Seriously though, some call this the uncanny valley, where a realistic human figure is so close that you can’t tell what it is that’s off, you just know something is off. There may never come a point when the art gets past this because people are use to seeing and interacting with real people everyday. It’s like an exponential curve trying to approach perfection. The closer you get, the harder it is to get there.
When looking to make a character less obviously CG, those are my top 8 things to look out for.
Skin. Real skin is complex and layered. Unless somebody has abnormally oily skin or sweats a lot, real people aren’t that glossy either. Too much or too little can make a character seem plastic or wooden. Skin also isn’t 100% opaque. Skin has this slightly translucent quality that can be mimicked through careful and measured use of subsurface scattering. Skin that’s too CG often looks like that of an actor who’s wearing too much makeup or foundation. Real skin is also flawed. Little nicks, bruises, burns, or blemishes give our skin character.
Eyes. Those can be a dead giveaway. If a character appears too alert, that can look fake. If the eye material looks too glassy , that can look fake. If the character blinks too much or in an identifiable pattern… fake.
Mouth. Facial posing/animation is tricky. The lips don’t just move up & down or side to side. There’s forward & back movement too. There’s also some tongue action, which lots of animators ignore. Real people aren’t cartoons so don’t over exaggerate. Real people tend to be more subtle. Also, when posing the mouth, break that symmetry a bit.
Proportioning. Sadly, most people aren’t the heroic 8 heads tall our art teachers taught us to draw. Unless you’re built like a supermodel, athlete, or movie star, there’s a good chance that you’re probably just a mere 6-6.5 heads tall. Nothing wrong with that. It’s just that real world proportions are kinda boring. Real people, even generally attractive ones, are less than ideal.
Symmetry. We’re used to modeling with it. However, nobody is 100% symmetrical. It’s like the old saying, “If something looks too good to be true then it probably is.” Real people are kinda asymmetrical and imperfect.
Rendering. This encompasses a lot including: Lighting. Camera settings/setup. Staging. Shadowing. ETC. Those things aren’t easy to balance, even with real world photography. I’ve seen a number of real world scenes look like CG because somebody went crazy and overproduced a shoot. Don’t rush the results. A good render takes time and patience. Not just at render time, but also at post.
Quality. Similar to the issue of symmetry, you have to keep in mind how imperfect the real world is. It’s dirty. It’s all scratched up. Colors fade. Materials wear down. Things dent, bend, or break. All of this stuff kinda applies to humans. We’re imperfect. Understand the difference between believable and realistic. Something that is realistic is also believable. However, something that is believable isn’t necessarily realistic. If you’re going to make something realistic, you have to sit down and embrace all of those flaws and complexities. The more you do, the more people will buy into it.
Delivery. Ever watch WWE wrestling before? Notice how the guys never get hit hard, yet always seem to scream in agony? That’s what you have to do sometimes when it comes to CG. Sell the scene. Even the best character will look fake and boring if you have him on a plain backdrop and doing nothing at all. Even in stills, a character can appear to be animated. I’m not necessarily talking about big actions. Sometimes, pose and nuanced expressions can sell the character. A slight raising of an eyebrow. A slight shift in weight. How the character “communicates” with the chair he’s on. We all live in context. We interact with our world. Our expressions reflect our thoughts. Those “dead” characters sometimes seem dead because they’re not thinking about much of anything. Get your character to act, even if it’s subtle. Get him to think.
Lots of things can go into making a character more alive VS wooden. Photograph and video yourself or others. Try to break the scene or still down. Observation is a very powerful tool. Use it.
@teruchan: I don’t think uncanny valley has anything to do with it. I think UV is more to do with how realistic they look. But sometimes the person may actually look real, but simply doesn’t look ‘alive’.
@cookepuss: Thanks for those points. I will take those into consideration
There are subtle things that many people don’t even know that they notice, that many productions may not have the time or capability for. Things like a pulse; the way some areas (fingers) can change color when pressure is applied; Convincing eye darts & pupil contractions; or the occasional involuntary muscle twitch.
Everything we do is with bitmap textures stuck to the paper thin skin surface that is directly animated.
Real skin has thickness and things under it that all react to light and movement in different ways. Even the pigment of freckles and scar tissue refracts light different than the other colored parts of skin.
there’s also peach fuzz, microbumps, and wrinkles that physically stretch and bulge
Skin in real life glides across the fat and is driven by specific muscles and bones.
Even with good facial capture, something is going to have account for the physics behind the shading or the shot is going to have to be stylized in some way to hide those details - crushed blacks, color grading, film grain, low output resolution, etc
I would have to respectfully disagree. I’ve worked on plenty of digi doubles in my time, particularly on the last two Harry Potter films, that were virtually indistinguishable from the actors. I agree that once they’re animated, believability becomes a hell of a lot harder, but from a surfacing perspective, we’re not only close, we are there already and have been for some time.
Yeah I agree with this entirely. Also the limits on animation seem limited to humans, creatures are much more likely to withstand scrutiny, so my feeling is it mostly comes down to consistencies with real actors. i.e. I wonder if a complete digital human (as opposed to a digital double) used in a short scene would be noticeable.
And just for the sake of argument I’m not a believer in the Uncanny Valley as per it’s original definition. And research seems to agree with that statement. And I see nothing more difficult about emulating humans than emulating fire - it’s just the level of scrutiny we apply to them.
I don’t think digital doubles are a fair comparison because they are primarily used in action scenes. Show me a digital double framed from the chest up, standing their doing nothing, and I suspect it won’t fool anyone. I haven’t really seen fire done convincingly in a film yet either.
Digital double doesn’t strictly mean it’s used in action scenes, it means it’s a digital copy of an actual actor.
And you are bound to have seen a few standing there pretty, but the fact you’re not aware you did goes to show you have been fooled.
As Leigh said, surfacing wise we’re there, have been for a while. Animation wise, the examples are rarer.
Fire has also been done plenty times indistinguishably from the one shot. Plenty time I’ve seen plate fire replaced with digital one and nobody was any wiser for it, not even the people who shot the plates until told.
I suspect even on these forums a lot of people are simply unaware of the amount, extent and quality of work going into shots you take for granted were plates, but in reality weren’t. This thread at least seems to indicate so.
well ok, for movies and faking things with texture maps for typical movie shots with color grading, yeah we’re there
scientifically and medically, we’re not even remotely close if you’re talking about zooming into an object by a factor of 10,000 and expecting the geometry, lighting, and shading to hold up on the micro level the same as it does on the macro level.
As stuff moves and is animated, the shading is dynamic and needs to be accurate to be accurate. People do fantastic jobs right now adding a lot of animated texture maps that change as face shapes change etc, but those techniques are still an approximation.
We can get shots to look good, but if you rendered them at higher resolutions and higher frame rates, or could dynamically interact or move around the character like in a video game and really focused on them, they’re not perfect because well…they’re just not. Current computers and software have limits
Things like the current SSS shading algorithms that only account for an object’s surface - until people start using heavy translucent refraction transparency for all objects and model underlying all anatomy with GI based SSS , the shading is going to be technically off for anything other than movie shots.
Even animating - if you look at 5000 fps video footage, things ripple and flub around in real life as they move even though at 24 fps and motion blur, you think it’s moving in a simple way. It’s no wonder at higher frame rates that people notice how much more off animations are.
We’re always going to think things look realistic until we see something that looks even more realistic - hence why effects from the 80’s and 90’s don’t always hold up to modern expectations.
I wouldnt say ‘we’re there’ yet shading wise. We can fool people in certain types of cinematic shots with certain lighting, but I dont feel a blanket ‘we’re there’, which implies full mastery, is an accurate statement.
For example, even if you just take a still from this, I have never seen any shading that is as convincing on this level, and certainly not in motion. you look at this and you just know that it isn’t CGI.
The problem with your examples (the 5000fps one as well) is that they assume that you’d use the same technique to deal with all types of shots. I have no doubt an indistinguishable double of a 10,000 times zoom shot could be made but no one would bother doing it with the same geometry, lighting and shading because that would be probably be a stupid method.
Method is not the important thing when discussing photo-realism, the result is the important thing. And the results can be fantastic. It’s just a case of time, money and skill.
And you think all the super slo-mo you see in movies is all shot in lens on Phantoms and there’s no CG?
And at higher frame rates, if anything people had an issue with sets and actors and generally thought only the CG scenes were bearable, so I’m not sure what you’re talking about here (as there aren’t that many hfps movies that have been discussed by “people” I make the assumption we’re talking the Hobbit).
Rest, in all honesty, sounds a bit too much like rambling out of context. The topic isn’t “the current fractal inifnite detail models of humans are disappointing”, it’s “some CG people look dead”, and if they look “dead” it sure as hell ain’t because the current shading models are insufficiently advanced, because they long ago surpassed what photography reproduces in more cases than not. Often enough in fact it’s the hyper reality effect throwing people off.
I think that statement referred to stills, not animations. The Galleries here have plenty of stills that look perfectly realistic, as judged by the responses in the related threads. We apparently have a harder time spotting a fake human in a still image, but an easier time spotting the flaws when it is animated.
We add grain because we have plates to match. Chemical film has chemical grain, and digital shoots have sensor noise.
The grain and colour offsets we add are meant to align to the plate when we have one we work on top of, or to make the CG sequences fit into the general photography and not stand out as too pristine when you have full CG scenes.
I think a strong example of this being “really close, but not quite there” would be TRON: Legacy’sClu and the digital Flynn near the beginning (apologies that not all those images are from the film). In still frames they look to me like they exist as real, physical objects, but as wax figures rather than living humans. I accepted this for Clu since he only existed in a digital world, but for Flynn in the “real world” I found it distracting.
I should emphasize that I have huge respect for the artists who worked on those characters, and the effects are much better than anything I’m capable of doing. Still, at the end of the day I didn’t find the characters’ rendering convincingly “alive.”
I’m sure there have been digital doubles that totally fooled me in the 2-3 years since, but I’m inclined to think that when they’re 100% convincing, it’s often helped by framing, action, and motion blur that shields them from scrutiny.
Digi doubles are certainly not only for action scenes. We did stuff for the one Harry Potter film where the digital versions replaced the actors’ faces for a transformation sequence. The animation was a little strange for me, but the shading? Spot on. Also, regardless of how a digi double is being used, they’re still lookdev’d to a high degree, including close up renders and comparisons which are shown to the VFX sup.