/cyb/ - Cyberpunk Fiction and Fact

>>2017
>>2018
>>2019
>>2020
I'm speechless.

>>2021
I see hundreds of pics of this quality on discord every day. I just only bothered dumping some now.

>>2017
>>2018
>>2019
>>2020
Absolutely amazing!
>>2022
More, please!

>>2017
>>2018
>>2019
>>2020
Wow. They look great.
Do you know what model was used to create those images?

Purplesmart AI has been improving:
https://discord.gg/4QDFxbRmjV

>>2024
Program: automatic1111
Model: Astralites Ponyv1
Sampler: Euler a
Prompt: detailed fluffy fur, unicorn, (by taran fiddler and chunie , Michael & Inessa Garmash, Ruan Jia, Pino Daeni), volumetric lighting, ambient bloom, 4K, ((((solo)))), (((golden medieval full armor))), (golden full muzzle helmet with metal ears), (face covered), ((Fire for hair)), holy light

>>2026
Where can I get the Astralites Ponyv1 model?
I tried that prompt without the model and got this.

>>2027
Tested and loos like this is the model.
https://huggingface.co/AstraliteHeart/pony-diffusion
https://mega.nz/file/ZT1xEKgC#Xxir5udMmU_mKaRZAbBkF247Yk7DqCr01V0pDzSlYI0

>>2030
How does the model make art that good when most fandom art is show tracing- I mean vector art?

>>2031
I think that is one of the wonders of Neural Networks. No one knows exactly why it works so good as it does only that it works. But it is hit and misses in depending on seed number.

>>2030
In case people want to know. For that image (that looked best so far and first I ran) used the Prompt by >>2026
>detailed fluffy fur, unicorn, (by taran fiddler and chunie , Michael & Inessa Garmash, Ruan Jia, Pino Daeni), volumetric lighting, ambient bloom, 4K, ((((solo)))), (((golden medieval full armor))), (golden full muzzle helmet with metal ears), (face covered), ((Fire for hair)), holy light
>Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 3081589709, Size: 512x512

>>2034
I've heard of training girls, but this is ridiculous!
What is this, the next Kantai Collection spinoff?

>>2034
>trains and anime
Add ponies to the equation and we'd have enough autism to go to Mars and back.

>>2037
And Sonic. Autists love Sonic for some reason.
Maybe they wish they were Sonic? Maybe Tails the shy small genius child who wants to be cool like Sonic appeals to them? Maybe they just like fast things.

>>2040
I didn't make it, but thanks.

>>2041
Where's the source?

>>2039
It's good, but it makes me dizzy.

>>2039
Videos like this are further proof that AI generated images doesn't "hurt art", but expands it. The concept of art is being democratized, so that even random people who have no talent for drawing or no time to paint can realize their creative ambitions and silly ideas.
It's said that beauty is in the eye of the beholder. A machine may be able to make immaculate images, but only a human can appreciate them. Likewise, only a human would have thought to use a machine to compile a bunch of anime girls into the Rick Rolled video, and AI helped them do it.

>>2049
Things only have value when they are scarce

>>2052
Art has always been abundant. There's trillions of artworks out there. The fact that unpracticed artists can create their own images with memes just adds to the sphere.
Art didn't suddenly get worse once acrylic paints enabled artists to buy their paint instead of mixing it from the elements.

>>2052
That's only for market value, which is irrelevant. Art's value is subjective.
Just look at the boorus and see how there's tens of thousands of lewd pics of the mane six, but people still get excited when new

>>2053
There aren't. Painting well is hard

>>2054
All value is subjective. That value is based on scarcity. If you have infinite art then all of it becomes boring

There's lots of pics but a tiny fraction of them actually appeal to you. If you can press a button to get a new one perfectly suited to you, you'll get bored of it

EVERYTHING becomes boring if you have an endless supply of it

>>2055
>If you have infinite art then all of it becomes boring
I've seen thousands of conventional pony artworks and thousands of AI artworks in the past couple weeks alone, but I'm not bored at all.
>That value is based on scarcity.
I beg to differ. When it comes to art, the more the merrier.
>If you can press a button to get a new one perfectly suited to you, you'll get bored of it
If I became bored of it, it must have not suited me perfectly.
>EVERYTHING becomes boring if you have an endless supply of it
I already have a virtually endless supply of mare porn for free, and I'm not even close to bored of it yet.

>>2052
Things can have inherent value unrelated to their scarcity and how desired they are.
Incredibly rare randomly generated NFTs unlikely to exist don't have any intrinsic value granted by their rarity.

>>2057
>Incredibly rare randomly generated NFTs unlikely to exist don't have any intrinsic value granted by their rarity.
This, regardless of how scarce something is, it's true value is based on one of two things:
1. How useful it is to people
2. It's subjective, sentimental value
AI art is useful, and also enables people to generate artworks that meet their own subjective tastes.

>>2061
Amogüs

>>2064
>A.I. generated
I doubt it. I believe A.I. assisted might be a more accurate term.

>>2066
CUTE

>>2066
Post steamy honeymoon pics pls
Also, checked devil dubs

>>2120
Nice and hot.

>>2150
Pongle 2.0

Does anyone want a grid of 35-50 images made with your chosen prompt? My PC is pretty powerful.

>>2160
Ooooooo. I can try to think of something but from my testing crafting an effective promt is a science in itself so I am not sure what a good promt would be (my testing on my slow computer have not been too successful).
But if you can try to make some statues like in >>2115 (pic 3) it would be awesome. I think they look awesome.

>>2162
I'll need the prompt and model used for those pics. If I just said "(My Little Pony) garden statue" with the default model that came with the AI when I downloaded it I doubt I'll get good results, but I'll give it a go.

By the way, what does "Tiling" do? Whenever I check that box it just makes the output look inferior. Is there some purpose to it I'm missing?

>>2163
The one I had limited success with (but takes 2-3 minutes per image so I was not able to do to much) was:
>pony statue, majestic, alicorn, 4k hi-res, detailed, granite, cracked surface, weathered
>Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 174254891, Size: 512x512

>>2164
2-3 minutes per image? That must suck. How many images can you generate at once before your PC runs out of memory?
One time I left my room with an autoclicker generating more sets of images every minute. By the time I got back, I had a few thousand.

>>2166
>How many images can you generate at once before your PC runs out of memory?
I can only run one at a time, and at times have to close other apps that uses memory in order for it to run.
Are you using the Stable Diffusion Model or Waifu model instead of the Pony model? The Pony Model is good and you should try it
https://huggingface.co/AstraliteHeart/pony-diffusion
https://mega.nz/file/ZT1xEKgC#Xxir5udMmU_mKaRZAbBkF247Yk7DqCr01V0pDzSlYI0

>>2167
Yep, that was the problem. It's turning out a lot better with the pony model.

>>2168
Nice

>>2167
Pony model loaded.
Here are some new AI generated pony statues, image sets 3 and 4 were made with the seed you posted instead of a random seed.
Had to split them up because trying to post the next 4 intact broke some kind of limit. "Request Entity too large".

>>2172
Strange that the first image in the set using same seed and prompt didn't yield the same image as I got as it should have done that. Subsequent images when given a seed should give previous_seed+1 when running batches.

>>2174
Maybe there's a second invisible seed somewhere. All the settings were accurately replicated.

>>2176
Strange. For reference the arguments I use to start Stable Diffusion is:
>--ckpt pony_sfw_80k_safe_and_suggestive_500rating_plus-pruned.ckpt --lowvram --opt-split-attention
Other than that I have the standard settings (no scripts or face fixes or anything) and standard txt2img

>>2177
>lowvram --opt-split-attention
I was using those exact settings except for that part. They just reduce the load on your computer and have no effect on the resulting image, right?

>>2178
They should just reduce the memory used and not change what is produced. Just posted full list of arguments in case I'm wrong on that. It was just strange that you didn't get the same image using same prompt and seed.

>>2179
Was Restore Faces turned on?

>>2180
No, none of the extra features are turned on on my end when I generated.

>>2179
>It was just strange that you didn't get the same image using same prompt and seed.
You see, these AI's no longer "mash" images together with the hope of producing a coherent pic.
They basically emulate human creativity.
Artists need external inputs as well. They consciously and unconsciously absorb bits and pieces from every other artwork they may have seen before. Sometimes it may even be apparently unrelated life experiences.
Regardless, when artists draw inspiration from other art sources. They are pretty much doing the same thing that AI's do.
It's impossible to trace back the source pics. What AI's make is about as original as what any human could've conceived.

It makes sense you don't always get the same results, even when all things are equal.

>>2182
>It makes sense you don't always get the same results, even when all things are equal.
Running same seed and prompt I get same image every time. Also running others prompt and seeds (I seen people post) I get same image they get when using the same model they used.

>>2182
I don't think the AI is updating its own model with data from the pictures it generates.
I hope it isn't. I don't want my AI to give itself dementia by using its garbled output as new input.
By now its garbled output probably outnumbers the number of images it was trained on.

Does anyone have any good images? I tried doing a cursory search on ponerpics and the first few pages were horrid dogshit.

The era of abstract merchants has reached a new golden age.
Merchants are being produced faster than the ADL could ever hope to flag them.

>>2234
How do you make these?
I've got Stable Diffusion working, what's the secret?

>>2235
Idk. I got them from Discord.
If you find out any methods, do share. I want to mass produce merchants.

>>1884
That's not what I meant when I said I wanted racks on racks.

>>2241
Racks on racks on racks on racks. Stacked rat stacks in hats on mats.

Pixiv has lost its collective mind recently (more than before).
It's impossible to use the site without being recommended pics of [i]pregnant toddlers!

>>2248
Oh fuck. Don't some models use data taken from pixiv or anime booru sites?

>>2250
>you will never haiy her lcat or her spry aliet hany
Why live?

I got the new Everything V3 model.
It seems more varied than the previous smaller Waifu model I was using.
But it also loves to wash out the colours of anything I make with it. So I often have to saturate this desaturated art output until it looks right.

"Twilight Sparkle"

>>2273
Nice. Not knowing prompt but adding "(unicorn)" to prompt and "(text), devil" to negative prompt might fix some of the devil horns and and text. Love how AI art is evolving with better and better models.

>>2275
Wow. It looks fantastic.
Is this using upscaling and inpainting or...? if you know.

>>2274
Speaking of AI evolution how do I update my installation of Stable Diffusion? I heard the new version has no limit on text input and is better at drawing hands.

>>2277
The way I do it is that I used git to get the initial code and run an git update in the bat file used to start SD.

So initially do an
>git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui

Then in bat file I do
>...
>set COMMANDLINE_ARGS=--ckpt pony_sfw_80k_safe_and_suggestive_500rating_plus-pruned.ckpt --lowvram --opt-split-attention
>
>cd stable-diffusion-webui
>git pull
>call webui.bat
>cd ..

Wtf is he doing to edit the AI generated art and get new ones like that? https://m.youtube.com/watch?v=aFnGO74KsF0

>>2281
I assumed he did, but looked at description and he is using the Krita plugin for Stable Diffusion that runs an img2img with prompt. It can take basic input and you have to tell it what it should draw and weight how much it should resemble the input image. I am not fully sure how the weights should be but I think the Krita plugin is fairly easy to use.
First get Krita https://krita.org/
Then install the Krita plugin https://github.com/sddebz/stable-diffusion-krita-plugin

You can do the same in Stable Diffusion webui directly by going to the img2img tab.

>>2281
>>2282
I see that he also uses inpaint to fix areas.
I don't have full knowledge on how to use it effectively as my computer takes forever to generate a single image so I can't really play around with it without it taking forever.

>>2283
Have you tried using that setting where your generator makes the image tiny and then scales it up?

>>2287
I have not tried upscaling or any of the fancy options as the biggest I can generate is 512x512 and even that is pushing my card to the limits. And it takes a couple of minutes to generate an image. So experimenting with settings and what it does is an exercise in patience and remembering what I did half an hour ago and if my changes actually had an impact.

>>2288
Damn, that sucks. Back when I had to make do with a piece of shit laptop that would take multiple minutes to boot up and do basic tasks like open the file explorer or open a webpage, I ran Fallout New Vegas at barely 10 frames a second even with the lowest graphics settings and as few mods as possible. The game was borderline unplayable especially when combat started so I had to rely on companions basically doing all combat for me. The lag didn't fuck their attacks up as much. Even the fucking word processor lagged with that thing. A word processor!