Comments/Ratings for a Single Item

I did get a kind of Centaur when I asked it to draw a chess piece of a doe. But it was a lady centaur with bare breasts and a weird horn on one side of her head. When I repeated the prompt, I got a hornless deer, which is what I was originally going for. See the Withdrawer page for that image.

Oh! The lady centaur (a centauride) would've been great for the Amazon page, as an Amazonrider, which I call Centauride... well, if not for those two other details.

I was speaking of illustrators in the context of the boardgames edition, not in the case your activity. What I'm saying is a fact, a reality that has already started. For authors, I was meaning authors of games, not books, in the context of my discussion. It includes chess variants too. It is a not reality yet although you had yourself presented your own experience with Chat GPT few monthes ago. Time will tell

H.G., it is indeed solely based on the product for me. It is not about the AI itself, I'm completely fine with that. It is simply not ready yet to produce pictures of actual value. I can say that pretty much all AI generations lack something that I would call harmony or function. I just feel it this way, possibly because I had formal exposure to this area. It's a picture merged from thousands of pictures. An approximation. Despite approaching it very closely, there are no real harmonical relations between parts, and if you don't notice it, just see that if you request an AI image of a human, they will have from 3-4 to 7 and possibly even more fingers, a very bizarre phenomenon on the subject which shows that as of today AI still misses many details. My feeling here is solely based on the properties of the shape itself. AI can produce interesting and sometimes unique results, especially with tight prompting, but the image will never look really finished, and, therefore, its value is the value of a sketch. Many artists and other people are fine with that. If that works, then it's okay. This chess piece looks really great, but, there's something that is just left incomplete.

I myself generated many pictures based on works by a single artist that I like. Without the correct prompting, it looked tacky as hell, but when AI seems to have limits in what it does, the results approach the definition of astonishing.

Instead, I would find it to be worth developing new and simplest possible designs that can be associated with a particular chess piece if it had no physical design before. Chess pieces are generally not art items, but rather products of industrial design that can be mass-produced, so a different approach is required. They should be simple to be made by a hobbyist, using a lathe or 3D printer, ergonomic, and distinctive. Why not include such files (which will be eventually made by someone) in every Piececlopedia article? There's no need to create a design classic, just something that can be very simple, yet different and efficient. It should be compatible with a standard Staunton-like chess set. All of these images are instantly incompatible if placed together with usual chessmen, which is also an important criterion aside from their artistic side.

In 9x9 Modern chess, they used a minister's hat for the minister (BN compound), it was simply designed as a usual chess king with a cylinder hat instead. It looked really simple and efficient and could be easily made on a lathe. In a slightly more sophisticated way, they made excellent designs for the Champion and Wizard in Omega Chess, It's sort of sad that they discontinued selling them, but with enough skill, they can be also made out of wood or be printed, although I'm not sure about the copyright issues.

just see that if you request an AI image of a human, they will have from 3-4 to 7 and possibly even more fingers, a very bizarre phenomenon on the subject which shows that as of today AI still misses many details.

Leonardo.Ai allows the use of different models. I guess these are different algorithms for drawing that have been trained on different sets of images, and different ones give different results. While some act as you describe, the PhotoReal model, which I'm currently paying to use, does not, and I've also been getting really good results with DreamShaper.

They should be simple to be made by a hobbyist, using a lathe or 3D printer, ergonomic, and distinctive. Why not include such files (which will be eventually made by someone) in every Piececlopedia article? There's no need to create a design classic, just something that can be very simple, yet different and efficient. It should be compatible with a standard Staunton-like chess set.

I fully agree here. Of the designs I've seen, I think Jean-Louis's come the closest (certainly, mine don't fit with Staunton designs, or at least scant few do).

Chess pieces are generally not art items, but rather products of industrial design that can be mass-produced, so a different approach is required.

They are both, and there are plenty of examples of the former.

They should be simple to be made by a hobbyist, using a lathe or 3D printer, ergonomic, and distinctive. Why not include such files (which will be eventually made by someone) in every Piececlopedia article?

We can do that if someone supplies them, but it's not an either/or matter of doing only one or the other. The AI art is for stimulating the imagination, not for constructing physical pieces.

... I've also been getting really good results with DreamShaper.

Nice that you mention this. I have it a while with about a dozen or more possibilities (not including three or four tries at getting it to understand what a centaur is), and all I got decent results for were Exorcist, Tax Collector, and Vulture (and an almost good one for a Bharal). That's actually more than I expected.

To temper my previous comment for the Withdrawer, this illustration for the Advancer is much meaningful. Here it looks like a chess piece.

Ultraadvancer Chess