Ads
Discovering my first ad on my website brought an unexpected sense of joy. Probably because it’s our own site, and these ads are here to help cover hosting costs. I’ve lost count of how many attempts I made to activate ads on www.usmalbilder.ch, but the latest one finally hit the mark.
The only tweak I made was adding two menu entries where I talk about the project and explain the value of coloring pages. But hey, no complaints! 🙂
It’s fun to see how the ads are automatically placed and I guess I’m fine with these ads, because I know how the whole website was built and how the ads are integrated. But for most users it may feel pretty bloated.
Bear in mind it is sort of an experiment and if we get a few bucks to pay for the hosting expenses, I’m more than happy. More important was the fun to create something using a lot of AI technologies and having a project with my kids, so they could learn some basics of doing business (with the internet and generative AI).
Stable Diffusion
Initially, we have been using MidJourney. It was a great experience, but there are alternatives and as part of the project I also wanted to try out Stable Diffusion. I was a bit sceptical at first, but I read in an article, that there is a pretty simple solution to run the model on a local machine. I don’t have a powerful machine. It’s a pretty old gaming PC with a GTX 1080, let’s see if it can manage the model!
Getting started
I found this setup to work very nicely: GitHub – AUTOMATIC1111/stable-diffusion-webui: Stable Diffusion web UI
I basically followed the automated install guide:
Most likely it worked without any issue, because I already played with some other LLMs and I had the right Python version, the Cuda libs and stuff installed. If you want to try it out, better read the manual.
When rnning webui-user.bat, the web ui is started automatically I was able to test the model for the first time.
First results
The first results were horrible. I guess one really needs to understand how to write good prompts for the individual models.
My machine was able to generate rather small images with ease, but it still takes quite some time to get a final result with higher resolution. It’s OK to generate an image once in a while.
Models?
The UI is great, but it allows to fine tune a lot of stuff. I’m still in learning mode, but one of the first thing I realized is that there are different models that you can choose from and they may be optimized for one task or another.
I found Civitai: The Home of Open-Source Generative AI a great source of prompt ideas and models.
You can basically download the models and put them into the models folder of your stable diffusion installation:
That’s it!
Prompt ideas
On Civitai: The Home of Open-Source Generative AI you can also browse tons of AI generated images. The great thing is that they usually tell you which model was used and what the prompt was, this is a great source to learn how to write effective prompts.
Conclusion
Getting ads up and running was not as easy as I initially thought, but I’m happy it finally worked out. It is a bit frustrating to not understand exactly how it works. But it seems that also small tweaks may help with convincing the algorithms. 🙂
Running an image generating AI on my local machine is fascinating, the access to these models has become increasingly simpler and I’ve also been able to generate some Xmas coloring pages. Playing around with image generation is great fun.
