MisoLabsAI/MisoTTS — reverse-engineered prompt

Build me a simple local app for Miso TTS that lets me type text and turn it into natural sounding conversational speech using the public MisoLabs model from Hugging Face. I want it to work from a clean setup, download the model automatically the first time, and save the result as a wav file I can play right away.

Please include an easy example script, plus a small interface where I can enter text, choose a speaker number, set the max audio length, and generate audio. If possible, add an optional prompt audio upload with a transcript field so the model can condition on prior audio for the same speaker. Make the app handle CUDA when available and fall back gracefully or explain if the machine is not powerful enough.

Keep the generated audio watermarked by default and include a short safety note about not impersonating people. Look up current docs online if you need to.