- Dec 9, 2020
- 135
- 106
But is it hentai?If you want to be truly amazed by AI generated art, search for whatYou must be registered to see the linkscan do...
What is truly amazing is the soft body and clothes deformation are fairly accurate as well as the fluids.Like the AI gereated waifus are so good, it almost feels wrong. The only problem is on pixAI they give you very little credits, like enough to generate 5 to 10 images a day at most which is not enough for how good they look. Check these out:
View attachment 2602930 View attachment 2602932 View attachment 2602933 View attachment 2602938
No, what it truly amazing is how easily AI enthusiastic can miss so many things when they looks at an image.What is truly amazing is the soft body and clothes deformation are fairly accurate as well as the fluids.
That means the AI internalize all the physics and material properties.
You know, it don't suffice that there's shades for the lighting to be correct. The shades also need to be consistent with the light(s) source(s).Not to mention the lighting as if it got that wrong you would have immediately noticed.
But everything is shaded and in shadow as it should based on the light source with enough ambient light for things not to be too dark.
Well, if it's the case, then they have the same understanding of those rules than a preschooler.This is what AI Skeptics do not understand and cannot comprehend, AI do not just merely mimic art, they have already internalize the "Rules" and Anatomy that make up the Technical Skill of Drawing.
Yeah that looks weird but maybe you haven't tried to fold the boobs hard enough lol.[Note: All position are "of the screen".]
No, what it truly amazing is how easily AI enthusiastic can miss so many things when they looks at an image.
Take the time to really look at the second image, and come again tell that the soft body and clothes deformation are fairly accurate.
In no world the boobs fat can fold up on itself
I'll just repeat this again as someone who minored on information science (not the library kind) and neural nets: no they don't. absolutely the last thing they do is any kind of cognitive processing. they 'just' copy things and interpolate between stuff the programmer exposed them with. the result is always 'stuff that LOOKS like X' rather than 'creating new thing Y'. we don't know how to do the reasoning part, yet.That means the AI internalize all the physics and material properties.
This is what AI Skeptics do not understand and cannot comprehend, AI do not just merely mimic art, they have already internalize the "Rules" and Anatomy that make up the Technical Skill of Drawing.
It works with hentai logic which is exaggerated.Take the time to really look at the second image, and come again tell that the soft body and clothes deformation are fairly accurate.
In no world the boobs fat can fold up on itself ; especially on a so large distance since it cover entirely the bra. And what amazing asymmetrical bra ; the right part is plain and large, the left one is small with a big all in the middle.
No, you just do not understand what an achievement it is for the results we got.You know, it don't suffice that there's shades for the lighting to be correct. The shades also need to be consistent with the light(s) source(s).
Those are stylized material reflections on the skin, similar on the tight, similar on the hair, same as the braw.And the frontal highlight spots on front of the girl imply a top (a bit on the right) front light that cast no other shadow than for the left boob.
I totally agree with this. In fact it's precisely why I consider those images as not being bad at all.Either way, a lot of those images at first glance look great to me, then I notice the mistakes and weird anatomy later but those can be corrected by cropping the image where the mistake happens (lazy solution)
or an artist can correct those in photoshop or redraw them and improve. I am not that good to do that, but someone with experience in digital art could easily fix mistakes.
They don't just "copy".they 'just' copy things and interpolate between stuff the programmer exposed them with. the result is always 'stuff that LOOKS like X' rather than 'creating new thing Y'. we don't know how to do the reasoning part, yet.
Exactly.THAT's why it can't understand how it attached the black dude's hands on the girls arms, this interpolation result just happened to fit its training images closer than the girl retaining her own hands with the black dude behind him groping her tits.
They are pretty good for AI for sure, and ofc it is basically copying and remixing styles of actual artists to get something new, but I am very happy with the anime style there because I have been trying stable diffusion for free and they do get anime style look but not as sharp and precise as on that site.Does brutally good mean something other than what I think it means? These all look like the super generic crap you'd find in someone's first ever doujin, plus some AI artifacts like extra limbs ruining the already mediocre quality of the images. Sure it's good for something made by an AI, but it's not so good I'd want to hang it on my wall or make it the background on my computer.
Yeah, it basically remixes styles and patterns of human authors so it is difficult to get a specific thing sometimes. I think it will improve in the future where you will have greater control over image generation and add more poses, and specific things to it which will enable people to create their vision more easily. For now, it takes a lot of tries but the results can be cool.AI is cool for the first 200-300 images you see then once you start to notice patterns in the images you're seeing, you start to realize its all icing and no cake. I've made about 10k images with it, got frustrated and gave up. I have a little skill with art and when I try to get it to generate an image of something I have in my head, I end up just fighting tooth and nail against the data it's trained on, ultimately to no avail. Something as simple as a bikini nipple slip, if the data wasn't trained on it, will never show up no matter how much emphasis you place on it. All my ventures in AI just really made me appreciate human art a lot more, and its actually made me get back into art as well.
Ha. You should see what you can do with a local AI install with Stable Diffusion using something like Automatic1111.Like the AI gereated waifus are so good, it almost feels wrong. The only problem is on pixAI they give you very little credits, like enough to generate 5 to 10 images a day at most which is not enough for how good they look.
This is just WRONG. I'm an artist AND an AI enthusiast and everything you said is just wrong.What is truly amazing is the soft body and clothes deformation are fairly accurate as well as the fluids.
That means the AI internalize all the physics and material properties.
Not to mention the lighting as if it got that wrong you would have immediately noticed.
But everything is shaded and in shadow as it should based on the light source with enough ambient light for things not to be too dark.
This is what AI Skeptics do not understand and cannot comprehend, AI do not just merely mimic art, they have already internalize the "Rules" and Anatomy that make up the Technical Skill of Drawing.
Don't know if you've used a local install of AI image generation, like Automatic1111, but it gives you immense control now.AI is cool for the first 200-300 images you see then once you start to notice patterns in the images you're seeing, you start to realize its all icing and no cake. I've made about 10k images with it, got frustrated and gave up. I have a little skill with art and when I try to get it to generate an image of something I have in my head, I end up just fighting tooth and nail against the data it's trained on, ultimately to no avail. Something as simple as a bikini nipple slip, if the data wasn't trained on it, will never show up no matter how much emphasis you place on it. All my ventures in AI just really made me appreciate human art a lot more, and its actually made me get back into art as well.
Yeah, I went the lazy way instead of following instructions and installing plus I don't know if my graphic card is strong enough to support using it on my computer. If it is I would need to spend a couple days learning the tools and commands but it sounds very very nice.Ha. You should see what you can do with a local AI install with Stable Diffusion using something like Automatic1111.
All the settings are exposed for you to just go nuts, with no censorship. And there are hundreds of models that can do pretty much whatever you want, and these things called LORAs that can let them understand really specific concepts.