The model is made for creating weirdcore-like images. It was trained on a mix of weird images, that I associate with weirdcore, dreamcore, y2k aesthetics and other Y2K mad vibe stuff.

Versions

Article with comparison

I dont feel, that one version is better than another. In case of blurry, distorted weirdcore pictures (our case) it shouldnt be compared by quality, so take different versions as just different "blends". They just different :) Check each and take the one, that you feel the best with

Details about each version in 'about section' of every model

Usage

add 'y2kmad' activation word for it to work. I find it work better with smaller weights with small prompts like 0.75-1 and if you have detailed big prompt, increase wights to 1-1.5

Support

You can support me and request new lora's/checkpoints at: https://boosty.to/smgbuddies/donate

描述:

Same dataset as v0.0.1 Mix, but at first was trained as full finetune, and then extracted to LoRA

训练词语: y2kmad