Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Robsoie]

I had never heard of Stable Diffusion, but the "open source" got me interested as unfortunately most of the AI "artwork" generator are closed source and more or less heavily censored that even what is normal attempts (i mean obviously not involving any kind of porn words) sometime fall wrongly into their censoring crap .
Looks like whoever code those things would have had dozen of heart attacks at once if they entered an actual art museum i guess.

From googling it seems there a beta of that Stable Diffusion here :
https://huggingface.co/spaces/stabilityai/stable-diffusion

From all the art generators, i think the worst offender in the censoring department going silly is the "Wombo Dream" thing in which it's hard to generate anything that is involving a woman without their censor hammed thinking despite there's nothing not "PG" in what you wrote triggering whatever shit algorithm they're using instead of generating what should be a normal artwork on every PG websites around.
That's probably not helped by the fact those generator often ignore half of the words you enter, maybe that's what does not help in their censoring algorithm

[TamerVirus]

Alright if y’all are interested in Stable Diffusion and if you have a powerful enough computer I recommend the NMKD gui for ease of use and installation. Local install, no censors, nada

The most popular and developed GUI people use is the AUTOMATIC1111 gui. It’s updated frequently and has all the new bells and whistles. That’s the gui that got accused of stealing code btw. (NovelAI got accused of taking some of AUTOMATIC’s code too). It requires some knowledge of git and whatnot but it shouldn’t be too hard to install.

There’s a bunch of online options regarding using Stable Diffusion as well

I think Stable Diffusion and imggen is big enough to warrant its own thread

[Strongpoint]

I had never heard of Stable Diffusion, but the "open source" got me interested as unfortunately most of the AI "artwork" generator are closed source and more or less heavily censored that even what is normal attempts (i mean obviously not involving any kind of porn words) sometime fall wrongly into their censoring crap .
Looks like whoever code those things would have had dozen of heart attacks at once if they entered an actual art museum i guess.

From googling it seems there a beta of that Stable Diffusion here :
https://huggingface.co/spaces/stabilityai/stable-diffusion

From all the art generators, i think the worst offender in the censoring department going silly is the "Wombo Dream" thing in which it's hard to generate anything that is involving a woman without their censor hammed thinking despite there's nothing not "PG" in what you wrote triggering whatever shit algorithm they're using instead of generating what should be a normal artwork on every PG websites around.
That's probably not helped by the fact those generator often ignore half of the words you enter, maybe that's what does not help in their censoring algorithm

For stable diffusion, I use https://colab.research.google.com/drive/1jUwJ0owjigpG-9m6AI_wEStwimisUE17 but there are many other colabs, with different options and tools

It is one of my major destressing tools in the current world: making AI draw silly stuff. Helps to keep my sanity

[King Zultan]

I have used said AI to generate exciting new Scoops Novel thread titles. My prompt was all of their thread titles from the last 3 pages as of this post. Morpho preset (i.e a blank slate, with no settings or biases curtailing pure, unadulterated word salad).

Damn some of those feel authentic, it's almost as if that's where he gets the ideas for his threads.

[Robsoie]

Alright if y’all are interested in Stable Diffusion and if you have a powerful enough computer I recommend the NMKD gui for ease of use and installation. Local install, no censors, nada

Looks like it's running on the GPU and is only compatible with NVidia ones, sad for me and my AMD crappy system.
Too bad the other GUI, regardless of code stealing stuff (putting code in open source isn't really the best way to ensure none will take your code), is mostly for people that know the how to with github and compilling things and i'm not there yet

For stable diffusion, I use https://colab.research.google.com/drive/1jUwJ0owjigpG-9m6AI_wEStwimisUE17 but there are many other colabs, with different options and tools
It is one of my major destressing tools in the current world: making AI draw silly stuff. Helps to keep my sanity

Never used one of the colab thingy on google, will have to look into this.

Meanwhile, that stable diffusion thing is really useful for subjects of extreme importance

Spoiler: yeah (click to show/hide)

There's lot of fun potential with those silly pictures made with those tools, maybe someone more knowledgable about the various available ai image gens can get a thread going.

[lemon10]

Holy shit. I saw a picture that was allegedly AI generated, immediately went to downvote it because no way was it AI generated.

But no, it simply *is* that good. A lot of the stuff I found is was just flat out AI generated anime stuff that I can't tell at *all* from real anime stuff. Sure, the stuff I saw was obviously the cream of the cream of the crop that people spent hours generating, but the fact that I can't tell at all for some of them is mind boggling. (Also a lot of the new innovation over the last week or two is due to the Novel AI program leak, which Ima try out shortly myself).

A bunch of it is really good and contains crazy details. I mean, its *been* good for a while, but this stuff is a complete game changer.

Doing it well obviously requires you to learn some stuff and put in some work, but its many OOM easier then spending years and years learning how to draw art.

Obviously given time normal artists will use it as a tool to make great art much easier and quicker, but "anyone can make real art without a real time investment" is going to be a seismic shift in art that's gunna cause a shitload of problems for them as well.
---
I would share some links of these pictures, but uh... the proof thread I found was on 4chan and thus not exactly something to share here.

[Strongpoint]

Obviously given time normal artists will use it as a tool to make great art much easier and quicker

Some artists already do. And in a few years, it will be very standard practice. Also, many low-level artists will lose their job.

For many tasks, it will be much easier to ask AI to generate a few thousands option and pick one you need.

[KittyTac]

I had never heard of Stable Diffusion, but the "open source" got me interested as unfortunately most of the AI "artwork" generator are closed source and more or less heavily censored that even what is normal attempts (i mean obviously not involving any kind of porn words) sometime fall wrongly into their censoring crap .
Looks like whoever code those things would have had dozen of heart attacks at once if they entered an actual art museum i guess.

From googling it seems there a beta of that Stable Diffusion here :
https://huggingface.co/spaces/stabilityai/stable-diffusion

From all the art generators, i think the worst offender in the censoring department going silly is the "Wombo Dream" thing in which it's hard to generate anything that is involving a woman without their censor hammed thinking despite there's nothing not "PG" in what you wrote triggering whatever shit algorithm they're using instead of generating what should be a normal artwork on every PG websites around.
That's probably not helped by the fact those generator often ignore half of the words you enter, maybe that's what does not help in their censoring algorithm

I found an unfiltered Stable Diffusion AI. https://www.mage.space/

[Robsoie]

Very interesting that this version has options working (as the other version had them disabled unfortunately), that's nice, there's even an "enhance" function once you generated an image to have the AI trying to improve it.

[lemon10]

I set it up to run on my PC first, but its quite a bit more difficult and my GPU had some issues handling it so I just went over to cloud AI.
---
I've been running NovelAI using Stable Diffusion on a paperspace account for the last week which comes with some pretty beefy free machinery (GPU with 8GB VRAM) and super beefy stuff if you are willing to actually pay eight bucks a month (GPU with 16GB(!) VRAM). I paid, cause 8 bucks a month for unlimited generation and training is pretty sick.

It takes a bit of setup, but there is a very good guide and codebase that does all the heavy lifting for you. Still took me quite a few hours to figure out cause I have never used github before ect, but its easier then setting it up on your own computer as well as safer.

The 4chan /h/ Diffusion thread has a ton of very good very useful resources on everything from setting it up on your computer or the cloud to training and hypernetworks to simple image generation settings and inpainting. I won't link it cause its uh... 4chan on a porn board, but its worth a checkout.

The midjourney AI also seems really sick with a super neat style, but you can't get your own version of it so you have to pay a pretty hefty fee, so I haven't and likely won't use it.
---
Honestly, the stuff you can do, generating images that have never existed before according to your exact specifications on your own computer in minutes(!) or on the cloud like a professional AI learning specialist feels really powerful and intoxicating.

[KittyTac]

Very interesting that this version has options working (as the other version had them disabled unfortunately), that's nice, there's even an "enhance" function once you generated an image to have the AI trying to improve it.

Also, you can put this in the negative prompt to instantly improve quality. "lowres, text, error, cropped, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, out of frame, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck"

Yes, all of it. Put it in and leave it there. Severely reduces the likelihood of body horror.

[TamerVirus]

The 4chan /h/

Dude just use the thread on /g/ or /vg/. Those are blue boards…

[Robsoie]

Very interesting that this version has options working (as the other version had them disabled unfortunately), that's nice, there's even an "enhance" function once you generated an image to have the AI trying to improve it.

Also, you can put this in the negative prompt to instantly improve quality. "lowres, text, error, cropped, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, out of frame, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck"

Yes, all of it. Put it in and leave it there. Severely reduces the likelihood of body horror.

Didn't knew this also was a feature as so far i hadn't logged in when testing it, it's rather amazing as you can then give more direction to the generator.

Spoiler: useful when generating more serious images (click to show/hide)

 

[lemon10]

The 4chan /h/

Dude just use the thread on /g/ or /vg/. Those are blue boards…

I don't use 4chan normally lol, so I have no clue what any of the boards are or that there were other threads about it on other boards (although of course there are). I just got directed to that thread from a reddit thread on AI image generation.
I uh... still don't want to link the other boards anyways, cause its *still* 4chan lol.

[Robsoie]

And so as it was feared, it has actually happened.
https://www.businessinsider.com/ai-art-wins-competition-angering-artists-2022-9

Really sad to see this, regardless of the quality of the AI produced image, those art "judges" ditching all the artists that puts tons of works on making their artwork and giving the win to a guy that just typed a "prompt" in that ai image generating tool.
Makes me want to puke.