nari-labs/dia ? reverse-engineered prompt

Build me a simple app for generating realistic spoken dialogue from text using this model.

I want to paste in a script with speaker labels like S1 and S2, hit generate, and get an audio file back that sounds natural, including things like laughs, coughs, or pauses when they are written in the text. If possible, let me also upload a short voice sample so the output can match that voice more closely.

Make the experience easy to use, with a clean page for trying prompts in the browser and a basic command line way to run it too. Save the finished audio as an mp3, and give me a small example I can try right away. If there are any setup details needed for GPU use or current model loading, handle them in the project and look up the latest docs online if you need to.