My second attempt at the small-sized lora series: Dakota Johnson.

It's not a very compelling proof, as Dakota has a rich set of priors in various models, making training quite easy. But still, it goes to say: don't waste the prior knowledge of the base models.

The generated images in this version might be a bit blurry; this is not due to the small dimensions, but rather the lack of a carefully curated training set. However, it still proves that a very small lora can achieve sufficient similarity.

STOP POSTING HUGE LORA.