Less than four years ago, Epic Games developers presented new tools at the GDC conference that they wanted to use in the coming years. to get as close as possible to reality in the sense of facial animation. The digital Andy Serkis, who was moved in real time by the real one, seemed intoxicating, and it seemed that a great revolution was really heading into the games. But time has moved relentlessly, and it is certainly not possible to say that realistically moving faces are finally the norm. Nevertheless, individual companies are trying again and again and another aspirant warranty The technological advancement is Ziva Dynamics. The name of this company may flash the memory of Miles Morales from Insomniac Games – this is where the company interfered with the muscle animation in which it specializes. But now introduces the new ZRT Face Trainer tool – currently only in beta and only for selected applicants – who scatters everything we have seen in the field of video games with the first example. Including a unique technology that moved the faces in LA Noire.
Of course, it is necessary to slightly slow down expectations, if you count on automatic deployment in all games that are created on the Unreal Engine. The people at Ziva Dynamics are currently providing access to a trial version, in which you will send your model to cloud processing, where artificial intelligence reaches the bank of tens of thousands of so-called 4D movements and builds the appropriate virtual muscles from them. The returned model will then be ready to respond adequately to the reading of scanned data from the faces of real people and present a believable result in real time. Of course, everything depends on the quality of the original model, respectively. with which the graphic designers will proceed specifically to use it in the ZRT Face Trainer. However, the better the work will be done at the very beginning, the better the result from the tool falls out, so you can really work your way up to the form you see in the video above.
However, Ziva Dynamics also offers a simpler solution that does not require such careful preparation or long calculations in the cloud. Even the basic tool ML Face can process data from the actors’ faces in real time and very satisfactorily. Another short video, which no longer looks like this, is used to demonstrate this tool bombasticky, but it would still be above standard if the developer used only coarse output without muscle application using AI. Of course, it also depends to a large extent on the sophistication of textures or global lighting, which adds much-needed liveliness and credibility that not so well-crafted models cannot offer. On the other hand, Ziva Dynamics does not only want a realistic representation of us humans. On the relevant page in the official presentation, it also presents the application of the same data to faces from the fantasy world – such as orcs or elves – but even there, the ZRT Face Trainer does not seem to have more serious problems and the results are consistent.
Of course, the question remains as to how useful this solution will be in the end, but Ziva Dynamics boasts in the technical presentation that, unlike current practices, it is a much more effective way to achieve the same, but rather better results. Reportedly, even an adequately modified model should not have such large memory requirements, which would load the implementation, and something tells us that it may already be happening. When you look at the only more meaningful example of what can be Senua’s Saga: Hellblade 2 and immerse yourself in familiar details, you will get information about the use of 4D data and special rigging of facial models, with which 3Lateral officially helps. However, it is possible that the ZRT Face Trainer is just one part of the whole development mechanism, and the second Hellblade – which is powered by the fifth generation of the Unreal Engine – can finally ignite that revolution.
–
