- This topic has 37 replies, 23 voices, and was last updated 2 years, 11 months ago by
.
-
Posts
-
To paint something which looks realistic one needs a reference. This applies both to humans and AI. Looking at photos or around for inspiration is not stealing. As far as I can tell AI doesn't copy elements like glasses or rings. As a result such things often look strange.
When future generations rely on AI to think then we have a BIG problem.
No I won't participate in feeding the machine however tempting.Computers hate us
They've been eavesdropping on all our hatred and ignorance for years now
They're bloody well sick of it
Even though we are cooking our planet making enough electricity to keep them all going
they are not satisfiedThat's how I see it too @gremmie
And we are getting more and more stupid. We just type in a few words and think then "Wow.... Look at This......COOL".I've just seen a post on Reddit where someone has used AI to take some clips from Stanley Kubrick's 2001, one of the most famously meticulously cinematically framed films of all time and reframe them vertically and some comments seriously suggest that it's an improvement over the original.
Prior to seeing that I was mostly either cautiously positive or ambivalent about AI art. Now, saying I'm unimpressed is a bit like saying the sun is 'a bit warm'. When one of those chuckling idiots finally unleashes Strong AI trying to create a waifu girlfriend, and it wipes us off the face of the planet, we deserve it.
I use Daz because I'm either to lazy or to unskilled to created what I want. I effectively use resources others create to make my image.
I'm going to assume the same from AI "artist". They are to lazy or unskilled to create their own content.
That said I think it's a bit more drastic.
The artist that creates the texture for G8 probably had nothing to do with the mesh. The artist that created the environment we stick G8 in had nothing to do with G8. The lighting bundle had nothing to do with all the above. The poses even less so. We grab all that stuff and design an images.The AI artist likes image 3 and 8. He tells the AI to mix them and clicks a button.
It takes lazy and theft to a new level. It's a pirates wet dream. Steal everything with little to no effort and let something else think for you.
I like the images people REcreate with it but I'm not a fan.
Rank:
I think its neat, but will get much much better...
It's an amazing piece of software programming, but it relies on too many variables that make it hard to work with to guarantee a result that the end user is happy with. You gotta play with the description a LOT. And it's NOT affordable, a lot of what I have seen so far you practically have to pay for in some way or another. And the ones that are available in the Microsoft store for purchase aren't very good.
I read recently that Adobe is taking a leap into the programming. Good luck with that!
If you're talking about automatic1111 you can go to Civitai where everything is free. There's 50-100 new items there every day. I agree that there's no way to guarantee consistency but reusing the initial seed with variation seeds you can make changes without generating a completely new image.
What is totally annoying is when you get so close to the image you want but just one thing (usually hands/eyes) is off and you can't fix it even with inpainting. I must have discarded thousands of images that were 99% good but spoiled by one thing.The trouble with the current state of AI (requiring training) is that it needs original art (or writing, or code, etc.). If, over time, a majority opts for AI 'creations', original art will decline and AI creations will stagnate because all AI has to refer to are extant art and various user manipulations of it's own output. It's a little like digital incest.
That being said, users could use AI as a 'toolbox' to prototype new concepts, but the conceptualization remains with the artist, AI might let the artist more rapidly try out new ideas, although current limitations can somewhat put the artist in a 'box'.
I've been testing Stable Diffusion and it's useful for some purposes. In general though, any attempt to throw something truly creative at it usually winds up with a mutated subject that looks like a horrible genetics experiment gone wrong.
Even in the best of outcomes it usually produces hands that look terrible. Some of the hentai/anime models don't look so bad because drawn-by-hand anime/hentai artists often take shortcuts with hands and fingers. However, for realistic models - a perfect human specimen to have mutated sausage fingers ruins the effect.
In any case DAZ can probably take some tips (and maybe even code) from Stable Diffusion in the future. The tension on clothes digging into flesh just the right amount is really good in many SD models. The hair is much more realistic than DAZ. Clothing rarely has pokethrough, though it does fall off from time to time.
- You must be logged in to reply to this topic.
