It has definitely been a while since I have made a post here. Last time I was still in college and having a hard time getting to new projects but now that I am done I have started testing a lot of new things. One of my new favorites has been testing stable diffusion. I just started working with it the other day and I am so surprised at the ease of use and some of the great results you can get from it. I still have a ton of information to learn but I have been able to produce some images that I am very impressed with.
I am a big fan of anime and there is already a lot out there for producing high quality anime images. Below I have included some of the better ones I was able to create yesterday.
Here’s another one I created using a simple prompt “cute anime girl long hair”.
I created these images using stable diffusion webui. I got a new graphics card around Black Friday (PNY RTX 3060) which has allowed me to start working with some machine learning and AI tools. It took some time to get it setup as I am currently using Windows 11 (for gaming). I had a lot of headaches removing an old version of Python that kept holding the PATH so I couldn’t use Python 3.10 (required for stable diffusion). I also had some difficulty with the PATH for git but once I got through issues it was relatively easy to get stable diffusion running.
My first images were not very great using the default checkpoint. I got images like what you see below.
After doing a bit more research I was able to find some useful blogs and videos to point me to using hypernetworks, VAE, extensions and what settings would be useful. I still have a lot of work before I can learn about how stable diffusion works. I am early in the learning process but impressed by how a little information and low level of knowledge can produce some excellent images.
I am going to include some of the resources I have been using to learn more about the process of generating these awesome images below. Also, I plan on either writing more articles about this subject or updating this one as time goes on. If you have a computer with a powerful enough GPU, I highly recommend giving it a try yourself.