jd-opensource/JoyAI-Echo ? reverse-engineered prompt

Build me this JoyAI Echo project so I can run both parts locally, one for long video generation and one for the interactive world model.

I want a clean setup where I can follow the README, install the right Python environment for each project, download the separate checkpoints, and run inference without the two setups interfering with each other. For the long video side, make it generate long horizon audio visual clips with continuity across shots. For the world model side, make it work as an interactive demo where I can navigate and watch the scene, sound, music, and speech change together.

If anything is unclear from the docs, check the project pages or current model docs online and fill in the missing pieces in a practical way. Please keep the structure simple, make the entry points easy to find, and add whatever basic fixes are needed so a person can actually launch both demos from a fresh machine.