Small benchmark on MetaHuman Animator Markerless processing speed
I wanted to share a quick benchmark in case it helps other people testing the new markerless workflow.
Hardware:
CPU: AMD Ryzen 9 5900X, 12 cores / 24 threads
GPU: NVIDIA RTX 3090, 24 GB VRAM
Driver: NVIDIA 596.36
RAM: 64 GB
Storage: SSD / NVMe system setup
Unreal Engine: 5.8
Clip: 300 frames at 30 fps, so about 10 seconds
Tracking: Body ON + Face ON
Same footage and same project settings for all tests
Results:
-
Run Preview Pass OFF + Visualization Object EMPTY
Processing time: 5m37s -
Run Preview Pass ON + Visualization Object EMPTY
Processing time: 5m53s -
Run Preview Pass ON + Visualization Object ON
Processing time: 9m41s
In this test, disabling Run Preview Pass gave a small improvement, but the big difference came from leaving the Visualization Object empty. With Visualization Object enabled, the same 300-frame clip jumped from around 5m53s to 9m41s.
So for longer takes or batch processing, my current recommendation is:
- Leave Visualization Object empty unless you really need to preview the character inside the Performance asset.
- Turn Run Preview Pass off when you only need the final exported animation.
- Use preview/visualization only for short test clips.
Credit: thanks to Solomon Jagwe for pointing out the Visualization Object performance tip in his UE 5.8 MetaHuman Animator Markerless tutorial:
He discusses the Visualization Object slowdown around the 26:40-28:10 section.
Also credit to the UE 5.8 MetaHuman Performance source/UI discovery: “Run Preview Pass” appears to map internally to the skip-preview behavior (bSkipPreview). That helped clarify that this is optional preview work, not required for the final exported animation.
Hope this helps others compare results. If anyone has benchmarks with different GPUs/drivers, especially AMD vs NVIDIA, it would be useful to collect them here.