cocktailpeanut/hallucinator — reverse-engineered prompt

Build me a simple local web app called Hallucinator where I can upload or record an audio clip and have an AI continue it, like audio autocomplete. It should run from a Python app with a clean browser UI, show the input audio and the generated result, and make it easy to try again with a different clip.

This is for NVIDIA GPU machines only, so set it up for CUDA PyTorch and use the Hertz audio model from Standard Intelligence as the base. Keep the setup straightforward, with requirements, a launch command, and any helper scripts needed so someone can install it manually or run it through Pinokio. If flash attention is annoying to install, use a faster to install attention option instead, as long as generation still works.

Please make the app practical rather than fancy. Handle uploaded audio safely, give clear status messages while it is generating, and include a short README explaining install and run steps. Look up current docs online if you need to.