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

[The Mass Shooter Ron Soye]

3D for everyone? Nvidia’s Magic3D can generate 3D models from text

3D for everyone? Nvidia’s Magic3D can generate 3D models from text | …

archived 22 Nov 2022 10:10:06 UTC

archive.ph

Comments page 1

Magic3D: High-Resolution Text-to-3D Content Creation

It over

 

[Involuntary Celebrity]

in minor drama news people have started snatching future sdupdate urls to debate discord mods and nanachi, since the author had only reserved them up to https://rentry.org/sdupdates5

 

[Catgirl Tyranid]

To thank the farms helping me to laugh longer and harder than I have in months, here are some kiwi related images.

 

[Dumpster dived waifu]

Wondering if anyone can tell me what the fuck I did wrong. I've installed the CPU Retard version of the AI, but no matter what I get the error:
"ModuloeNotFoundError: No module named 'numpy'

Can someone smarter than me help?

 

[Catgirls are Love]

Wondering if anyone can tell me what the fuck I did wrong. I've installed the CPU Retard version of the AI, but no matter what I get the error:
"ModuloeNotFoundError: No module named 'numpy'

Can someone smarter than me help?

That's a python module. Did you enter all of the commands (from the right directory) to install the needed resources? (if it's the rentry guide, Steps 5/6)

 

[Dumpster dived waifu]

That's a python module. Did you enter all of the commands (from the right directory) to install the needed resources? (if it's the rentry guide, Steps 5/6)

I did, and to make sure I'm not fully retarded, I tried manually installing with: pip install numpy and conda install numpy

 

[Catgirls are Love]

I did, and to make sure I'm not fully retarded, I tried manually installing with: pip install numpy and conda install numpy

Unfortunately, I can't help further without running through that setup myself to see if any issues pop up. Someone else seems to have had the same issue.

 

[Dumpster dived waifu]

Unfortunately, I can't help further without running through that setup myself to see if any issues pop up. Someone else seems to have had the same issue.

Some futzing about later, I've gotten it working I think. My computer is very slow for creating art.

 

[DeadwastePrime]

i have bought a 3060 and hopefully more vram will be quite helpful in doing the ai arts
might even consider training the ai off of certain artstyles cause its really bad at some

 

[seri0us]

Some futzing about later, I've gotten it working I think. My computer is very slow for creating art.

What kind of hardware, render times, and prompts?

 

[Human Institute Director]

Wondering if anyone can tell me what the fuck I did wrong. I've installed the CPU Retard version of the AI, but no matter what I get the error:
"ModuloeNotFoundError: No module named 'numpy'

Can someone smarter than me help?

Looks like you've already solved it, but since I've run into this issue previously this might be helpful for some: if you've already run 'pip install numpy' but still don't have the module, chances are you have multiple versions of python running on your machine (e.g python 3.7 and 3.8 ) and the pip you're running is associated with one of the python versions, but the python command you're using is one of the other. So you've installed the module for a version but not the one you're actually running.

Check which version of python is associated with the default alias "python" ( 'python --version' or 'which python' for Linux users), which version of python is associated with the default 'pip' command ('pip --version'), and when you need to install something for a given python version, e.g. python3.8, just run 'python3.8 -m pip install <your_module>')

 

[Dumpster dived waifu]

What kind of hardware, render times, and prompts?

Hardware: potato with 8gb of ram and an i3
Render times: ~20 minutes
Prompts: so far nothing but test prompts to test what I can do, how long things take, and if I need a fire extinguisher for my safety

 

[WelfareNiggerQueen]

No depressed bird girl gf. why even live (crying_pepe.webp)

 

[Involuntary Celebrity]

Stable-Diffusion v2.0 is out: https://stability.ai/blog/stable-diffusion-v2-release
Pros: cool new functionality, native 768x768 training
Cons: they're scared of titty and reportedly pre-filtered the dataset with LAION's nsfw filter with a threshold of 0.1 (where 1.0 means legitimately nsfw lol). Not that previous versions were great with nude anatomy anyway. Might complicate model merges I guess, but I'm guessing it wouldn't merge out of the can in any case due to the changes.

Needs some extra wrapping to work so WebUI doesn't support it yet but they're working on it, should come very soon.

 

[Epic Fail Man]

"Jesus"

Spoiler: NSFW

"Goku"

Spoiler: NSFW

 

[Cloacan]

Spoiler: buttocks

A peek into Nick Fuentes' dreams...

 

[Kiwi Fairy With Bow]

Happy Thanksgiving Kiwis

 

[AmpleApricots]

Cons: they're scared of titty and reportedly pre-filtered the dataset with LAION's nsfw filter with a threshold of 0.1 (where 1.0 means legitimately nsfw lol). Not that previous versions were great with nude anatomy anyway. Might complicate model merges I guess, but I'm guessing it wouldn't merge out of the can in any case due to the changes.

Weren't they really, really big about being anti-censorship no matter what? What happened? I'd usually guess the usual suspects (pearl-clutching trannies) but this degeneracy is right up their alley and I doubt they'd run storm against it.

 

[Involuntary Celebrity]

Weren't they really, really big about being anti-censorship no matter what? What happened? I'd usually guess the usual suspects (pearl-clutching trannies) but this degeneracy is right up their alley and I doubt they'd run storm against it.

IIRC they originally announced this change (which was going to be applied to 1.5 but some goodguy researchers released it early from under their noses) pretty shortly after feds orchestrated that AI kiddy model scare. Which turned out to be bullshit and a blip everyone's forgotten about, but the boss is a wall street dick or something so overreacting is probably part of the business plan.
And this time they aggressively pre-filtered everything going in so there's no real model to get leaked. So fuck you if you're interested in like, whole eras of art, I guess.

Luckily there's a lot more people entering the game now thanks to the current popularity so there'll be plenty of alternatives soon enough, but it's still a shame because the depth preservation mode in particular looks fucking dope.

 

[SomeBirds]

Not very Thanksgiving-y, but here's the Boris Vallejo test results:

Female Prompt:
a beautiful young queen with large green eyes and red lips, wearing ornate blue and gold robes, standing in a ruined cathedral, (((colorful))), prismatic, hyper-realistic, detailed, illustration, boris vallejo

Male Prompt:
a handsome knight with serious green eyes and angular features, wearing an ornate blue and gold cape, standing in a ruined cathedral, (((colorful))), prismatic, hyper-realistic, detailed, illustration, boris vallejo

Results from prior to starting training (ie, results from the examples of Boris Vallejo already in the module):


Not bad, but pretty generic. Now here's the exact same prompts after our training session.


The post-training results do not include any of the Boris Vallejo content that was originally included. We trained it for the prompt 'Boris-Vallejo' which prevented us from accidentally over training it.

Our 'Boris-Vallejo' prompt combined with the 'Boris Vallejo' prompt inherent in the module. For funsies.