LoRA scaling is now enabled in production for cloned voices. It keeps the foundation model's pronunciation quality while holding on to the style of the cloned voice, so clones stay accurate without losing what makes them sound like the source.

The default CFG weight for LoRA voices has also been lowered to 0.1. The lower weight gives more expressive, natural output from those voices.

Both changes affect how LoRA cloned voices are generated.