Okay, so today I decided to mess around with Sid Fernandez. No, not the former MLB pitcher, but the Stable Diffusion model. It was a totally random pick, I just stumbled upon it and thought, “Why not?”
Getting Started
First things first, I had to get the model. I grabbed it from, well, you know, that place where all the models hang out. It was a pretty hefty download, a few gigabytes, so I went and made a cup of coffee while it chugged along.
Setting Up
Once it was downloaded, I had to get it into my Automatic1111 setup. I use it, because I’m too lazy to learn other ways. I just copied the file into the models/Stable-diffusion folder. Seemed simple enough, right?
My First Attempt
I fired up the UI, selected the Sid Fernandez checkpoint, and typed in a really basic prompt: “a photo of a cat, cute, fluffy”. I hit generate, and waited.
The result? Well, it was a cat, I’ll give it that. But it was a bit…off. Like, the eyes were kinda weird, and the fur looked more like a bad hairpiece. Not exactly what I was hoping for.
Tweaking the Settings
I figured I needed to play with the settings a bit. I’m no expert, so I just started messing with stuff:
- Sampler: I switched from Euler a to DPM++ 2M Karras. I heard that one was good.
- Steps: I bumped it up to 30. More steps, better quality, right?
- CFG Scale: I lowered this a bit, to around 7. I read somewhere that too high a CFG scale can make things look wonky.
More Attempts
I tried a few more prompts, still keeping it simple:
- “a beautiful landscape, sunset”
- “a portrait of a woman, realistic”
The results were definitely better than the first try. The landscape was actually pretty nice, with some decent colors and details. The portrait was still a little strange, but way less creepy than that first cat.
Adding some Negative Prompts
I remembered about negative prompts, so I added stuff like:”blurry, deformed, bad anatomy, extra limbs”. I guess these helped,because I did see a slight improvement, so I used those to start with.
Final Thoughts (for now)
Honestly, it’s a mixed bag. Some images are pretty good, others are…not so much. I definitely need to spend more time experimenting with different prompts and settings. And maybe do some actual research instead of just randomly clicking things. But hey, it was fun to mess around with, and I did manage to get a few decent-looking images out of it. I’ll probably keep playing with it, because who knows, maybe I’ll accidentally create a masterpiece. Or at least a cat that doesn’t look like it’s wearing a toupee.
