Marvels of AI Art

The ultimate hero of the cosmos, merged together with the power of the stars! Now Elemental HERO Divine Neos Kirby… bring light within the absolute darkness of nightmare!

As promised yesterday I wanted to make a post today answering all the most common questions i've received over the months about prompting and a tutorial on how to do it as well as possible and also my thoughts on AI itself (which i've also been questioned about a lot) First half will be more of a tutorial and second half about the general state of things in regards to AI. 1) "What AI do you use" As of today, I use Dall-E 3 almost exclusively apart from some fringe cases where Midjourney v5 provides more accurate understanding. I've often read on forums that Midjourney v5 is more artistic and Dall-E3 has higher prompt accuracy, therefore it's up to taste which one you like better. I STRONGLY disagree on this. I would argue that prompt accuracy is by far the most important thing in generation and the reason for this is pretty simple. You can always prompt for more beauty and expression but you can almost never prompt for better understanding. Sure there is reference imaging in Midjourney but in its current state it falls apart extremely fast when it gets even just slightly complicated (basically a form of short-circuiting, something i'll get into more detail on in the next point). Even for future versions of AI generation, I would recommend to stick with the one that offers best prompt accuracy (but i actually also believe that there will soon be higher degrees of specialization so depending on the topic of the picture you want to make, a different AI might be better suited for the job, so prompt accuracy is for now an universal advantage of Dall-E 3 but everything will probably become more narrow towards specific sub-genres in the future) Using an AI that does not evaluate every word you prompt for and instead hallucinates too much tends to build bad habits in a person that uses it because if you don't get "punished" for having a suboptimal word in there and you're not clearly shown that it is suboptimal, you tend to get lazy with it and can't learn to be as precise as possible, which gets me to my next point 2) "How do you prompt" If you are using an AI that follows prompts closely, the next step is to prompt with a high degree of intent. The first step is to fully get rid of words that basically mean nothing - this is actually quite an easy step because it involves doing less but i keep seeing it so i think it might just be the most important: Words like "colorful" "4k, 8k, 32k" (there are upscalers online, why would you need it in the first generation?) "trending on art station" theres hundreds of variations like this i could list but i think it's easy to sort them if you're aware of it. These words are suboptimal because they lack depth, you need to imagine yourself as an art critic that looks at art and describes it afterwards, you can't just use "colorful"...to better describe color, use saturation degree, value degree, degree of color harmony on the color circle and color triangle. For light you can't just use "beautiful light, bright". Google is again your friend here. Stuff like rim light, catchlight, chandelier light are far more specific and offer a way more natural look. Every prompt should have as many descriptions of texture as possible without short circuiting the engine due to complexity. A list of textures can again be found on google. If your prompt starts with "A picture of" you are wasting at least 50% potential. Because AI always focuses on the beginning, this is where you can add an insane amount of direction. Ideally you'd start with what matters most to you, be it camera angle, lighting, colors or texture. "A wide angle picture of" would for example be a variant that greatly increases the potential of your idea being realised exactly how you want it. Additionally, adding a style of painting to the prompt always helps,therefore it's vital to study art history itself (with words like impasto or sfumato) (or if it is meant to be a picture taken by a camera, add which camera) Generally, you want to make your prompt as long as humanly possible, just barely below the point where the AI cannot deal with the complexity of it anymore. This is an extremely fine line and it's very easy to throw in the towel and not try to seek this small landing point but i think this is what makes AI art fun in the first place, stacking up the ideas while steering it in a way so it won't short-circuit due to overload. 3) AI and the future of art in general Ai art in general is a thing i don't think can be put back into the box. It's here and the programs for running AI will become easier and easier to create for anyone, so banning it now would only cause it to resurface later by someone who can make it cheaply on his computer a few years later. I believe AI art and regular art can peacefully co-exist, if they are properly labeled. I made it my point to put the word AI art in my name because it is not in my interest to deceive anyone, and i feel like a lot of bad blood has arisen out of people that mislabeled art as human-made when it was AI made or partially AI made. Additionally, I have not fixed flaws in my pictures with human input like photoshop because that would oversell the current capabilities of what AI can do (and make it a human-AI mix) I would like to see full transparency everywhere so people that would like to enjoy human art can do so and people that want to see AI can do so as well. Hell, depending on my mood and day i also like to see one category only and I believe that AI has the potential to expand the general interest in art rather than cut into the marketshare of human artists, as long as the labels are always transparent for which is which. Also AI pictures have ignited a lot of passion for me for human art as well that i previously did not have. In fact, i was never going to draw in the first place. Yes i could have picked up a pencil but given my life i just don't have the time to study drawing and my life trajectory for "pictures created in my lifetime" would simply remain zero. Additionally, I feel like despite my inability to draw i have gained a strong sense of which art has AI in it and which doesn't, so if you struggle to tell which is which i would recommend to actually prompt a lot with everything that is out there so you can get a feel for the general main quirks of AI images. I did not anticipate that i'd be able to notice these small details but sometimes I can actually adequately guess what words were used in a prompt when I look at AI art (and its a fun game to do, too!) Hope that answers everything, if you have any more questions feel free to ask and i will answer in the comments. Next picture is coming tomorrow, also i'll be taking a small break over christmas. Also thanks a ton for 7k followers!