MEIKO LORA XL

=================================

----READ THIS HERE FIRST--------

=================================

This is an experimental version of locally-made SDXL LORA specialized to create an image of the first generation of VOCALOID character, MEIKO. I choose this because it is easier to train characters with less details and easier to manage it. However, since SDXL requires enormous amount of VRAM which requires at least 24GB to work it with few hours, training it with consumer-level gaming GPU eventually took longer time, at least for 12 hours when trained with Nvidia RTX 3060 12GB.

May requires anime SDXL models to get an accurate image representation of the character even in realistic modes. See the image samples here for prompt example. Different strength levels defines the realistic levels:

Strength 0.5 = Semi-realistic image

Strength 0.7 = Realistic image

If the image appears to have "double parts" eg. absurdly tall person etc, you might need to lower it down its resolution. Also, hi-res fix may be needed to fix image that apparently low-res despite setting it to higher resolution. Use Adetailer to fix distorted face, fingers etc.

Also, this LORA is compatible with SDXL Illustrious model, may adjust the prompt to give a better representation.