Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump

[Puff]

I can't seem to make a good Kanye as a Peanuts character. It doesn't seem to like to do people as that style.

Spoiler: Early attempts

Rei Ayanami with same prompts for reference

 

[Catgirl Tyranid]

I can't seem to make a good Kanye as a Peanuts character. It doesn't seem to like to do people as that style.

Spoiler: Early attempts

View attachment 4027965View attachment 4027968

Rei Ayanami with same prompts for reference

View attachment 4027962

Have you tried referencing Franklin in the prompt?

 

[Ahriman]

I bring you more seething

Brian Serway on Instagram: "To be more specific, f*ck AI art generators like Lensa. You may be someone who has used the app for yourself - and if you are - know that what I am examining here is in no way an attempt to shame you… I have become more t

December 8, 2022 - 25K likes, 868 comments - Brian Serway (@bserway) on Instagram: "To be more specific, f*ck AI art generators like Lensa. You may be someone who has used the app ..."

www.instagram.com

To be more specific, f*ck AI art generators like Lensa. You may be someone who has used the app for yourself - and if you are - know that what I am examining here is in no way an attempt to shame you… I have become more than a little obsessed with that AI Pokémon generator (Summer can attest to this). But I do hope you will pause what your doing for a moment and consider the very real implications AI generators like Lensa can and will have on artists..

Stable Diffusion - the AI image generator that powers Lensa - was trained on over 2.3 billion captioned images, many of which are watermarked & copywrited; those images also include thousands of images from apps like Pinterest, DeviantArt, Flickr, Shutterstock, Getty, and more. The consequence of this is that the pool of content Lensa uses is made from the original work of artists who did not consent or opt-in to having their work used in this way. In many cases, artists signature lines, and even their signatures themselves, have been spotted in these AI creations.

Why pay a real artist hundreds of dollars for a portrait when Lensa can give you 50 for $7.99? And do it using the style of very real artists who have spent years refining the app. How will this transform our modern art world and other creative industries in the future?

This post is here to spark discussion & deeper knowledge on the topic - I would love to hear your thoughts in the comments & help me get a discussion brewing. I have my opinions… which are in many ways very personal and emotionally driven.. but they are also still evolving with new knowledge and perspectives - so I want to hear yours! Also, apologies for baiting you with the graphic.. but it’s the only way our other computer overlords ensure you see this. If you would like to learn more about Lensa, there’s a link to an article in my profile you can check out too!

#lensa #aiart #traditionalartist #digitalart #brianserway

 

[Dread First]

Hm... Today I will put this in the Stable Diffusion Web prompt.

 

[SomeBirds]

Not sure about that. Im sure your kiwi model would be fine. Ive uploaded a few and nothing has happened yet. I didnt even know they review them

I thought they did but no problems with uploading the kiwi model.

Kiwis

I'm not some special exemption and if there's anything wrong with the model it would be news to me, but take the good advice in the thread and do a safety scan if you DL it because you should be doing them no matter what.

This is a 100% kiwi bird trained model merged with the regular base model, so that means it's 30% kiwi birds and should be used as a stand-alone. If you merge it the amount of kiwi-ness will be halved and it's unlikely to give good results. If someone wants the 100% model for merging yell at me and I'll see that it gets added.

Have fun!

Also, next time I will be back to talk more about DreamBooth and the results from feeding it 1,536 images. I want to generate more results before explaining what's going on.

 

[Pocket Dragoon]

I finally got around to playing with Midjourney; here's the best from my first three attempts:

"Slavic James Cameron remake of Schindler's List"
(I'm impressed with the juxtapositions & atmospheric quality).


Impressionist painting of Titanic leaving port for the last time:


Vintage propaganda poster of Ukrainian War Racoon and burning tanks in winter.

 

[FrostedDix]

Noodie ai

Kickstarter for a uncensored, open source unstable diffusion already hit half the goal in hours.

I'm hoping this team doesn't buckle to investors eventually like every other one so far

 

[Pocket Dragoon]

Then ran out of freebies before I could upscale #4 of that last set.

This is what I was playing with.

style of alex maleev, realistic, confederate soldier on horse, dramatic background, cloudy, strong shadows, winter colors, detailed realistic, beautiful colors, single panel.

I still hate discord though; if they made midjourney standalone, I'd consider paying for it.

 

[SomeBirds]

Pumped 1,536 unique images into the furry model. Since each image gets mirrored it wound up being over 3,000 total images. Almost nobody wants to talk about what they're doing to train their AI, but I recall seeing that the furry model was trained on far fewer images than the anime model or the base model, so it seemed like it would be easiest to see results in.

If you could curate enough images it's possible that this could be replicated for any model. It will probably take a few months, but if we can process enough images there might be a more generalized version with closer to 6-8k images involved.

Initial testing was 10 very simple prompts. These were run through an unmodified version of the furry model with a limit of 3 re-rolls. What I'm posting are the top 2 results for each prompt. The only negative prompt was 'bad anatomy.'

Spoiler: Prompts

A pony with a colorful mane, wearing an ornate dress, standing in a sunlit meadow, my little pony, detailed, hyper-realistic, highly detailed face, illustration, digital painting

A peacock wearing a suit standing on a city street, detailed, hyper-realistic, highly detailed face, illustration, digital painting

A horse wearing ornate red and gold robes kneeling in a temple, detailed, hyper-realistic, highly detailed face, illustration, digital painting

A duck wearing a hoodie standing on a city street, detailed, hyper-realistic, highly detailed face, illustration, digital painting

A tiger woman wearing an embroidered dress walking on the beach, detailed, hyper-realistic, highly detailed face, illustration, digital painting

classy pony in a tweed suit, english countryside, manor house in the background, detailed, highly detailed face, hyper-realistic

lion wearing a tweed suit and fedora, lounging, leaning against a brick wall in a dark alley, detailed, highly detailed face, hyper-realistic

badger wearing a kilt and tartan, holding a sword, standing proudly, rugged mountains in the background, detailed, highly detailed face, hyper-realistic

parrot druid, wearing dark robes in a dimly lit forest, detailed, highly detailed face, hyper-realistic

owl wearing ornate armor, ornate helmet, standing proudly in a forest of surreal giant redwood trees, correct hands, detailed, highly detailed face, hyper-realistic

badger wearing an ornate silk dress, sitting on the bank of the nile river, palm trees and pyramids in the background, life-like, highly detailed face, hyper-realistic

Results from Epic Furry_4

Spoiler: Results

Homebrew/Modified Epic Furry

Spoiler: Results

It seemed like the backgrounds improved more than the animals. The biggest difference is that requesting a pony will return a MLP style pony. If you want a horse you have to ask for a horse even without any MLP related tags.

It's probably not going to put any furry artists out of business yet, but as a test to see whether a large chunk of curated art could get better results than a spider grabbing bulk images it feels like a success.

 

[DeadwastePrime]

ryan gosling if he was fat i guess lmao

 

[Allanon]

I bring you more seething

View attachment 4032588

Brian Serway on Instagram: "To be more specific, f*ck AI art generators like Lensa. You may be someone who has used the app for yourself - and if you are - know that what I am examining here is in no way an attempt to shame you… I have become more t

December 8, 2022 - 25K likes, 868 comments - Brian Serway (@bserway) on Instagram: "To be more specific, f*ck AI art generators like Lensa. You may be someone who has used the app ..."

www.instagram.com

Mostly just a casual that thinks AI art is cool and has potential, but I thought this has already been disproven? The supposed 'signature lines' that the AI generates is not any artist's signature but the AI aping the concept of adding signatures and being unable to write words in any legible language, same with watermarks. I'm pretty sure nothing of the original art remains in any AI art, the whole 'database' is only used to essentially teach the computer what things look like, so it can then copy that with its own effort.

 

[Pendulous Dong]

This famous piece by engraver Robert Liefeldus (1489-1561) demonstrates a knight on horseback along a narrow path. Unusually, Liefeldus's inability to draw feet (commonly hidden by perspective or objects in the foreground) is on full display in this piece.

 

[AnOminous]

I'm pretty sure nothing of the original art remains in any AI art, the whole 'database' is only used to essentially teach the computer what things look like, so it can then copy that with its own effort.

Someone should tell these bitches that's what's called "transformative."

 

[Pocket Dragoon]

Some of my favorites this week:
Miyazaki Star Wars


Oil-Rig psychological horror disaster movie:

 

[Abeezo]

A post from a close net buddy who used to like my own AI instagram acc but has a spergout today all of a sudden becuz the tism
Funniest thing was he liked all of those posts but then unflowwled and deleted all of his comments on those posts the same day he posts this.

 

[Stephanie Bustcakes]

I tried to make le pol face, but I just couldn't get good results. So instead...Who else but Quagmire?!

 

[Potatoherder]

Noodie ai

Kickstarter for a uncensored, open source unstable diffusion already hit half the goal in hours.

I'm hoping this team doesn't buckle to investors eventually like every other one so far

Don't get your hopes up. It's been fully infected by troons. They also haven't ever released a model and the most they have been involved with one is giving funds to Waifu diffusion devs for compute.

Embarrassed_Stuff_83 comments on Unstable Diffusion has reached their…

archived 12 Dec 2022 11:06:38 UTC

archive.vn

 

[Woe-B-Gon (TM)]

Don't get your hopes up. It's been fully infected by troons. They also haven't ever released a model and the most they have been involved with one is giving funds to Waifu diffusion devs for compute.
View attachment 4054269

Embarrassed_Stuff_83 comments on Unstable Diffusion has reached their…

archived 12 Dec 2022 11:06:38 UTC

archive.vn

And yet they got, if you'll forgive my use of their jargon, the most cishet looking white guy on the planet - one who also seems to believe in marriage - to do their pitch vid. Hmmmmm....

 

[AnOminous]

And yet they got, if you'll forgive my use of their jargon, the most cishet looking white guy on the planet - one who also seems to believe in marriage - to do their pitch vid. Hmmmmm....

Maybe even troons are realizing putting a troon up front of any kickstarter scam is enough to scare people off.

 

[Drain Todger]

Damn, Midjourney v4 is amazing.

Spoiler: Some Midjourney v4 Stuff

"tsutomu nihei, horse, equine, cyborg, robotic, armored core, zoids, bionicle, plating, armor, futuristic, biopunk, armor, feral, military, weapons, aggressive, muscular, cyberpunk, black and red with gold trim, hump, bulky, armored, heavy, heavy, heavy, thick, thick, thick, --v 4"
"sonic the hedgehog --v 4"

"battle angel alita --v 4"

The UNSC's had some budget cuts, so the Chief ordered his new armor off Wish.com:

Wait, what if I actually used that as the basis of a prompt? Let's try, "master chief, but the UNSC has had budget cuts so now he orders his MJOLNIR armor from China off Wish.com, and it looks really cheap and fake --v 4"

Wait, those are actually awesome looking. Let's try it with cardboard.

"master chief, but his armor was designed in the 1980s by the Soviet Union --v 4"

Jesus Christ, my fucking sides are in orbit:

I asked Midjourney v4 for "70s science fiction book cover painting of a guy in a desert wearing a space suit with a retro laser pistol, moody --v 4"

This one, same prompt, just without him being in a desert:


Midjourney is GREAT for doing architectural visualizations:


"A shopping mall by Le Corbusier":

"A house in concrete by HR Giger":

I had a long chat over Zoom with some guys in India, and I guarantee you, AI art generators like these are already being used for architectural concept art. Midjourney has very likely already been used to design a building. Why hire someone to do this in Photoshop, Max, and/or AutoCAD if you don't even know what you want, yet? This is so much faster and cheaper, it's not even funny. I could have this thing spew out literally hundreds of pieces of concept art in the time it would take for someone to do one detailed architectural sketch.

Your Therapist: "Swolenic isn't real, he can't hurt you!"

Spoiler: Swolenic

Wait, what if I prompt it to do a Sonic-Pikachu hybrid?

Spoiler: Sonichu?

I regret everything.

Okay, then. How about Stable Diffusion Twilight Sparkles?

Spoiler: Twilight

Yep, that looks about right.

Okay, what about in armor?

What about with a crown?

Holy shit, that's a big crown.

It's crazy how the AI "knows" how a character's face should deform to make different expressions:

Not only that, it can transplant features onto characters that they never had, before. Skinny, Fat, Cyborg, whatever.

What about a Midjourney v4 Twilight Sparkle?

Spoiler: Terrifying Horse Demon

Oh god.