tencent/MimicMotion ? reverse-engineered prompt

Build me a working app for generating human motion videos from a reference image and pose guidance, using the MimicMotion approach described in the repo.

I want to be able to give it an input image, a pose sequence or pose video, and get back a smooth generated video with realistic motion, good detail, and support for longer clips. Please wire up the inference flow so it can load the model weights, run pose detection guidance, and save the output video in a simple way from the command line. If there’s a sample config already, make sure it works end to end and is easy to change for frame count, resolution, and other settings.

Please also make it practical for someone with limited GPU memory, so include sensible defaults and any memory saving options that help. If you need to check current docs for dependencies or model download steps, do that.