“HF_TOKEN is not set”

Follow the model access setup steps: create a HuggingFace token and save it to ~/Library/Application Support/video2text/.env.

403 / gated repo error

This means either you haven’t accepted the license on both model pages yet, or your token doesn’t have Read access:

• pyannote/speaker-diarization-3.1 — click “Agree and access repository”
• pyannote/segmentation-3.0 — click “Agree and access repository”

First run is slow

The first conversion downloads model weights (hundreds of MB to ~1GB per model). This only happens once — everything is cached locally afterward, and subsequent runs are much faster.

“Intel Mac not supported” / app won’t launch

video2text depends on torch and mlx builds that target Apple Silicon specifically. There is no Intel Mac build, and none is planned, since MLX (Apple’s ML framework) doesn’t target Intel Macs.

macOS blocks the app as “unidentified developer”

Right-click video2text.app → Open once. macOS will remember your choice for future launches.

Code changes aren’t showing up in video2text.app

Run ./build.sh again — it re-syncs app/ and static/ into the bundle. If you already moved the .app out of the original project folder (e.g. to /Applications), it can no longer be re-synced this way; rebuild before moving it, or clone a fresh copy.