It would be difficult for a large volume of uploads, but TensorFlow does this in PhotoPrism for facial recognition, etc on a Lenovo thin client running CasaOS in my house. I uploaded over 6000 images and it finished in a few hours. 

That’s just a quad core old i5. 

 GIFs are not images, the action requires a different model on top of images and yet one more model to tie it all together 🐶🐾🫡 

 Ok. I guess I’m too smooth brained. 

 You are not, it’s just not the usual thing people think about. And then you also need a search engine on top of that and embeddings, etc 🐶🐾🤣🫂