I have been working on artwork using various AI models for a couple months now. During this time I've been in several discussions with both laymen and other artists as to the ethics, growth, and potential for the industry. Each person I've spoken with I've tried to explain why AI isn't going to replace "real" artists. Below is something I stumbled upon from a thread discussing these very issues. There was one person trying to introduce AI Art to a community and most of the people responding were negative and some accused him of "stealing" other people's styles for his work. The following is the best response I've seen to date...
Hi. My name is David and I have been into AI for 7 months, first 3 with Disco Diffusion understanding the system, prompting, models, settings, schedule, etc. and nowadays using Stable Diffusion as a personal model trainer (sometimes private for IG AI artists). My currently job is professional material artist at Adobe and I’m ex concept artist (and more things) in the industry for 11-12 years but got interested in AI since May this year. I will answer and correct a lot of things said here thata were wrong.
First of all, it seems that some people still don’t know what is a diffusion or how AI works. I alreday saw Steven Zapata’s video where he told that AI replicates works. Utterly false, AI won’t replicate any input and no matter how much you train it to do it, actually it will mix the “style” of that inputs into the prompts you put for a guidance. So even you try to replicate same artworks the only thing that get similar is the style.
Let’s talk about the style. The style is something you build upon other works as inspiration, references, films, comics, books, etc. on a daily basis, that’s called a visual library and everyone have it’s own. Professional artist library usually is HUGE, AI’s VL is limited to the dataset which actually can be tweaked, changed and extended.
Before all of you start screaming it’s stealing, etc. No more steal than professional artist when inspiring in other works, other artists, films, netflix, comics, references, etc. They never (and me actually) paid copyrights for it. Same as when people do fan art, etc. Nothing is original, we work on a mix of things already created called “our own style” but the fact is that we neither invented anything new, we just put pieces in a unfinished puzzle.
The diffusion once again (“rendering”) its inspiring on the artist’s style or in the dataset but never will be using actual works in the final image, will be using the style AI thinks was used to create those.
About training. Training a model its easy, train a good model not so easy but never close at those prices in the first post IMO. A normal person can buy a 10$ subscription for 1000-2000 images generation in DreamStudio (Official Stable Diffusion site) or like me, can use VisionOfChaos or WebUI to run it locally which his or her own GPU card FOR FREE (paying electricity bill) OR use Google Colab also for free. I trained a model yesterday for 14 hours for a friend based on his works to 17.000 checkpoints, more or less.
It’s ok they’re lifetime but the images are SD 1.4 and can run for free even on Google Colab s on non paid accounts. Not worth IMO.
There are lately thousands of sites with ridiculous prices trying to jump in the AI art wave, my advice as artist and AI trainer, don’t go this way. You don’t need to pay a single $ and you don’t need (if you’re willing to pay) more than 20-30$ month to get a MidJourney account with better results (actually with V4) or just save to get a nice RTX 3000 series and use it locally on your PC.
That was a mic drop moment for that thread. I'll revisit AI later and discuss my use and the possibilities I see going forward.
