AI generated art

[gaogao]

Let's all calm down. Whatever he says is not something anyone needs to be triggered. We're here to get a cheap but quality erection, speak and choose with our penises (or whichever genitals you have), and not politicalizing AI technology.

Anyway, I'm no simp (feel like I must make that clear) but AI can read images. So checking on the correct image to solve a captcha is not out of the possibility.
You can read about it here:

You must be registered to see the links

or

You must be registered to see the links

If you don't have time then:

Spoiler: image

You don't have permission to view the spoiler content. Log in or register now.

 

[Meaning Less]

Anyway, I'm no simp (feel like I must make that clear) but AI can read images. So checking on the correct image to solve a captcha is not out of the possibility.

And did I said they couldn't?
The thing you are missing is that captchas are made out of images that google AI already failed to identify, when they aren't clear enough or don't fully match the training data the results are inconsistent, that's why they delegate them to people.

And if AI could have properly detected them in the first place then those captchas wouldn't exist. So it is a self-defeating argument when you understand why captchas exist.

Only the day those captchas stop existing is the day one could argue "AI has no issues at all detecting images".

 

[gaogao]

And did I said they couldn't?

And did I say you were wrong? I said it's not worth arguing over, and here you are - being yourself, Meaning Less.

 

[XcentY]

This is AI

You must be registered to see the links

 

[Pervtron3000]

Is anyone here up to date on the cutting edge of machine art?

Iterating on what I said on page 2, if (WHEN) a company like Adobe were to fully embraces machine art and empowers it with more context and awareness, and make things super easy for creators to touch-up and iterate on the works, I think we'll have a golden age of higher quality sex games.

Let me give a very very basic example of touching up an image for something specifically made to remove text overlays and watermarks

You must be registered to see the links

You must be registered to see the links

Apologies if those two images don't work, maybe these will instead

You must be registered to see the links

You must be registered to see the links

This is a machine learning tool that is purpose built with contextual knowledge about what text and overlays are.
Its quite possible that such a tool could be confused if an area highlighted were on top of a billboard which also had text. But I was lucky and the text was on top of essentially empty space, and the content-aware fill technology was able to do a decent job at removing it.

Human artists are contextually aware of so many things that we take for granted. We are evolved creatures that have millions of years of instinctual knowledge that appreciates certain things that AI have no chance of understanding without being taught.
Don't get me wrong, AI can learn and imply miraculous things with finding patterns and connections in data that humans don't because of how time consuming and unintuitive it would be for us.
We recognize a lazy eye instinctually, but might not be able to effectively communicate it to a machine.
We know to appreciate symmetry in faces. A machine doesn't know that more symmetrical = more betterer.


In this image we know it looks wrong in... many many ways, but lets just go with these two things. A machine doesn't know that the eyes are looking in two different directions, and even if it did, it doesn't know that that looks bad.
For the arm, the machine doesn't know that its drawing an arm, and it did, doesn't know that it looks off.

But I do think that the tools will one day exist to be able for them to do so, but it will take tools built by humans to teach the machine artists context and judgement.

A completely separate AI could look at this image and be able to identify that it as a picture of an asian woman.
It would be able to identify the sky, the green bushes behind her, her blue shirt, all of her facial features, her hair/neck/arm/hand, etc.
That identifier AI could then chop the image up, with it's associated meta-data, possibly create a 3D-wireframe if the image has depth, and feed it back into the AI artist's workspace for iterating.
(Edit: not all that different than myself reading my own forum submission and identifying that my grammar and wording is dogshit with many mistakes)

The AI artist could then know that, "Oh, I am drawing an asian woman with long strawberry blonde hair", and then know to automatically do some extra work before presenting a final draft to the human.

The only way for it to know if what it has drawn is anatomically accurate, would be to zoom out, and skeleton the entire asian woman. Upon doing so, the machine could then look at the original draft's meta-data, and determine that it is horribly flawed.

But it just feels like machine art is extremely primitive right now. Many many tools will need to be created to support it. I don't think simply relying on advanced pattern recognition will ever be able to fill in the blanks of the missing context

 

[Pervtron3000]

And honestly I think that with time the concerns over theft/legality will sort of die out when far more mature AI tools exist, because I don't think purely-AI created images will stand a chance against human creators with AI tools when it comes to non-abstract art works.

I will say that AI may continue to be a problem/preference when it comes to abstract works. Midjourney can abstract like a chaos god.

But for non-abstract works, the

You must be registered to see the links

will slay the robot.

Heres how I think art process will evolve with powerful AI tools, but I have to admit that I am ignorant of how any art workflows work so don't get mad at me, I just don't know.

1) Inspiration
But I believe that, midjourney can serve the role of inspiration. Inspiration comes from everywhere. Could be God's works, other people's works, sounds, smells, drugs, your own imagination. Its foolish to think that an AI couldn't legitimately also create inspiration.

2) 3D wireframe
Midjourney may not even know wtf it created, and thats the beauty of it, midjourney can indeed create. Midjourney can create feverdreams, which an identifier AI wouldn't be able to usefully catalogue it's meta-data.
Thus the human needs to have an idea of what is to be created. We already do this with midjourney prompts, but a higher resolution idea will result in a higher quality outcome.
There will be tools for extracting/creating/editing 3D wireframes of scapes that humans would then make corrections to.
Humans can establish a size to scale somehow. Maybe place a dummy human somewhere in the scape, and scale the size of the human, so that the tools can know.. how big a chair should be, a table should be, the fruit bowl on the table should be, the fruit in the fruit bowl should be.

3) Human organized meta-data
Once a 3D scape is established, the human can start identifying different areas, objects, environments, etc about it.
These things will be used to iterate over.

4) Scape aware, machine-iterated content generation
Heres where you could use tools to create another inspirational work, but within the parameters of the 3D scape.
From here on, the human will probably iterate from #2 through #4 many times building the scape and refining the meta-data, and something like midjourney wouldn't be sufficient. From here on, the AI would need to know a lot more about what it is drawing. Heck, the AI may even need to prompt YOU for some help.
These tools will be aware of anatomy and skeletons, of eyes, of hair, of motion, of day and night, of gravity, etc.

And I think these tools will probably have licensing in mind when you tell it what styles to draw from as inspiration.
The "AI stealin my art" discussion will evolve throughout the rest of our lives and after our deaths.
The "Prove that your AI didn't learn from me" suggestions, I don't even know if thats possible. An AI's knowledge isn't a database of it's past.
And would we apply the same principle to AI identifiers? Probably not. Technologies that would detect if your art was used in an AI's training would likely also need to be trained with your art. Same training data, different application.

I don't know what should be done other than that I know we should take into account the livelihoods of the humans that are still alive and currently invested in their life's work. I want them to be provided that which they have worked hard for and deserve compensation for.
Whether or not they will get to decide if AI can learn from their art, well, we'll see.

 

[SloppyTurdSlapper]

Might not be part of the subject but i thought to let you know, that there are going to be webbrowsers with integrated AI features. Is this a privacy concern or could AI potentialy increase our privacy just like in iron man?

 

[♍VoidTraveler]

Might not be part of the subject but i thought to let you know, that there are going to be webbrowsers with integrated AI features. Is this a privacy concern or could AI potentialy increase our privacy just like in iron man?

Probably going to be the usual:
Dude X starts abusing AI for his nefarious purposes, dude Y appears and starts using AI to counter dude X.

^This archetype can be applied to many things, such as: Spam, advertisement, tracking, etc.

 

[lfg_dev]

Are there any models trained on Daz3D characters ?

 

[XcentY]

Just to share

You must be registered to see the links

 

[FamilyDev]

Is anyone here up to date on the cutting edge of machine art?

Iterating on what I said on page 2, if (WHEN) a company like Adobe were to fully embraces machine art and empowers it with more context and awareness, and make things super easy for creators to touch-up and iterate on the works, I think we'll have a golden age of higher quality sex games.

Let me give a very very basic example of touching up an image for something specifically made to remove text overlays and watermarks

You must be registered to see the links

You must be registered to see the links

Apologies if those two images don't work, maybe these will instead

You must be registered to see the links

You must be registered to see the links

This is a machine learning tool that is purpose built with contextual knowledge about what text and overlays are.
Its quite possible that such a tool could be confused if an area highlighted were on top of a billboard which also had text. But I was lucky and the text was on top of essentially empty space, and the content-aware fill technology was able to do a decent job at removing it.

Human artists are contextually aware of so many things that we take for granted. We are evolved creatures that have millions of years of instinctual knowledge that appreciates certain things that AI have no chance of understanding without being taught.
Don't get me wrong, AI can learn and imply miraculous things with finding patterns and connections in data that humans don't because of how time consuming and unintuitive it would be for us.
We recognize a lazy eye instinctually, but might not be able to effectively communicate it to a machine.
We know to appreciate symmetry in faces. A machine doesn't know that more symmetrical = more betterer.

View attachment 2755266
In this image we know it looks wrong in... many many ways, but lets just go with these two things. A machine doesn't know that the eyes are looking in two different directions, and even if it did, it doesn't know that that looks bad.
For the arm, the machine doesn't know that its drawing an arm, and it did, doesn't know that it looks off.

But I do think that the tools will one day exist to be able for them to do so, but it will take tools built by humans to teach the machine artists context and judgement.

A completely separate AI could look at this image and be able to identify that it as a picture of an asian woman.
It would be able to identify the sky, the green bushes behind her, her blue shirt, all of her facial features, her hair/neck/arm/hand, etc.
That identifier AI could then chop the image up, with it's associated meta-data, possibly create a 3D-wireframe if the image has depth, and feed it back into the AI artist's workspace for iterating.
(Edit: not all that different than myself reading my own forum submission and identifying that my grammar and wording is dogshit with many mistakes)

The AI artist could then know that, "Oh, I am drawing an asian woman with long strawberry blonde hair", and then know to automatically do some extra work before presenting a final draft to the human.

The only way for it to know if what it has drawn is anatomically accurate, would be to zoom out, and skeleton the entire asian woman. Upon doing so, the machine could then look at the original draft's meta-data, and determine that it is horribly flawed.

But it just feels like machine art is extremely primitive right now. Many many tools will need to be created to support it. I don't think simply relying on advanced pattern recognition will ever be able to fill in the blanks of the missing context

The eyes are already very easy to correct, here is an example. But it doesn't work with the hands. We have to hide them behind the scenes) The only thing the AI can't do is the lines from the pupil. They can already be added in Photoshop.

 

[nebtop]

Look at this lazy people happy to find a way to generate art made by an Ai using stolen data. "I don't need skills anymore and ruin other creators live with boring art that I didn't made that looks like 500k similar accounts while shadowing the same artists the Ai stole from, yay"

The audacity of this guy to say this after asking people to pay for character "edits".

 

[XcentY]

check this video. It's definitively improving

download link

You must be registered to see the links

 

[koiflo]

The eyes are already very easy to correct, here is an example. But it doesn't work with the hands. We have to hide them behind the scenes) The only thing the AI can't do is the lines from the pupil. They can already be added in Photoshop.
View attachment 3022263

Any tips for getting tongue and teeth correct?

 

[Fate stay]

AI is definitely improving

 

[Melkor99]

The eyes are already very easy to correct, here is an example. But it doesn't work with the hands. We have to hide them behind the scenes) The only thing the AI can't do is the lines from the pupil. They can already be added in Photoshop.
View attachment 3022263

Have you tried using controlnet and depth maps to get the hands right. I've had some success there, but TBH in a lot of cases, unless the hands are important to the scene, it's easier to just hide 'em.

 

[wrathgazer]

whut ai are you all using ?

 

[FamilyDev]

Any tips for getting tongue and teeth correct?

Teeth and tongue are always VERY difficult. The only thing that will really help 100% is to make a ControlNet SoftEdge mask and draw the teeth, tongue yourself and generate a mouth using the mask in Inpaint. This is the only way SD can do it normally.

Have you tried using controlnet and depth maps to get the hands right. I've had some success there, but TBH in a lot of cases, unless the hands are important to the scene, it's easier to just hide 'em.

Sorry for not answering, if it's still relevant, the ControlNet retains the shape of the hand well, but with a large denoise strength, everything is still bad. In fact, I found Lora, which really works, although not always.

You must be registered to see the links

whut ai are you all using ?

I'm using Stable Diffusion.

 

[Melkor99]

So, anyone have advice on interactive drawings, IE, guy boinking girl? Because even with Loras from Civitai, that seems very hit and miss, especially if the characters are supposed to be someone specific.

 

[XcentY]

So, anyone have advice on interactive drawings, IE, guy boinking girl? Because even with Loras from Civitai, that seems very hit and miss, especially if the characters are supposed to be someone specific.

1) build good prompts
2) build negative prompts
3) give a name to the person you want to create (it's like you give birth to a new person)
4) be precise
5) use embeddings
6) use controlnet
7) have good settings especially if you don't want to let AI have too much freedom which seems to be your case