Most of the comments that I can see below are restricted to discussing the impact of announcement and demo on the Unity vs. Unreal situation. However, it turns out that here are other very relevant and important issues which arise from the technological specifics behind the demo which, although they obviously directly impinge on the rivalry, have potentially broader implications.

'Uncanny valley' issues related to face-based (as well as body motion-based) realism in simulation for both games development and video production, whilst there have been recent very significant advances in addressing them are still a (pun unavoidable) rapidly moving target, a target where this demo highlights both substantial advances and still as-yet unresolved serious shortcomings.

The advances mostly relate to the introduction of machine learning and 15TB of training data into the content production workflow to reduce a rendering job from 6 hours per 50 frames to 3 milliseconds per frame as detailed in this article:

https://www.globenewswire.com/news-release/2021/12/10/234977...

Returning for just a moment to the UE vs Unreal thing, those results were discussed in an article about a demo created using Ziva Dynamics's Real Time Trainer and Unreal Engine 4.26, but Ziva was subsequently acquired by Unity.

The discussion I'd be interested to see here (or somewhere else!) is about how ZRT compares with (or perhaps even fits into?) Unreal Engine's Metahuman workflow and resulting content creation and realism results.

Perhaps more importantly, how might the ML-driven capabilities of ZRT (will they still continue to be available under Unreal as well as Unity?) and ML-based impact upon realism be expected to manifest themselves

Here's a link to the ZRT 'preparation workflow', which a Z-brush user who was trialling the trainer setup says took them about an hour to produce a workable result:

https://drive.google.com/uc?export=download&id=15J6vX-6vU5EU...