İ think it works best at 0.5. Tongue out improves it too.
Hey, do you mind sharing your training data?
ahego.safetensors
