Deploy InsightFace Swap Pipeline

I need the open-source InsightFace face-swap model installed and fully configured so I can produce high-quality swaps for media production on both videos and images at 1080p.

Here is what I expect once the job is complete:
• A repeatable setup (bash script, Dockerfile, or clear step-by-step notes) that provisions the model on my own Ubuntu server or an inexpensive rented GPU instance.
• A test run that proves the pipeline can process sample footage and photos without quality loss, including a short 1080p clip and a few stills.
• Guidance on how to queue jobs, control face detection, and tweak blending parameters so each swap costs only a fraction of a cent in compute time.

Please make sure any dependencies—CUDA, PyTorch, FFmpeg, and InsightFace weights—are pulled automatically or documented precisely so I can rebuild the environment later. If you prefer another lightweight orchestration tool, I’m open to it as long as the result remains simple to maintain.

Once everything runs end-to-end on my machine, the project is done.