Initial. (Experimental release)

The effects of incomplete captioning are clearly visible in the generated results.

The following tags need to be counteracted in the negative prompt:

glasses, open mouth, blush, speech bubble

If these are not specified, the above features will almost always be generated.

If I update this LoRA in the future, I plan to address these issues by correcting the captions.

dim32/alpha32, CAME, UNet LR 3e-05, cosine 848 steps, 53 images (1024²)