A LoRA model trained on the Krea 2 base model, optimized for consistent character representation with realistic details, trained for 700 steps using Rank 32 and Alpha 32 in bfloat16 precision.