inematds/VideosDGX — reverse-engineered prompt

I want this turned into a working Docker based video generation setup for a DGX Spark style machine, with a simple web interface and clean APIs so I can generate videos from text prompts without babysitting the system.

Please make the app run the four models already implied here, LTX 2, Wan 2.1, MAGI 1, and Waver, each in its own container, with health checks, queue based job handling, lazy model loading, manual unload, persistent storage for models and generated files, and an easy way to download finished videos. I also want a basic browser UI where I can pick a model, enter a prompt and settings like duration or resolution, submit the job, watch progress, and grab the result when it is done.

There are already scripts and web interface files in the repo, so please connect and clean that up into one reliable flow. If some models are still flaky, fix what is realistic and leave clear notes for the remaining blockers. Also make the quick start and first run experience simple, and look up current docs online if you need to.