Trigger: happy new year

This LoRA is very experimental and unstable at this point.

Low generation successful rate, it depends on your base model, and it gets even lower when prompts get longer.

Batch test on AOM2 and short prompts, v1: 6/16 => v2: 12/16