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

  • 🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
SS Rangers

download (5).jpg
 
Just some general thoughts:

I’ve been fucking around with Stable Diffusion for about two weeks now. In that time I’ve updated from a 3080 to a 4090, this did speed up as now I can generate 2 batches of 3(6 images) in about 20-22 seconds, which I have found to be the most consistent size.

I have found the best results with 45 sampling steps. Some say it does nothing after 32 steps, but 45 seemed to be most consistent in my case.
My CFG scale usually works best, for me, around 15-17, and a denoising strength of 0.75 usually offers the best results without going crazy random.

What I often do is swap between Euler A and Euler to get the closest results to whatever my prompt was. Occasionally I’ll have to use photoshop to change a color that was incorrect, or to copy/paste a better hand over the hands which fixes the AI’s general dislike of the hand and foot.

I have found that the image size plays a big part, and I generally keep it at 764 X 448.

I have also found that the negative prompts do more for the final result than the positive ones. As such, I add these prompts to every image before adding in the character specific ones

Positive
masterpiece, best quality,

Negative
lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry

My honest opinion is that this will not fully replace traditional artists yet. It takes a lot of work and fine tuning with tremendous trial and error to get the best results, and often times it requires a little extra work in different programs to get things to work. However, I think for quick concept art, easy fan art, or even to get a base pose/design for your actual art, this is one of the best tools around. I would compare it to artistic equivalent of auto tune. Yeah it can help a bad artist, but it sounds best when used with a good artist.

All that said, this has been a lot of fun to play with, and I highly recommend it. Better than any video game I’ve played in a while, and the most use I’ve gotten out of my 4090 thus far.


Here are some of my better results. I have yet to try classical painting styles.

Most recently I was asked by a friend to try and get it to make the girl from that lady bug show using Text2Img first. The pattern proved difficult but after around an hour I’d trial and error I mostly succeeded. Her mask was occasionally drawn, but in the best result it was removed

2E0A60E6-234A-435C-B6DB-B5508F82B8FD.jpeg

Another friend asked me to try making Krile from Final Fantasy XIV. This character is a dwarf with a rather complex cat hood jacket

This is what she looks like

12720898-D82E-4409-A0CC-9BA47800261B.jpeg

For this one, I used img2img so it could closer emulate the outfit. The cloak itself came out really close, but it never managed to draw her with the cat hood on properly. That said, the overall appearance is close enough to be compared with any standard fan art of her. The biggest issue was that her hair kept coming out pink, so I used photoshop to color it brown again.

2543103F-4D0E-4066-A41F-6EC6DBDD7028.jpeg

Here are two quick text2image grids I did.

One was Bill Nye in One piece’s art style, this was a small set of prompts to see how close I could get without saying much

09FE0649-CCE2-405F-BCD8-1625CAB4784E.png

This one was Dawn from Pokémon smoking a cigarette. I mentioned her by name since I figured she’s popular enough to get good results. It did draw her mostly intact, albeit never on model.

4959EEF3-4B1C-495C-93E9-8C5F8AD03BF1.png

These next ones were all Princess Peach with a Gun. And they all came out mostly decent. It kept trying to draw her in a bikini though, those I did not save.

2F8BEDA5-54AF-427A-A73D-9B8B52C4BA14.png

82902255-2AF0-4953-9279-B6F0CA7E0E36.png

95D7F7B8-90A2-4428-A78E-87FB9E69C70F.png

630D2C80-1D9A-428C-8194-C5D11412FE4F.pngE68732D6-E822-47B7-9EF7-32DE85BCE005.png39D10CB4-7C9D-404D-98FF-EBEF39BAF3D4.png51B17116-32D0-4EE8-9F84-6473FF11BEA2.png

My most difficult test was when a friend asked to see if it could draw the paper bear from The Amazing World of Gumball. For reference, she looks like this:

4AB95A30-5576-4DDD-9E54-7E6D476D9636.png

I used img2img and changed the prompts various times, both including origami and removing it at times. It never came close to drawing the character on model, but we did get some fascinating output.

DA4123EE-7AA1-4FBE-B486-5BB4EED3E17A.png
ECA577B3-26EF-4C7E-BC2B-75F37E3CCE99.png
ECA577B3-26EF-4C7E-BC2B-75F37E3CCE99.png
CE5ABDA4-5BAD-445F-8462-1528DD47D07B.png
1A788929-5C22-47AF-8D65-4A486E08708D.png
285E4881-2B77-421D-BDF1-2D3ABF4A3573.png
D0D97A61-5FD0-481C-BF42-A09159556C1E.png
C7EEE69D-E63F-466A-B3D7-5E25528B06A5.png2D139D33-3481-4B82-9341-EFA224780018.png

C8484DF6-E902-4FCD-A1D8-0850CBDBA6D1.png2AFC8526-545B-4AAC-A262-5B32AA7BBF7D.png382EF716-EE51-485B-86C9-3541FBB504E5.png54B9E752-35E1-415D-BDD4-1C77DFDF976C.pngF8BE1CB6-F7F1-4F87-B32C-6286FDAC7232.pngF1C34C8F-4CF4-4E45-B520-EE15077262D5.pngE44DADE8-0A59-4517-BDD0-D29DAE1CB165.pngA6A69280-7F17-4379-B831-DAAA54910B6B.pngBC457861-3793-454A-8216-C18F3D068246.pngF25B4DF4-9F30-48C3-801D-5D58652FF1BB.png0CA84073-B781-48C4-AEDE-CF1624A04759.pngCAADFA52-1C5E-4147-97EF-F7322232B2C8.png9A557643-9234-4EB3-A265-81D42EC5F87B.png9AC35B2D-614F-415E-BB02-9123062B866F.pngF9CE2354-9211-4382-AE96-DF03BE9A8032.pngBF2F1234-3CBE-4238-B1BF-C51648AB0C33.png9846FB8D-3F73-47F2-8568-6FE0A7659A83.png996900A9-475D-4F6B-BB6A-51480C74A791.png
 
How angry were you when the pen and paper were invented?
Pen and paper didn't take pictures of little girls and use them but to make more pictures of naked little girls but whatever clearly I'm in the wrong in this situation somehow.

God forbid we tweak the robots to stop them from lewding children. The sky would fall, the water would turn to blood, hell would spill out upon our realm because the fucking robots couldn't make "art" like this

You people bewilder me
 
Last edited:
I've been learning to draw with friends for a while, it's one in a small list of things I enjoy.
This shit makes me feel like there's no point. It's difficult to enjoy something when a machine could easily create something better or of similar quality in much less time.
It is retarded to let some program on the internet affect your real-life emotions but goddamn, if it isn't demoralizing.
My friend, don't despair. None of this AI art stuff is going to matter after the bombs drop and we're all blasted back to pre-industrial society. People like me who push buttons for a living will have to adapt or die quickly, but people like you who can work a trade with physical matter will flourish… after we re-attain some level of sustainable food production, anyway. Your future is as bright as an atomic explosion.
 
i feel like this whole AI art craze is going to go to the same path as NFTs
My friend, don't despair. None of this AI art stuff is going to matter after the bombs drop and we're all blasted back to pre-industrial society. People like me who push buttons for a living will have to adapt or die quickly, but people like you who can work a trade with physical matter will flourish… after we re-attain some level of sustainable food production, anyway. Your future is as bright as an atomic explosion.

this post can fill someone with either positivity or existental dread.
 
I could certainly see this used for stuff like video essays on yt, it doesn't seem all that useful for stuff like a cohesive visual or narrative sequence. That obviously could still change.

Stuff like logos will now be easier for regular people to develop. Though sometimes logos contain certain subliminal visuals that relate to the company, it would be interesting to see the ai replicate that.

Commissions were such a shit way to make money to begin with, even for the more successful of artists, it was only a matter of time before that whole thing hit the skids.

Stilllll, some of dem animu tiddehs do look "neat".
 
Last edited:
We are only here to share generated images... Mass Debates is probably the better home for ethical discussions about fake naked children.

By a request of @ConspicuousArdiunoDue:
Late, but here you go.

6xd4w5.jpg
 
@LiquidKid (no reply option REEE)

This is a crime against firearms
There were several where the gun was actually facing backwards and it looked like she was gonna shoot herself in the head. The AI does not draw guns consistently

Anyways the artists who will come out unscathed and probably in a better position are the ones who actually play with the artistic medium using real creativity rather then generic anime style #24629

I agree, the only artists that will be harmed are the ones that draw basic pinup porn of whatever flavor of the month waifu for a living.

People who actually develop unique art styles and tell stories with their compositions will be fine.


I wonder how specific you can make it if you feed it the right data sets. Can you just feed it UI images and have it help you make a video game UI.
It gets very wonky with text. The code would become a garbled mess
 
No, I did care about those things, I don't want people to lose their jobs.

There was a discussion on this years ago.


You can see I was optimistic that art couldn't be replaced as a purpose, but that's being demolished. It's looking like my negative prediction is the one coming true.
Sadly those who sound the alarm always seem to act as if Skynet AI will completely redefine "art" to replace the human element. It does not. It is merely a tool to enable the less talented to create images from their own ideas. What is so tragic about this?

A response from the amateur art scene that is all about personal greed and entitlement will not win sympathy. If an artist's work is good, it will speak for itself, no matter what tools the competition employs.
 
Back