Yup, see article, video uses a lot more, my argument is maybe only one in a hundred is a creator, the rest just consume content, so not everyone will be high energy user of AI
I believe my text-only interactions with LLM cost him a lot more energy then average Joe ones. What about video generators it is sad when I see a youtube video with a really expensive generated picture because I feel somewhat guilty not to look at footage and listen to only words. A lot of water boiled for nothing.
Using tools like remotion can help, then AI is not generating massive images, it's helping compose scenes, where the local CPU then renders the video. I use this technique, so my videos may have only 10 seconds of AI generated video in a 10 minute video, whilst still having motion and keeping people's attention through engaging visual elements.