This is what hCaptcha is currently doing, they are switching the image category every 24-72 hours. How useful is it? Not very. Modern ML models such as mobilenet, resnet or yolo require only a few hundred images for it to be accurate to solve those captchas.

You don't need few million samples, with 500-700 images per category you are more than ready to solve current captchas.

btw hCaptcha has an accessiblity page for you to sign up and never solve a hCaptcha again.

here is the link https://dashboard.hcaptcha.com/signup?type=accessibility

*edited typo

I tried it doesn’t work

you need to enable cookies

Yep, the cost of keeping the model up to date would be negligible compared to the hosting bill.

I wonder if stable diffusion / dall-e type offerings could procedurally generate images?

hcaptcha already seems to be using ml to generate their challenges