This uses the dataset from my dreambooth checkpointed Pepe Frog models. It combines the ideas of both cartoon and rendered/realistic pepe frogs. You can use cartoon PepeFrog or realistic PepeFrog tokens to switch. Also included in the fully captioned training data were various art styles of Pepe, so the model should have pretty good flexibility.

One thing to note is that many of the images were obviously meme's and so the background was a solid color. Generally the model will output solid color background all the time for cartoon renders and half the time for realistic ones. So if you want pepe in an environment, you need to describe the setting well. But the blank background are best for making meme's so that why I wanted the focus there.

Some example pics are using high res fix. All were generated at 512x512 and a few were upscaled. You can generate at higher base resolutions just fine as well.