Shrinks - [CONCEPT] [LORA] [FLUX] - V3.0

New training config (again), resulting in:

• training image count reduced from 25 to 15 images

• minimally bigger LoRa size

• improved flexibility

• less bias

• better coherency

• more consistency