@davidrevoy

> But lately I started to enjoy this behaviour in a new way, because this faking could be what wastes all the data sets and training models

AI artists from civitai have already made a million of loras, trained on the midjourney outputs. For now it's more like compiler bootstrapping rather than "eating ourselves".
To be honest, I think the main problem with using generated images in training is that someone would say that even though the datasets contain no copyrighted hand-drawn images, it's still "unethical" , because they are generated using old models and continue this "ethics" shitpost.
In order for this to make sense, we first need to create some kind of model to which no one will have any complaints, and then start bootstrapping.

> I will continue to publish all under a permissive Creative Commons licence and with the source files, because that's how I like my art: free and libre.

Just at the moment when computer geeks came up with idea how they can use your images to create somewhat free and open source it turned out that it's forbidden. This is funny.
I understand why you may not like midjourney, firefly, dalle2 and other non-free stuff (i don't like them too), but SD is completely free, open source and can be used or modified by anyone...

> With AI, AIArtists simply type in keywords for a subject.

No, we don't, stop it. With no control and editing at all ai draws random broken shit.