Captioning photos with KI I

I am a hobby photographer. Over the last 20 years I did around 100k fotos, which are now sleeping on my storage array. Never will I be able to give each one a caption. I tried to give whole photo shots some keywords but this was only particular. I tried some Photo software which aim was be to make photography easier. It does not work. So I am here with 100k+ fotos and no software to help me. Can KI and GPU near limitless GPU power help me out?

So how do I get automatically captions for all my images.

In IMG2IMG mode Stable Diffusion has a button "Ask CLIP" which gives you a textual description of the image you have uploaded.

This looked promising. But a search for CLIP does not was conclusive, but offers BLIB. 

So I wrote a small code utilizing the great huggingface eco system to use BLIB to caption my images. It took a few iterations to convice the software to utilize the GPU. But then my captioning code runs fast and was only limited be the SQL write rate into the sqlite-DB I used for the Filename -> Caption storage.

After some hours on a NVIDIA 4090 system 60000 Images were captioned.