How Hard is Your Face?
May 8, 2011
Got a face you have to render? CG faces are hard. Of course, some are harder than others so I’m introducing the “How Hard is Your Face” test.
How Hard is Your Face:
- Does it have real human proportions (vs. creature-like)
- Is it light-skinned (vs. dark skinned)
- Is it younger (vs. older)
- Is it attractive (vs. ugly)
- Is it a woman (vs. a man)
Of course, ALL faces are hard. Any time someone does a good CG face that moves, it’s a real accomplishment. That being said, some are harder than others.
Ask yourself those five questions and give yourself a point every time you say yes. If your answer is 0 it shouldn’t be too bad, and if you answer 5 then you’re screwed. In general I’ve found that shots of realistically-proportioned light-skinned cute 20-something women are much harder than stylized creature-like dark-skinned ugly old men.
Looking back, the first CG face that most people say “worked” was Gollum. You could choose other ones, but he’s a good benchmark. So how hard was Gollum to do on the scale?
- Creature-like proportions. +0
- Got dark skin. +0
- He’s over 100 years old. +0
- He’s not exactly cuddly. +0
- Gollum was a dude. +0
Gollum was a big 0 in the difficulty, which shouldn’t be surprising. Of course he was a huge accomplishment. But it should make sense that the first big CG face success was on the easier end. You have to crawl before you can walk. Let’s do another one.
Next up I’m going to Benjamin Button. Personally, I thought it was incredible. You could nitpick a few shots, and some of my friends thought it got a little “creepy” at times but I thought it was great. How hard was he though?
- Realistic proportions. +1
- Light skin. +1
- An old man. +0
- Empirically, he’s an attractive guy. +1
- Dude. +0
So Benjamin button gets a 3. But solving the “Head Problem” is still a long ways away. In particular, getting women to look good is sooooooo much harder. That one should probably be worth two points.
The main reason I got inspired to write this was because LA Noire is coming out. One silly discussion that will inevitably begin is which characters look better than others. Certainly, artist talent and time spent will play a huge part on how good a certain head looks. But the ugly old black dudes will probably look better than the young hot white girls simply because they are easier to do.
Even if rendered brilliantly, the biggest problem I see LA Noire having is the disjointed nature of the motion. It defines the uncanny valley for animation. Not for a lack of talent, mind you; it’s just the nature of the capture technology.
It’s interesting that you rate ‘Old’ as easy, since it always seemed to me that age greatly complexify a face by making it less homogeneous : the texture, the thickness as well as the looseness of the skin can look and move very differently (but often in subtle ways) between the eyelids, neck, forehead and cheeks.
These differences usually give a lot of small visual clues that any viewer can identify as rightly or wrongly reproduced (both in terms of look and motion), and since younger faces seems more coherent and uniform, I’d say the margin for error would be bigger on these.
Christopher Plummer in The Sound of Music vs him in Parnassus is a good example of what I mean.
Cowbs: Doing the head animation separately from the body definitely has a tradeoff. Here is a link to the Ucap demo where the head was captured separately. The neck animation wasn’t quite right, but it still looks much better than if we would have had standard rendering with combined animation.
Actually, with this demo, the number one complaint we got when showing the demo live and up close was the cloth. It’s like they could accept the head and the animation, and the next thing they focused on was the lack of a cloth sim.
http://www.youtube.com/watch?v=DZuMMevcjHo&feature=player_embedded
As far as LA Noire goes, I try not to talk much about specific games. (-:
Vimes: Certainly older people adds to the animation. But it seems to make the shading easier, maybe because they have less subsurface scattering. Getting the subsurface scattering right seems to be the hardest part.
With younger skin, I think you also get more pigmentation changes, but I haven’t studied it in depth. As you stretch the skin, there is less blood which seems more pronounced with lighter skin.
In general though, I’ve really found that in any game or movie that has both CG young and old people, the young look freakier.
Hi,
I wonder what’s your opinion about Triss Merigold? She appeared in Polish Playboy.
Hi Pablo: Triss gets disqualified for two reasons. First, it only counts if you are going for truly photoreal. It looks like they were trying to make it look lie a video game. Second, pictures don’t count either…it has to be video. Getting the shading right is really hard. There are quite a few artists that can get it right in stills, but making it look right in action is an order of magnitude harder.
Would you say that your high-scoring faces are harder only because faces that are supposed to be ‘conventionally attractive’ just come under more scrutiny?
Difficult/Beautiful = Realistic, pale, young, female. The western (Hollywood) ‘beauty ideal’.
Easy/Ugly = Bestial, dark, old, male.
Shouldn’t the things that make a photorealistic face convincing be largely independent of ‘how pretty’ the face itself is? It’s not like the same defects spotted on the ‘attractive’ faces that make them look weird don’t also generally apply to the ‘ugly’ ones – you’d no doubt spot a similar pattern of defects on all of the games’ faces if you looked at each with an eye for detail.
In the relevant sense the ugly black guy isn’t going to look any better (or worse) than the pretty blonde girl, since the quality of that render isn’t going to be any more faithful than that for the girl – they’re both coming out of the same tech.
The bias is that the old black guy just isn’t ’supposed to be beautiful’, so no-one particularly cares when the render gets it wrong and he looks weird. It risks reinforcing this bias that only certain young, white, female faces are allowed to be beautiful.
Hi Anon. Not really. I’m just telling you what I see as well as virtually every good CG artist I’ve ever talked to. I think part of it is subsurface scattering. That’s the hardest part to get right. It doesn’t seem like bias though…you can’t control what your eye sees.
This is a +5 and pretty close to escaping the uncanny valley:
http://vigilantcitizen.com/latestnews/japans-newest-popstar-outed-as-cgi-creation/
Hard faces are ugly like Pollard the football players wife who play for the ravens, her face is harder looking than a man
I think it comes down to exactly what the title states, hardness. With softer skin the subsurface scattering is much more visible, as are the features under the skin. While we’ve had some remarkable progress in the last decade simulating this, our models are still shells with nothing but empty space inside. If we expect to keep with that trend, we’ll probably have to approach it as multiple phases to color calculations.
At a minimum:
A phase to calculate the color under the skin, with geometric shape intact. We can probably fake this to a large degree with a good displacement mapping algorithm. Then a phase to calculate color and opacity of the skin itself, this might be a bit hard unless you happen to know people with exceptional knowledge of human skin. Then a phase to cover particulate buildup on the surface of the skin, as well as hairs extruding out of the skin.
Each of those phases likely requires several sub-phases to get over the uncanny valley, but it’s just to point out how much we are over simplifying things.