I have been building PhotoScribe, a free and open source tool that writes titles, captions and keywords for your photos using a local AI model. Everything runs on your own machine through LM Studio or Ollama, so your images never leave your computer.
You point it at a folder, it looks at each photo with a local vision model (Gemma works well), and writes the results to standard metadata: IPTC, XMP and EXIF, plus XMP sidecars for RAW files. It supports the darktable and digiKam sidecar naming (photo.cr2.xmp) as well as the Adobe convention, so the tags turn up wherever you catalogue. You can feed it a keyword vocabulary for it to ‘snap’ to, batch whole folders at once, and it reads any existing tags and GPS so captions can name the actual place rather than guessing.
It’s been through quite a few iterations to get to this stage and received some excellent contributions and user suggestions that made it into the build. (There’s also a Lightroom plugin version but I’m guessing that’s not of interest here?)
It runs on macOS and Windows, and from source on Linux.