Adventures in AI Image Generation · 6:20am Oct 25th, 2022
I've been interested in trying out ai image generation for a while, ethical questions aside. While there are plenty of ai image generators online, they usually have limitations or want to charge you money eventually, and I thought it'd be more fun to set it up on my own computer, if it'd run.
Well, there's a free open source one called "Stable Diffusion". The computer I have with an RTX580 seemed like a good one to install it on until I realized Stable Diffusion is geared around Nvidia cards.
I do have a different computer with a GTX960 with 4 GB of VRAM, and it had 16 GB of memory. I decided to upgrade it to 32 GB after getting memory bottlenecked, but it actually works on this computer. I did have to pass it some parameters to tell it that my graphics card doesn't have much memory, though.
So, here's some prompts I ran on it after finding a pony image model.
Sweetie Belle as a bat pony, cute, adorable:
A cute blue bat pony wearing glasses sticking its tongue out.:
A purple bat pony with green eyes and a red mane eating a mango.:
A grey bat pony eating a sandwich.:
This is where I got Stable Diffusion from:
https://github.com/AUTOMATIC1111/stable-diffusion-webui
These are the command line arguments I used in webui-user.sh (This computer has Linux on it) --lowvram --opt-split-attention
Here's a pony model:
https://huggingface.co/AstraliteHeart/pony-diffusion
Here's the actual model people normally use. Though 1.5 just came out, so a lot of people probably still use 1.4.
https://huggingface.co/runwayml/stable-diffusion-v1-5
Seems pretty promising, anyways, and fun to play with...
AI art is both cool and scary to look at.
5694369
Yep! Though sometimes it totally misses pieces of prompts or clearly has no idea what I'm looking for. A lot has to do with how common what I'm looking for was in the images it trained on, I think.
And the other question I end up with is whether the images are close enough to someone's drawings to be ripping them off, too, and that's a question there will be a lot of discussion on for quite a while, I think.
--Sweetie Belle
You are the right pony for this job! I had been looking at Stable Diffusion with a view to trying it, but I still find AI art to be an absolute horror show, so it's probably not a good idea. Although I guess this is the time of year for it :)
5694379
This is definitely a job for "Cute, adorable fluffy (Sweetie Belle) wearing (glasses)."!
i.imgur.com/Uhj3jbn.png
It varies a lot. Sometimes it's great, and sometimes you're wondering where that body came from...
i.imgur.com/HbY8mLl.png
Also, sometimes it knows pony names and sometimes it doesn't. Any of the CMC were fine, and the mane 6 I've tried so far, but it didn't have a clue about Twist or Silver Spoon (though it could get close to Twist if I described her). Part of the trouble, I think, is it was trained on popular pictures, and if a pony isn't particularly popular, they aren't in the training as much.
It's a lot of fun to play with, though. Coming up with prompts, typing them in, and getting pictures a few minutes later is really neat. And I still have to play with all the img2img stuff, or most of the settings, really. I just found out what () and [] mean...
--Sweetie Belle