Okay, guys, so today I wanted to mess around with Stable Diffusion and try to get a picture of Ghostlord. Here’s how the whole thing went down.
First, I fired up Automatic1111’s web UI. I’ve been using this for a while now, and it’s pretty straightforward. I make sure it’s up-to-date.
Next, I needed a good model.I have serveral model,I choose “rev আবারওanimated_* [773c6c557d]”. Because it looks cool.
Then came the prompt. This is where the magic happens, right? I typed in something like:
“zzz take a photo of ghostlord.” I wanted to see what happened, so I kept it pretty * I added some Negative prompt,like:(worst quality, low quality:1.4),monochrome,
The First Few Tries
- First Image: It was…weird. Not what I expected at all. Lots of distorted face, not scary, just…off.
- Second Image: A little better, some improvement on the light, but still looked nothing ghost.
- Third Image: Okay, we’re getting a bit closer, It’s almost like human.
So, I realized I needed to tweak the settings a bit.I change the Sampling method from “Euler a” to “DPM++ 2M Karras”. I heard that one was good for details.
I also messed around with the CFG Scale. I usually keep it around 7, but I bumped it up to 8 to see if it would help. I also set the resolution to 512×768, which seems to be a sweet spot for some.
More Attempts, More Tweaks
I kept hitting that “Generate” button, and I started getting results that looked a bit more like what I was going for.I make them look more like photography.
Finally, after many, many tries, I got one! It was much better!
So, that’s it. That’s how my “zzz take a photo of ghostlord” adventure went down. It took some trial and error, but that’s part of the fun, right? Hope you guys enjoyed this little recap of my Stable Diffusion session!
