Its the training of AI that uses lots of energy, once its trained actually running it uses essentially fuck all. I can generate an image for like 0.0005kWh, it would use significantly more power to create the image myself as the PC would need to be on for much longer.

I don't mind AI generated images too much, depends a bit on context though. Background images to a low budget video which is mostly speech and just gives something to look at is something I have seen a few times and doesn't seem too bad. Probably won't be looking at them much anyway.