large(maybe) tests show:

rv1.2 performs well with easynegative

rv1.4 performs well with no easynegative

and the only difference between them is training steps.

100% game textures in dataset.

Highly recommended:

furry, rabbit ears, rabbit nose, :-<, yuran_face, yuran_body, yuran_tail,yuran_hairstyle in prompt and

simple background, black background in negative prompt

Negative prompt above is highly recommended because of the feature of the dataset.

The fact is, prompts depends largely on your base model.

After I run all the posts, I changed the Lora file name for easy understanding. Better pay attention when copy the generation data.