As the possibilities for creating photo realistic, computer generated creatures advances, post production and VFX studios have yet to reach the ultimate CG goal: to create believable CG humans in close up. But there’s a potential pitfall in such an endeavour. The more lifelike you try to make a CG character, the more empathy for it turns to revulsion as the viewer’s brain realises there’s something ‘wrong’.
Japanese roboticist Masahiro Mori first described this phenomenon in the form of a graph back in 1970. He plotted ‘familiarity’ against ‘human likeness’, producing a dip in empathy for realistic humanoids that he named the Uncanny Valley. Mori hypothesised that the closer a humanoid object, whether a robot or an animated character, comes to resembling a human being in its motion and appearance, the more positive our emotional response to that object becomes – until suddenly, at some point of very close resemblance, our emotional response turns from empathy to uneasiness. It’s precisely because they inhabit Mori’s Uncanny Valley that zombies or corpses scare us, or why the unnatural movement of a humanoid puppet can freak us out. It also goes some way to explain why the first photo realistic computer animated feature film, Final Fantasy: The Spirits Within (directed by Hironobu Sakaguch) bombed in the cinema when it was released in 2001. Audiences just couldn’t empathise with the computer-generated characters.
In the 1970s Mori advised not trying to make robots too lifelike in order to avoid the problem and his advice is still heeded by many working in CG today. Lloyds TSB’s long running campaign featuring the life adventures of a highly stylised animated family and also Disney/Pixar’s 2009 film Up are cases in point where the makers of the work have aimed well left of the Uncanny Valley.
However, with CGI technology and skills marching ever forward, how close are we to bridging the Uncanny Valley and creating believable CG humans for commercial purposes? And will doing so be an exciting breakthrough or ultimately be even more unsettling?
“Yes, undoubtedly the technology will make this possible soon,” says Tom Painter of Bigman, a 3D company based in Hove. “I predict we will bridge the Uncanny Valley in five to 20 years, though there still may be moments where the illusion is broken.”
Mark Grosvenor of London-based Smoke and Mirrors isn’t so confident. “Hair and fur textures are now very realistic and movement, whether motion captured or animated, is perfectly achievable. But to mimic a human, whether in print or motion, is a completely different proposition,” he says.
Why? It comes down to the way humans are hardwired to scrutinise each other – and in particular the way we convey emotion and meaning through numerous subtle facial expressions. Then there’s the way hair and skin and muscle behave together, and the way light reflects and reacts with human skin and eyes.
“The closer you look at a human, the more detail you observe,” says Wesley Roblett, a digital artist at Saddington Baynes, who has created fully CG humans for clients including Smirnoff. “In fact, the process of creation starts with a symmetrical model before the real work of moulding abnormalities and asymmetry begins.”
Of course we live in a world where we are used to seeing images of humans that have been altered in some way – particularly in fashion magazines and advertising where limbs are lengthened and skin flaws smoothed out. “We’re so accustomed now to seeing real people who have been retouched, stylised and airbrushed,” says Saddington Baynes’ James Digby-Jones, “that the edge between what we perceive as real and what feels uncanny is now pretty hazy. In this sense it’s easy enough to create a CG person who looks like a retouched model – and various artists have. But creating a character who could walk amongst us undetected is much harder.”
Jonathan Fawkner, a visual effects supervisor at Framestore in London, regularly works on pushing CG realism to its limits on big budget films. He worked on creating the three cyclops characters in the Wrath of the Titans film that was released at the end of March. Not only did the mythical giants have to look like giant humans, albeit with one eye a piece, they were to be virtually naked and appear in super close-ups, meaning muscle, skin and hair and the way they all moved and interacted with their environment had to be convincing. The characters were created using a hugely complex combination of 3D rigging, 2D texture mapping, motion capture data and global illumination ray tracing lighting techniques, not to mention creative input on the character design to work out how to make a one-eyed humanoid capable of emoting. Fawkner estimates creating the scenes took 8,000 man hours and 1.25 million computer processing hours.
The main obstacle on the path to the far side of the Uncanny Valley, then, is time and therefore money. “We know all this information about how light bounces around and how it bounces around inside human skin and eyes, because academics and experts are researching these areas,” Fawkner suggests, “but when you look at a human face, there’s infinite subtlety. And so if you have infinite time and infinite processing power you could let a computer work through the hundreds upon hundreds of complex algorithms to work out how to perfectly synthesise a human on screen.”
However, Bolton University lecturers Angela Tinwell, Mark Grimshaw and Andrew Williams believe that no matter how much time and money is spent, the breach will never be crossed: the Uncanny Valley is actually, they argued in a paper published last year by the International Journal of Arts and Technology, an Uncanny Wall. “Technological discernment on the part of the audience generally keeps pace with technological developments used in the attempt to create realistic human-like characters such that, ultimately, the perception of uncanniness for such characters is inevitable,” they argued. The gulf, in other words, will never be breached because we will always stay one step ahead in our perception of reality.
And perhaps the quest is not one worth undertaking in any event. VFX teams wonder if they’ll ever need to create photo real human facial performances in CG. If a real human performance is required, then it’s always going to be a lot quicker and cheaper to get a human to perform. It’s also the skills of the artists at CG studios in characterisation and stylisation, rather than their ability to perfectly synthesise realism, that is key to the continued survival of their industry, some argue.
“Despite rapid advances in rendering algorithms, fracturing dynamics and muscle systems, which simulate the movement of muscles under skin, there is no replacement for good old-fashioned observation,” says Mark Knowles, lead CGI artist at Taylor James. “Observation and an artistic eye will always be 70% of the battle. While we place great emphasis on honing these creative skills with all of our artists, they are constantly going back to real-world references,” he continues. “The technology is a tool, and it can make us quicker, more efficient, and allow us to push the boundaries of accuracy. But unless you know what ‘physically accurate’ looks like, you will never achieve it, in spite of the technology.”