The server prints it at startup and stores it in .token in the app folder.
Xiaohongshu image notes use OCR; supported video sites use captions or ASR.
Or drop an audio/video file anywhere on this page — files are also picked up from the
inbox/ folder automatically.
First run of a local model downloads it once (~small: 460 MB, large-v3-turbo: 1.6 GB).