Notch Reference Manual

NVIDIA Virtual Background

Method

This node uses NVIDIA’s Virtual Background technology running on NVIDIA RTX enabled GPUs to pick out human forms in video / images and remove them from the background, outputting a greyscale mask. This eliminates the need for greenscreen keying, allowing for use in much broader contexts.

The background removal is achieved using AI and was primarily developed by NVIDIA for use on webcam feeds, for video calls and streaming, so has been trained primarily on those data sources. It will give best results on video feeds that resemble those – e.g. front shots of a person’s head + shoulders. However, it still achieves good results on a much wider range of scenarios. A 16:9 aspect ratio input feed at HD resolution is recommended for best results.

Raw texture Processed texture Modified texture
Example Nodegraph

Attributes

Name Description
Active Turn the effect on or off. Values < 0.1 will turn the effect off.
Quality Mode Select quality mode:
  • Best Quality (Slower) – The best quality, but lower performance. In this mode, the height of the input buffer must be greater than 360 pixels.
  • Fastest Performance (Lower Quality) – The highest performance at a cost of lower quality output. In this mode, the height of the input buffer must be greater than 288 pixels.