This week I continued to work on building the new float, as well as teaching the silly freshman how to floats work. I drilled and tapped the holes in the new piston, which actually didn’t go terribly. I also finally finished up all of the acrylic welding.

The other thing that I have been working on is getting Stable Diffusion working on the new computers. I got it working pretty well on Tuesday, with the Automatic1111 webui. It works pretty well on 512x512 images, but once you get above that it doesn’t do very well.

We then wanted to try training our own model, as this is essentially what these computers were built for. So with consent from our friend, who also thinks it would be hilarous, we started to try to train a model on his face. Unfortunatly, I haven’t got the settings to quite work yet, but I’m hopeful that I can get it to start to train over the weekend. This probably wouldn’t take suuuper long, but it also wouldn’t be the final model, as I don’t really want to download about 5000 pictures for reference. This aren’t the picture of the friend, but they are pictures of other people, which allow the model to stay focused on the right thing, and bascially keep it on a leash.

A reference photo
This is one of the reference photos that we are using. We made sure that he was the only subject, in the photos. Another cool thing is a feature of the LoRA training, is the auto captioning tool.