cross-posted from: https://slrpnk.net/post/16155961
Your photos reveal a lot of private information. In this experiment, we use Google Vision API to extract the story behind a single photo. Upload a photo to find out how much an AI sees: https://theyseeyourphotos.com/
Here is one example of what the output would be, while not specifcally very senstive - it did catch the model of my phone, which is a friendly reminder to use metadata remover for your images before uploading them anywere!
Simply using “they” is nothing but fear mongering. Be specific and say who “they” are and why “they” matter. Otherwise you’re just telling people to trust you without providing any reason that you are trustworthy and why “they” aren’t.
All those details are easily noticeable by a regular person. I’d be more impressed if it was able to figure out the location of a photo without using metadata.
Summon geowizard!
The image-to-text model is impressive. I could see it being useful for smart search of your library, allowing users to find photos with a high-level description.
I’m not sure why it’s being reported on as though the technology is a privacy or security threat, though. If you’ve given a storage provider access to your photos anyway, using a vision model isn’t going to give them anything extra.
That said, I do love self-hosted photo solutions like Immich and Ente. Hope they continue to grow.
