BreakPoint
AI Generated Images with Stable Diffusion
So I saw some news stuff about AI generated images. But once a friend of mine showed me it in action I had to set it up and play around. The technology is absolutely amazing and has tons of uses.
There are a few ways to set it up. I first went with the python development method mostly described here. But there is also a Windows installer with mostly everything you need from here. You do need an Nvidia GPU with 4GB of VRAM, 8GB is recommended.
But after setting it up oh boy have I been having fun. There are just so many good and bad ones I don't want to flood this post, but here are a couple of my favorites.
Ron Swanson as Ironman
Ron Swanson as the Terminator
Chris Pratt as Captain America
Phoenix on an alien planet
Blue Tesla getting struck by lightning in the city
Lighthouse on an alien planet
People can also take images and ask the program to change it in interesting ways. Artists are using this to get ideas of what to make and inspiration. Game developers are using this for tiling of textures. You can use this for AI upscaling of images too. So much cool development around this tech.
I used one of the online ones and asked it for a "LabVIEW block diagram" and it gave me a lot of images that looked like this:
I've seen a couple articles today on this subject, and they have some pretty cool uses. This one talks about a very abstract MSPaint level drawing of Seattle, and then turning it into some kind of alien invasion future.
https://hackaday.com/2022/09/06/stable-diffusion-and-why-it-matters/
https://andys.page/posts/how-to-draw/
And here are a few more from my Napoleon Bonaparte tests.
Diffusion is AWESOME! I'm impressed beyond words by what i've seen!
I've made me a logo for my discord with it. 🙂
Search LabVIEW like a graph! So you can add training information to it, but it gets even more demanding on your PC. And my computer is just barely good enough to generate images, not additional training data. Here is a video from a VFX crew that takes images of the employees, and asks the AI to insert them into scenes to a story.
@Hooovahh wrote:
And my computer is just barely good enough to generate images, not additional training data.
Never mind I upgraded my computer. I wanted to make a better Linkedin profile picture, since I've hated mine for years. So I trained the AI on Google Colab with 20 pictures of my face. Then I took that back into my computer running AI generating code locally. The result is my new Linkedin profile picture.
This office isn't real, the suit isn't real, and I'm not real.
More like "I am the Architect"
(I didn't want to waste too much time on this one. I could have made it better. I probably spent 3 hours or so on my Linkedin profile picture over the weekend. This took about 10 minutes.)