Ranma Style LORA

A LORA trained to imitate the style of late '80s early 90's anime specifically, the Ranma 1/2 anime. It's similar to my Ranma 1/2 checkpoint, but it's not an extraction from the .ckpt, rather I trained the LORA from scratch with a tweaked data set. IMO the the main differences between them is, that the LORA seems to do a better job with full bodies and it lacks the graininess, slight blur and the sort of analog feel the .ckpt has.

As with most other LORA you're going to have to change the strength to somewhere between .4-.7 otherwise it will probably overpower and distort your results. It was trained with Anything V3, so it should work with it's various offshoots.

My training parameters were:

• max_resolution: 512,512,

• learning_rate: 1e-4,

• train_batch_size: 1,

• epoch: 5",

• mixed_precision: fp16,

• save_precision: fp16

• text_encoder_lr: 5e-5,

• unet_lr: 1e-4,

• clip_skip: 2,