is on by default because a reference built from a clean vocal clones better, but material that is already a dry voice does not need it. You can still isolate any source by hand from the Studio.

Check the clone-readiness score on the Create Voice step. It flags clips that are too quiet, too short, clipping, or mostly silence. Trim the waveform to the cleanest few seconds — five to ten seconds of clear speech beats a long take — and the score updates as you trim.

Silo Clone accepts a direct audio or video file URL only. Streaming playlists and web page links are rejected with an explanation. The app does not download from third-party media services.

Settings → Audio → Input, or the picker in the Record sheet. Both list every Core Audio input on your Mac with its channel count, the way Audio MIDI Setup does. Output has its own picker for playback.

Publish the voice from the Voice Library and the Silo and Ohms apps installed on your Mac can read it from the shared catalog. Only apps actually installed appear. You can also export any voice or take from the share button.