from Diary 2022-10-13 The field of non-realistic painting has opened up.

@nishio: I saw someone recently making a cartoon using AI of realist’s comments when camera was born, but I didn’t write it down. I couldn’t find it anymore. Since we are witnessing the birth of a new area right now, this post reminded me that it would be useful later on to record what I see and hear now.

  • @HomeiMiyashita: When studying media history, it is quite difficult to imagine the impact of the birth of the camera on society, or the feelings of painters toward photographers. It is quite difficult to imagine the impact of the birth of the camera on society, or the feelings of painters toward photographers. But only now, in early October 2022, following the keyword “AI painter” on Twitter may help us to imagine them.

@yurijinitai: I made a cartoon with Novel AI “I tried to find out if people felt the same way as AI does now when the picture was created.” #NovelAIDiffusionNovelAI imageimageimageimage - daguerreotype

My story diverges from this panel in this one.

  • image

@nishio: my story diverges from the third of these where “the field of non-photorealistic painting has opened up” @nishio: what I found in Stable Diffusion was the ability to “make paintings that are not realistic” image/c3cats/20220826100310 The above is the original picture of the cat that I still use as my project avatar. This is the first physical product I made image This is one of my favorite recent births. image

In fact, past experiments have generated many photographic-like pictures. The only reason why there are no such pictures in the project’s image list is that we and our couples are rejecting them. We have only released 1,500 of the 40,000 pictures we have generated so far. Examples of rejected pictures (we never introduced them before) image I said, “Isn’t this cat well drawn?” Wife: “In a place where cute pictures and videos of real cats abound, generating a realistic picture of a cat will never beat the real thing.” I “Sure thing.” That wife is the one who made the choice and 24 people like it on instagram. image Personally, I think the above is too subtle to be recognizable as a cat without being told it is a cat, so I prefer this one, even if they are the same “strange creature” type. This one was even made into a book cover the other day. image/c3cats/20220912034858

@nishio: In the context of the history of art, “When the camera was born, people who liked the technology and were not realistic painters started to make photographic works by trial and error with the camera technology. In the context of the history of art, “people who liked technology and were not realistic painters started to create photographic works through trial and error using camera technology”. - Open Source Drawing AI and Cameras

@nishio: “Engineers who understand what’s inside of a camera” made a camera specialized for a certain field of art style. I think it’s fair to say that the target audience will be repulsed by the camera. On the other hand, even if I specify cubo-futurism, I still end up with a picture like this, which is less attractive to me. image

With the original Stable Diffusion, you get a picture like this. In other words, by specializing in a particular area, the success rate in that area is increased, and the success rate in other areas is decreased (attracted to the target area). image/c3cats/20220829080905

@nishio: in other words, this is about me and my wife’s “Definition of good” not fitting the NovelAI Diffusion. I gave this kind of explanation the other day and someone asked, “Do you have to go that far to get a good picture?” I was asked, “Does it depend on the definition of “good”? Seed and prompt relationship for Stable Diffusion One aspect of art activities is to define “I think this is good” and present it to the world What I’m doing is automating image generation and generating prompts through reinforcement learning, so to use a camera analogy, it’s like “the camera will keep automatically releasing the shutter, and the camera will use its own judgment to compose itself by moving its head around, learning human preferences for which composition is best”. So the “I think this is good” search part was replaced by machine trial-and-error instead of human trial-and-error to speed up the process. In addition to that, I am a “software engineer” kind of artist, so I also “think this is good” as a meta-artwork of “a system in which a machine automatically keeps creating pictures to my liking. So “creating a network of meaning based on a set of pictures created by AI” is another layer of artwork Art medium. I want to express “I think this is good” not as one concrete work “point” x, but as the “probability distribution” p(x) that generates it. Each piece is a point x1, x2,
 sampled from p(x) Art as a probability distribution, not a point


This page is auto-translated from /nishio/憙漟的でăȘă„ç””ç”»ăźćˆ†é‡ŽăŒăČらけた using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.