This model is trained on 81 images. Please leave feedback as I am still exploring in low-rank loras.
About Low-Rank LoRA series:
I am currently testing on performance of <10 dim LoRAs on characters and styles and found that you can get decent results for characters using 1 dim and 1 conv_dim, and 2 for styles (no regulation images). The models give less details, but are good enough to identify the characters/styles, and are also less likely to get overburnt/overfit. I will continue to train and upload other models. If you like this model, please consider training on low ranks as well!