AI [AI Art] - Show Us Your AI Skill *NO TEENS*

[Deleted member 395648]

since posting these to the game art thread they immediately got deleted so imma leave them here
trained LoRAs with game renders - will be posting onto civitai soon once I iron out some niches
(for now to counter the output to looking too Daz3D I slip in a few real lookalike people into the training data,
still finding that sweet spot)

EDIT: They have been released!

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

​

 

[PandaRepublic]

since posting these to the game art thread they immediately got deleted so imma leave them here
trained LoRAs with game renders - will be posting onto civitai soon once I iron out some niches
(for now to counter the output to looking too Daz3D I slip in a few real lookalike people into the training data,
still finding that sweet spot)

View attachment 2761896 View attachment 2761895 View attachment 2761899 View attachment 2761906 View attachment 2761904​

can u share the LoRA’s plzzz

 

[Deleted member 395648]

can u share the LoRA’s plzzz

like I said, will be posting onto civitai soon! Probably in a few days

 

[felldude]

since posting these to the game art thread they immediately got deleted so imma leave them here
trained LoRAs with game renders - will be posting onto civitai soon once I iron out some niches
(for now to counter the output to looking too Daz3D I slip in a few real lookalike people into the training data,
still finding that sweet spot)
​

I've used DAZ and its prerequisites for 20 years and it is still my favorite render engine even though blender is objectively better.

However trying to educate the general user of DAZ reminds me of quotes from two of the most famous/infamous books

'Give not that which is holy unto the dogs, neither cast
ye your pearls before swine, lest they trample them
under their feet, and turn again and rend you.'

'The moment you start arguing with an ignorant fool, you have already lost.'


I would challenge them to identify the program(s) used to make this image. (The open source material I've read noted the FBI and DARPA was concerned about AI images but I guess they have more skill)

 

[felldude]

A mixture of AI and hand painting with a custom lora trained on the works of Boris Vallejo

In my opinion it highlights the talent of illusion of detail and the difference of creating every detail from your own mind vs AI enhanced mixed medium.

 

[Deleted member 395648]

I've used DAZ and its prerequisites for 20 years and it is still my favorite render engine even though blender is objectively better.

However trying to educate the general user of DAZ reminds me of quotes from two of the most famous/infamous books

'Give not that which is holy unto the dogs, neither cast
ye your pearls before swine, lest they trample them
under their feet, and turn again and rend you.'

'The moment you start arguing with an ignorant fool, you have already lost.'


I would challenge them to identify the program(s) used to make this image. (The open source material I've read noted the FBI and DARPA was concerned about AI images but I guess they have more skill)

View attachment 2762008

Thank you for this! The worst part about that whole ordeal is that I actually didn't even caption anything, not leaving a single word, did not engage with the AI hating crowd because I talk enough about AI in my day job I don't feel the need to convince some online stranger that I know what count as creativity. But because the simple act of posting stirred up some insecure people yelling to shove that up my ass so admin probably find it easier to just delete "both" sides even though I didn't even engage with them (tbf I did left a laugh reaction to their comment but that's about it )

 

[Microtom]

AI haters will have to start liking it because everything will be AI, and I really mean everything. Your 3d models will be AI generated. Your renderer will use AI. Your animations, voice, story, etc. Renders will all look perfectly real like photographs. It'll be soon too. In 5 years the world will have changed completely.

Models are trained on images. All the information they have are visual and linguistic, and they make associations between them. Nothing prevents models from associating additional information. Like spatial information you'd get from photogrammetry. The model could know exactly how a hand is in space, so when it is asked to generate one in an image, it always does it perfectly. Adding more information to models isn't difficult, so that's why I say it'll happen fast.

People don't realize how close we are to a general AI. Once AIs have a statistical understanding of space, then they can start understanding the physical properties and function of matter and objects. Then you pretty much have an AI that has full understanding of everything, you just needs to give it a memory of tested knowledge as reference to correct its hallucination.

 

[felldude]

There is a practical limit on what AI can do, that is unless IBM can prove their cubit, we measure the performance in CLOPS quantum processors can actually do something of value.

For the USA its around 1500 watts, for the UK its 3000watts, that is to say the machine can't exceed the average wall outlet.

For reference the

You must be registered to see the links

that NVIDIA used to turn minecraft into a real world in real time was trained on a server using 8 32GB V100 Graphics Cards being feed by 32 processors (512 Cores I believe) that is a power draw of over 10000 Watts (Around 42AMPS, about the same as running your stove on clean...and they did it for 4 days)

I do realize that training takes processing power of a exponent of an exponent compared to just running a module but highest ends of AI are out of reach of the average person.

Or the compalation of faces by

You must be registered to see the links

(Don't worry they say it shouldnt be used to train facial identification) the Tensor flow files are 270GB
That is 100x times bigger then your average checkpoint. (It was Trained with 70k images over 2TB)

 

[Lorthard]

 

[devilkkw]

Spoiler: gothic style

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

Spoiler: one colorfull

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

 

[sharlotte]

Shopping with the girls and other images:

 

[felldude]

Not really on point for the site but...

 

[sharlotte]

felldude maybe we should do more of these. It's a great piece! Slightly more wintery and dreary but here are some I generated earlier:

 

[Dagg0th]

New Party members

Dancer​

Necromancer​

Paladin​

 

[felldude]

felldude maybe we should do more of these. It's a great piece! Slightly more wintery and dreary but here are some I generated earlier:

Thanks

I think the old Spade data set was trained for landscapes, It would be interesting to use one that was trained on 50 or 100k images of landscape to do the background's then if you wanted to add a human element do that over it.

 

[Ptah]

AI haters will have to start liking it because everything will be AI, and I really mean everything. Your 3d models will be AI generated. Your renderer will use AI. Your animations, voice, story, etc. Renders will all look perfectly real like photographs. It'll be soon too. In 5 years the world will have changed completely.

Models are trained on images. All the information they have are visual and linguistic, and they make associations between them. Nothing prevents models from associating additional information. Like spatial information you'd get from photogrammetry. The model could know exactly how a hand is in space, so when it is asked to generate one in an image, it always does it perfectly. Adding more information to models isn't difficult, so that's why I say it'll happen fast.

People don't realize how close we are to a general AI. Once AIs have a statistical understanding of space, then they can start understanding the physical properties and function of matter and objects. Then you pretty much have an AI that has full understanding of everything, you just needs to give it a memory of tested knowledge as reference to correct its hallucination.

What is also going quite fast is AI animation, the consistence and lack of stuttering has improved like CRAZY. As I said some months ago, I can't wait for people being able to create films at home, and fuck all movie companies...

Some dirty AI as celebration...

 

[Jimwalrus]

"Blackmail!!"
"Are you referring to me, or my proposal?"

 

[ecobot]

 

[Microtom]

View attachment 2766984 View attachment 2766982

She certainly isn't going to drown

 

[Deleted member 395648]

Hey I have a request can you do one for Our Red String by Eva Kiss. If it's not a burden. THANK YOU!

Hmm, I usually do photorealistic AI content and all my LoRAs training are with photo captions, I imagine any 2D images like drawings, western comics, anime might require a completely different workflow that I might not be awfully familiar with, but I hope you can find someone who do!