Sunday, September 26, 2021

Have We Crossed the Uncanny Valley?

This synthetic human face was created by Chris Jones using purely digital tools. 

The face cycles through a series of dramatic expressions. At the extreme holds there are a few involuntary twitches, which Jones hand animated based on observations from live action reference. (Link to YouTube)

The surface is a neutral gray, which helps clarify the stretching of the micropores and the compression into wrinkles. The surface could easily be mapped with realistic colors and subsurface scattering, which would make it appear photographic.

--

Chris Jones used Blender and Krita, rendered in Cycles on Surface Pro 4 & on a 2008 vintage PC. More information about the method 

6 comments:

arenhaus said...

Nope. The facial structure is cartoony. The skin stretches unpleasantly like silicone. Et cetera.

Vladimir Venkov said...

Not there yet. Even in the best examples there is always something that bugs us.

atanguay said...

I'm a CG/VFX artist by trade, and while this is REALLY nice work, it's largely "just" that - the work of a really talented artist using the tools at his command to do a really nice job. (free tools BTW, which is amazing) Meaning, most certainly, not everyone can do this. I've been doing this for 25 years, and I certainly couldn't. It's really stunning.

-atanguay

MasterVertex said...

I don't think that manually set up deformation systems will ever cross the uncanny valley simply because of the level of possible complexity. The amount of minuscule detailed movements and deformations that make a real human face feel as such, are in the millions. The amount that a rigger can manually design, are in the hundreds at most. The way to absolute realism has to involve deep learning somewhere, because that's the only way you could have the required complexity to create and tweak millions of variables at once.

Vladimir Venkov said...

MasterVertex - Spot on. We can always spot when there are manual deformations in the setup. Yes, the only way is or mocap or deep learning.

Unknown said...

bruh everyone in this comment section is just trying to be different