Raw LLM Responses

The funny part is that there is already very, VERY well established case law that human authorship is a PREREQUISITE for copyright protections. This actually goes back decades, to a case in which a monkey took a photo with a photographer's camera, and the photographer tried to claim copyright on it. He failed, as neither he, nor any other human, was the author of the photo, so the photo had no copyright protections and was effectively immediately made part of the public domain. Recent cases found that this applies, for all the same reasons, to wholly or minimally unmodified output from generative models, where an "AI Artist" attempted to copyright a 'comic book' made entirely of midjourney prompted images. The court found that simply arranging prompted images does not constitute artistic authorship for the images themselves, and the "comic book" had no more benefit of copyright protections than if the AI bro had just put together a bunch of unmodified public domain images and tried to claim they owned them. So even if the AI bros succeed, and imitate someone's style... by law, they can't actually monetize it in any exclusive way. Since ownership and monetization without paying artists is the entire point for these types, with no consideration to actual artistic communication, to what the art says or means or the perspective it communicates, and they can't own or exclusively monetize the output anyway, this renders the whole exercise meaningless and spiteful. An example of deep learning being used in a way that does constitute human authorship is Weta Workshop. They've developed a bunch of tools for things like the Avatar movies, where strictly simulating things like the water effects in Avatar 2 was just not possible with modern hardware, so they had artists and engineers build their strict simulations, and then built an entirely in-house generative model, trained on their own simulations, that acted as a tweener to provide generated approximate motion data to unify the small-scale simulations of bubbles and foam around individual characters and objects in closeups with the large-scale simulations of waves and macro-scale water dynamics. The generative model basically filled the gaps in the data between the different types of sims, so that a human artist could then take that now-unified water sim and use it as the starting point to compose the final effect. That's an appropriate use of deep learning tools.