LoRA scaling is now enabled in production, preserving the foundation model's pronunciation quality while retaining the cloned voice's style. The default CFG weight for LoRA voices was reduced to 0.1 for more expressive, natural output.