I have created this LORA to mimic the style of the artist "Vadaboob"This models trigger phrase was meant to be "flat color anime" however that seems to worsen results.I am aware that this model has very unstable anatomy and very dubious biases, id love to address these issues if i knew how (this is my first lora that is sorta functional) however i do have ideas and i will attempt to implement them within soon.Out of the models I tested this with Hassaku XL seemed to work best with the least biasing and ok anatomy.Version 2a and 2b seem to both work fairly well, couldint deciede which one to post so posted both, the main diffrence between them is that version b is trianed for like 3 more epochs, additonally it seems as if 2b functions better at lower steps

Vadaboob Style LORA

Decreesed learning rate and increesed epochsGot rid of some erroneus images in the datasetTurned on shuffle captionsFixed the max tokens (75 -&gt; 225)Somethin else myb idk

Files with SHA256: 435c05457c31becb4aacc1613b0d55943ca6006f0f90377a6f74a7e9c9430b07