ace-step/ACE-Step-1.5 ? reverse-engineered prompt

I want to build a powerful local AI music generation app that can run on my own computer. The main idea is to type in a description of the music I want—like 'a chill lo-fi beat with piano and rain sounds'—and it should generate a full, high-quality song. It should be able to create tracks up to 10 minutes long and support lots of different styles and instruments, and even handle lyrics in different languages.

Beyond just text prompts, I want it to have some advanced features. It should let me upload an audio file to use as a style reference for new generations. I also want it to be able to create a cover version of a song, separate a track into its individual stems like vocals and bass, and even generate background music for a vocal track I provide.

Please build this with a simple, user-friendly web interface that I can run locally. Make sure it's optimized for consumer hardware, so it should work well on NVIDIA GPUs but also support Macs and AMD cards. It should be a Python project, so please set it up with all the dependencies and models needed to get started.