Filter Auto Framing
Automatically frame your face, through the power of Machine Learning! This filter simulates what it's like to have a professional camera operator follow your every move on stage, never missing a beat. Get rid of your old manual cropping, and let modern technology handle what should be automatically happening.
⚠ This feature is not available if none of the supported providers are available. ⚠
Version Information
| Status | Version |
|---|---|
| 🔴Added | 0.11 |
| 🟠Unstable | N/A |
| 🟢Stable | N/A |
| N/A | |
| ❌Removed | N/A |
Any option related to Tracking.
The filter supports multiple Tracking modes, each with their own advantages and drawbacks. Depending on what kind of stream or layout you're planning, either mode may work for you.
Ideal if you only need to track a single face and don't expect to show more people in the video frame. Depending on the Provider used, this mode may be more lightweight while also being more precise compare to other modes.
When one person just isn't enough, there's Group tracking. It allows tracking many people at once with high accuracy, and combines all tracked people into a single massive frame. Great for couch-coop, talk shows, multi-person podcasts, etc. The limit on how many people can be tracked depends on the Hardware and Provider used. (Only available with Providers that support multiple regions of interest.)
The frequency of Tracking, in seconds or Hz (default if not specified). Lower frequencies reduce system usage, but also decrease tracking quality. Tracking quality can be balanced out with prediction and smoothing.
Motion related options.
How much smoothing should be applied to detected movement (in Percent)? Very small movements not visible to the human eye can end up being very distracting when they are picked up, which can be filtered out with smoothing.
How much motion should be predicted when movement is detected (in Percent)? This can significantly improve the quality of tracking when the frequency of tracking is reduced, but too much will have a negative impact on tracking quality.
Framing related options.
How much smoothing should be applied to the final frame? This can further reduce visible artifacts from imperceptible movements, but too high values may have significantly more negative impact than motion smoothing would.
The padding for each tracked frame in px or %, with the latter being relative to the tracked frame size.
The offset for each tracked frame in px or %, with the latter being relative to the tracked frame size.
Ideal Aspect Ratio for the framed area which will contain all tracked people. Can be selected from a drop down or manually entered as either X:Y or as X where X is the result of X/Y. If left empty, will default to the Aspect Ratio of the source the filter was applied to.
Select which provided tracking functionality should be used to scale up the footage. See further down for the available tracking providers.
Supports 'Group' mode with up to 8 people. Requires a Tensor-capable GPU (Turing 2060 or higher, Ampere 3060 or higher), as well as the latest NVIDIA AR SDK Redistributable.
- Turing (RTX 20xx) Redistributables: https://international.download.nvidia.com/Windows/broadcast/sdk/AR/nvidia_ar_sdk_installer_v0.8.2_turing.exe
- Ampere (RTX 30xx) Redistributables: https://international.download.nvidia.com/Windows/broadcast/sdk/AR/nvidia_ar_sdk_installer_v0.8.2_ampere.exe
- Ada (RTX 40xx) Redistributables: https://international.download.nvidia.com/Windows/broadcast/sdk/AR/nvidia_ar_sdk_installer_v0.8.2_ada.exe
This work is licensed under a Creative Commons Attribution 4.0 International License.
The wiki has been under repeated automated attacks by bots, with attempts to phish or infect unsuspecting users with malware, trojans and similar software. As a response to these attacks and GitHubs complete ineptitude at handling this situation, we have decided to write-protect the wiki for the time being. If you wish to submit changes to the wiki, please do so over Discord. Thank you for understanding!