rhymes-ai/Allegro — reverse-engineered prompt

I want this repo turned into something I can actually run end to end on my machine without a lot of manual setup. Please make it easy to generate short high quality videos from a text prompt, and also support the mode where I give a first frame image, or a first and last frame image, plus text to guide the motion. The main result should be saved as an mp4 and match the model’s intended output, around 6 seconds at 15 fps and 720p when possible.

Please wire up the setup so it installs cleanly, downloads or points to the model weights from Hugging Face, and gives me a simple way to run inference with options like prompt, seed, sampling steps, guidance strength, output path, and CPU offload for lower memory machines. If the training and fine tuning scripts are already part of this repo, keep them usable, but prioritize a smooth inference experience first. Add clear usage notes and sensible defaults. If anything in the repo is outdated, look up the current docs online and make it work.