Here's v2 with an improved dataset, now instead of everyone having big heads and small noodle bodies they're all really tall and skinny. It's even good with men now, but still not great with hands and definitely not feet.
v1 - After about 20 remakes I was finally able to get it to replicate those eyes just the way I wanted. It's a weak model and not great on its own, especially with hands unless it's a built in character, so I honestly don't recommend using it. Men always come out babyfaced so probably not good for that either.
Add hair spikes to the negatives, they have a tendency to sneak in.
Description
FAQ
Comments (3)
What a lovely art style! I'm experimenting with it, and I really like it save for the fact that it makes my output character much younger than I'm comfortable with when at full weight 😅 However, decreasing the weight allows you to find a sweet spot while keeping the unique and cute vibe it adds.
Yeah I started to notice that too at 1 strength, and I think I know what exactly in the dataset is causing that. This particular artist likes to draw shorter, noodly characters with big heads and I think AI saw the head to body ratio and figured they were all meant to be younger since they're in that weird spot between chibi and more detailed. I've already remade this model a few too many times but I think I'll have to try yet again to thin out those images that made it do that. The problem is that I think the colors in the model won't be as vibrant once I remove that data since I see a lot of those same colors in the generations, but that's just part of the fun when making these.
Awesome concept/style!