Alright, so I wanted to mess around with my Stable Diffusion setup and get Kane, you know, the wrestler, to show up as Diesel, his old gimmick. Here’s how it went down.
First, I fired up Automatic1111, my go-to for Stable Diffusion. I made sure I had a decent model loaded up – nothing fancy, just a reliable one that handles realistic images pretty well.
Then, I started simple. My first prompt was something basic like, “photo of Kane as Diesel, wrestling.” I hit “Generate” and… well, the results were all over the place. Some looked like a melted Ken doll, others were just random muscle dudes. Nothing close to what I wanted.
So, I knew I needed to be more specific. I started tweaking the prompt. I added stuff like “long black hair,” “black vest,” “black gloves,” “sunglasses,” trying to nail down Diesel’s look. I also threw in some keywords like “WWE,” “1990s,” and “professional wrestling” to help guide the AI.
- First try: “photo of Kane as Diesel, wrestling” – Total mess.
- Second try: “photo of Kane as Diesel, long black hair, black vest, black gloves, sunglasses, WWE, 1990s, professional wrestling” – Better, but still off.
- And so on.
I went back and forth, changing the prompt, adjusting the sampling steps, and playing with the CFG scale. It felt like a guessing game, trying to find the right combination of words to unlock the Diesel I was picturing in my head. I must have generated dozens of images, each one slightly different. Some were hilariously bad, others were getting closer, but none were quite perfect.
I remembered someone telling me about negative prompts, so I started experimenting with those. I told the AI what I didn’t want, like “short hair,” “beard,” “mask,” “ugly,” things like that. This seemed to help clean things up a bit, getting rid of some of the weird artifacts and off-model features.
Fine-Tuning and More Attempts
I kept refining the prompt, adding more details, and playing with the settings. I also started messing with different samplers, like Euler a and DPM++ 2M Karras. Each one gave slightly different results, and I was trying to find the one that best captured Diesel’s look.
I realized that getting the face right was the hardest part. I decided to try using a higher resolution, so I enabled the “Hires. fix” option. This slowed things down, but it did seem to improve the facial details, making the images look less blurry and distorted.
Finally, after a lot of trial and error, I started getting some images that actually looked like Diesel! They weren’t perfect, but they were close enough to make me happy. The key seemed to be a combination of a detailed positive prompt, a carefully crafted negative prompt, and the right sampler and settings.
It was a lot of work, but it was also pretty fun. It’s amazing how you can use AI to bring your imagination to life, even if it takes a bit of tinkering to get there. Now I’ve got a bunch of images of Kane as Diesel, ready to, I don’t know, use as my desktop wallpaper or something.
