AI Settings is optional. It powers three things in TvT: organizing videos with AI, guessing titles to fetch artwork, and writing captions. The page has three tabs: the AI provider, a TheMovieDB key, and Captions.
1. The provider
Two providers run on your own hardware, with no key and nothing leaving your network:
- Ollama on this or another computer: the server URL, Test connection, a model from the list the server reports (pick one that can use tools), the context length, and how long the model stays loaded after a job.
- vLLM on a GPU server: the server URL, an optional key, the model, max tokens, temperature, and a timeout.
The hosted providers, OpenRouter, Claude, and OpenAI, take your own API key from the provider, stored on this computer only, and a model from their list.
2. TheMovieDB key
With an API key from The Movie Database, Get thumbnail with AI on Manage Videos and Auto thumbnail on a watch folder can guess a title and fetch its real poster art.
3. Captions
The Whisper models write captions on this computer. Tiny is fastest, large the most accurate and wants a GPU; the English-only variants are a little sharper for English. Download and use one, switch between installed ones, or uninstall. The OCR packs turn image subtitles, the kind on Blu-ray and DVD rips, into text: the detector is required, then a pack per language.
Save settings applies any change. From then on the AI is used from Manage Videos, Organize Videos, and the switches on Video Locations.
Good to know
- Model downloads run in the background and show in System Activity.
- No provider set is fine: everything else in TvT works without it, and the AI buttons explain what is missing.