Upgrade your account to a Community Supporter account and remove most of the site ads.

Reply to thread

Message: <blockquote data-quote="Art Waring" data-source="post: 9577001" data-attributes="member: 7037141">I honestly don't. After years of extensive research, MIT has concluded that ai companies will effectively run out of new training data by 2026, forcing them to then train on "synthetic data" instead. Training an ai on synthetic gen-ai data is proven to lead to model collapse. IMHO, gen-ai has already peaked, now we are going to see the long and inevitable decline over the next few years as they try to find a way to keep feeding the oroborus as it eats its own tail.Additionally, artists are using tools like Glaze & Nightshade to poison their work, leading to further potential model collapse due to the gradual poisoning of the datasets. There is nothing about the technology that is inevitable, as plenty of defunct technologies lie forgotten on the ash-heap of history, this is no different. A few years back I was reporting here about how some artists I know were losing work to gen-ai, well that was when the tech was new and largely unknown (except by hype). Today, those artists that were fired are now re-hired in a lot of cases because the tech cannot perform at industry standards without significant editing by human artists. The entire process was wasteful, as they not only wasted time firing artists, hiring promter-types with no actual experience making art at an industry standard, but then had to fire the promters, and re-hire skilled artists (now at a higher pay rate), all just to get back where they were in the first place (and wasting time and money in the process).If you pay attention to the gen-ai news, they are already moving away from image generation towards video generation in an attempt to disrupt yet another industry for their own personal gain. I don't think we will be seeing significant advancements in image generation because these companies are constantly chasing the next new thing. Currently, their attention is shifting over to video generation like SORA (its initial release was on December 9th 2024, just a few months ago). "Good enough" will be enough for some people, but there are simply things that gen-ai cannot do...For example: • Specificity: gen-ai cannot achieve specificity, that means if you want to create an image of a unique character that you have made for your game, and have tons of details about the character that you need to include, everything from appearance (which can cover a myriad of details, unique hair styles or physical traits, scars in specific places, specific articles of clothing etc) to any items the character may be carrying, you will likely not be able to get the results you want, and will be forced to accept whatever it gives you instead. I hope you like making creative compromises based on the limitations of a machine, because human artists simply don't have this problem.• Genre Limitations: Which leads me to my next point. If you are trying to make fantasy-genre images using gen-ai, well that is easier because gen-ai was trained heavily off of fantasy artwork (primarily artists that made work for MTG & D&D are featured on "style lists" on midjourney, where they deliberately stole the work of thousands of talented artists like Greg Rutkowsky in order to mimic their work for profit). If you try and make anything with the word "cyberpunk" or other niche genres, you will start running into problems, as they obviously trained off of art featured in the popular game Cyberpunk 2077, you will get countless iterations of "cyberpunk 2077-esque" pictures with the obvious color schemes and a significant lack in any variety in terms of what you can achieve using promts to explore genres that have not had a significant amount of work in the mainstream. That means that for anything other than generic fantasy, you are simply better off working with an artist that actually knows what they are doing.In the short term (2022-2024), it did manage to disrupt the lives of a lot of working artists. In the long term, it actually helped to highlight just what its limitations are, and thus help artists get a much needed increase in their pay rates, because a lot of companies learned the hard way that if you want industry level work, you are going to have to pay for it.</blockquote>

Verification