Aurora 4x
C# Aurora => AI Content => Topic started by: AlStar on September 07, 2022, 05:38:01 PM
-
I've been playing around with Stable Diffusion, an AI image generator, for a couple days now. One of the first things I thought to work on with it was new race pictures - I could make* my very own! No need to scower through the internets for work of a correct type and in a style that I like that's been created by others to crop and resize... I can now get my computer to labor over new pics on its own!
Anyway, I've been running it through it's paces some, and I thought I'd share some of the more successful outputs.
This is "head and shoulders, cyborg soldier, sinister atmosphere, highly detailed, intricate line drawings, artstation", which gave me the following:
(https://lh4.googleusercontent.com/XnjstdEHF4JhLsKfCSrzto4lMKy1SZxgG4D84DwEGd_67qe7rL_kPcqKOH-KRHjbkP0=w2400)(https://lh4.googleusercontent.com/gwia1M9UyUjw8zGNTJC7y8Ggck5NGjrT-Xz42uJtPU5OqNzBT40UGDVtRB87Fv8n4YE=w2400)
(https://lh3.googleusercontent.com/AcygYoCzZuRaY0tlvwaR61NZNu4bz0Mk4m32aglhMAAdGXDRqC6EVTP43r8D7kxR5Bw=w2400)(https://lh4.googleusercontent.com/ceFqMEiTLMoY0Ejx5XX1jryNmQwwcDZ8KfvgVn9hWXKSjdzbUuyB-OYR-fyIB80fktU=w2400)
I then ran the one in the lower left corner through img2img, with the simple command of "trending on artstation":
(https://lh4.googleusercontent.com/9_jn4s7hlThznSlAUs4euAFZ6arOAF5sScj_16FwElgX3byN5qhyWbxJKaY9lXu0pcg=w2400)(https://lh5.googleusercontent.com/_5A5ZU0I5bNxsfTp5ssB8mjc3qF-D8LFcaRnHhx4pnxViXZXsDMI6iRxKBjbdSHtVPc=w2400)
(https://lh6.googleusercontent.com/79Et9F5D8_c6avPngh0hjU7RioSv6Fo1idYVDTj7_Vh8JkKRclvxSXzD_QgM9B3MVjA=w2400)(https://lh4.googleusercontent.com/cApyJZnxUk6PbcGZ5C0ivyeQBUg6cljpZEZQT7IuD6kFEQfsqu5RR828glmix9hOxDY=w2400)
The future is now, and it's pretty cool, I've got to say.
*For certain definitions of 'make'
-
That is really great :o . Experimenting with some AI stuff myself atm. How can I get access to this image generator?
-
That is really great :o . Experimenting with some AI stuff myself atm. How can I get access to this image generator?
I followed this (https://docs.google.com/document/d/1owAMJGe56sbocCdrv7IO8fM6I4NLqxZ2bJgfI7EsYAw/edit#heading=h.kvl3yl1apz15) installation guide (found through Reddit), which has excellent step-by-step instructions, with pictures.
If you've got a decently hefty graphics card (I'm running this on a 3060 Ti, and each image takes about 10 seconds to generate) and the space and ability to download several gigs of data to get everything set up; I totally recommend checking it out.
-
This is really cool. I'd love to have something like this hooked up to the ship designer to generate an illustration for my ships...
Just playing around with a random online image generator (no idea how sophisticated it is) got some results I wouldn't mind seeing in a thumbnail:
spaceship battlecrusier with spinal mounted laser (https://creator.nightcafe.studio/creation/1vLu8eBNjlDscJaNEB4o)
spaceship missile armed fighter (https://creator.nightcafe.studio/creation/qKySWLUAIb9KNAtoZ6ZX)
-
Bottom right on the top set looks like a cyborgified star wars First Order trooper
-
I've also been geeking out with Stable Diffusion, using it to make character images for my DnD character, my home-brew Traveller campaign (WIP), etc....
Personally I'm running it in a docker container on my Ubuntu machine (https://github.com/AbdBarho/stable-diffusion-webui-docker). Docker was the easiest way to get around Python's dependency hell. I only have a 1060 Ti (6GB) so image generation takes ~45s for 512x512 images, and twice that for 832x512.
If you don't have a good GPU, I recommended poking around Lexica.art and simply downloading something that is close to what you want. There is some drop-dead gorgeous AI-generated imagery on that site.
-
So, it's been a little while, but I finally got around to seriously building out an AI-generated set of race pictures. I figured that a good theme for an AI-generated image set would be cybernetic / cyborg entities, so that's what I ran with, although we've got some images that morphed over to more alien-looking or robot-looking, depending on the AI's whims, some of which I kept because I liked how they turned out.
Starting off with roughly 2,000 images, which were a mix of text-to-image (for new raw images) and image-to-image (for when a particular image interested me, and I wanted to see more permutations on a specific theme), I whittled down the selection to 375 images that I liked.
Of course, these were all the default AI-generated 512x512 images, so the directory weighed in at 141 megs. While that isn't entirely insane in this day-and-age of multi-gig downloads, I decided to trim things down a little bit. A batch resize to 75% (384x384) got the size under 100 megs - which is notably small enough to attach to this message - while remaining perfectly fine for the little window that Aurora uses for race pictures.
Overall, the process took about a day, running batches of 20 images at a time in the background while I was browsing the internet or watching something streaming, then a couple hours combing through the results to decide on what to actually include in the image pack.
Anyway, without any further adieu, a preview:
(https://lh6.googleusercontent.com/K_wW4aY03TZQ6IE76KwKxp8Szy7KMdCSWceB57tJ2S4qNOm9ZYptFdPljqEZS-vEiIM=w2400)
-
I'd argue that this belongs in C# Utilities/Image Enhancements, rather than in AI Fiction - I'm not using AI to make stories, just new art assets.
-
I'd argue that this belongs in C# Utilities/Image Enhancements, rather than in AI Fiction - I'm not using AI to make stories, just new art assets.
True. I was trying to group all of the AI related stuff together.
-
I'd argue that this belongs in C# Utilities/Image Enhancements, rather than in AI Fiction - I'm not using AI to make stories, just new art assets.
True. I was trying to group all of the AI related stuff together.
Maybe 'AI Content' would be a more suitable title for this section, and more representative of the broad range of stuff possible with it?