VFX_Veteran
Regular
Ahha! So it doesn't take for you to have worked in the field to notice the imperfections!
The move towards CGI. What consitutes 'the look' and how close are we getting?
- Thread starter Shifty Geezer
- Start date
Ahha! So it doesn't take for you to have worked in the field to notice the imperfections!
Well, I'm certain that there are some psychological elements there, although in my opinion it's a bit more about just how vastly different our (men's and women's) idea of female beauty can be. There's a huge range just for a start: the girl next door, fashion models, actresses both contemporary and classical, all kinds of traditional art (sculptures and paintings), celebrities, even porn stars. There's an incredible range of hair styles, make-up, cultural inheritance, and so on. There's a LOT of serious research as well, with things like the 'beauty mask', averages of hundreds or thousands of photographs, theories... Compared to this, men are really simple already
However, Drake is an interesting example because the team behind the character had both men and women in it, so I'm not sure if there's anything like this could be involved.
But I still think that the visual aspects I've mentioned are probably more important. Things like stubble, wrinkles, scars, more complex forms and such provide a lot more detail for the eye, or in other words the visual elements of the brain, and so it's a little easier to get away with less then completely realistic execution, the minute flaws are less noticeable because they're hidden under all the "noise". And of course Drake is still a stylized character, even if he's far more realistic now.
So males are probably easier to create no matter your gender.
Which makes me even more curious to see female characters from UC4. I'm quite sure that they're actually a much harder task for the team and this is at least in part the reason why we haven't seen any glances of them yet.
It's a really interesting question I've had for ages. Why is it so much easier to represent a man, especially a relatively old and wrinkly man, maybe with some beard, than a girl? Even accounting for the extra 'cheap' details afforded by the imperfections, perfection just always looks odd.
Then I thought... The vast majority of images we see of girls, in real life but especially in printed for or any other media, have a huge component male faces typically don't have: make up. A 'good' picture of a girl wouldn't actually show her real skin but what a make up artist painted on her face.
This creates an immense variety of different scenarios where skin just looks completely different from girl to girl - and this is on real people, all striving to look 'perfect', which is a completely unrealistic goal and half the time makes a girl look ridiculous. I mean I'm watching TV right now, some funny chat show, and these people look like they dipped themselves in paint.
At the same time, generally speaking, you know what the face of a guy cause that's what you usually see. His real skin.
So personally I think that part of the problem is from the origin, the basis of trying to make an already fake looking girl real, while trying to make a real guy look like a real guy.
The image above would definitely have less of a robotic feel if only the position of the face weren't be so perfectly centered. She already looks absolutely photorealistic but a more natural position, and maybe a bit more facial expression, would make her look like the other two.
You're definitely right about the issue with make-up - but only to a certain extent!
First of all, everyone on TV or in movies HAS to wear make-up. The amount of light you need for the cameras to work well in the studios would overemphasize the reflections of the oily layer on the skin and everyone would look way to shiny. So even males get a lot of powder on their face to dim the reflections.
Just look at Spacey here:
There's an almost complete lack of skin tone variation on his face, whereas in real life he's obviously very different - and in this case, yeah, I'm speaking from experience, having stared at close up photographs of him on COD AW. Still, I think that most people don't notice this when they're watching House of Cards.
Also, this is a very good example case of just how expensive even standard HD was for TV productions. They have to do a LOT of high quality make-up on everyone to keep the illusion, so I can't even begin to imagine what it'd mean to move to 4K...
The other thing is that realistic CG females won't get easier if you try to create a character without make-up, either. With the advent of cheap but super high quality DSLR cameras, there are lots of "amateur" pictures of young girls all around the net, and yet it's not a bit easier to try to create something like that in CG. The problems must run deeper, and I'm quite sure there's a lot of evolutionary training involved in this. Doesn't matter if you're straight or not, the wiring in our brains for pairing is just way too complicated and much much harder to fool, I guess.
First of all, everyone on TV or in movies HAS to wear make-up. The amount of light you need for the cameras to work well in the studios would overemphasize the reflections of the oily layer on the skin and everyone would look way to shiny. So even males get a lot of powder on their face to dim the reflections.
Just look at Spacey here:
There's an almost complete lack of skin tone variation on his face, whereas in real life he's obviously very different - and in this case, yeah, I'm speaking from experience, having stared at close up photographs of him on COD AW. Still, I think that most people don't notice this when they're watching House of Cards.
Also, this is a very good example case of just how expensive even standard HD was for TV productions. They have to do a LOT of high quality make-up on everyone to keep the illusion, so I can't even begin to imagine what it'd mean to move to 4K...
The other thing is that realistic CG females won't get easier if you try to create a character without make-up, either. With the advent of cheap but super high quality DSLR cameras, there are lots of "amateur" pictures of young girls all around the net, and yet it's not a bit easier to try to create something like that in CG. The problems must run deeper, and I'm quite sure there's a lot of evolutionary training involved in this. Doesn't matter if you're straight or not, the wiring in our brains for pairing is just way too complicated and much much harder to fool, I guess.
Well yes Kevin Spacey looks like a clay statue these days, in whatever he appears.
But we know what a guy's face looks like cause we wake up in the morning looking like shit, we see it in the mirror and we have that image in our heads the whole day.
But we know what a guy's face looks like cause we wake up in the morning looking like shit, we see it in the mirror and we have that image in our heads the whole day.
Last edited:
I just experienced something weird..
Do our expectations of what is looks real and what not real is a 'moving goalposts'? Simply because we consume more and more photorealistic stuff.
Here's the story. When ubi showed the division gameplay, I thought it was awesome photorealistic. But now I just stumbled the same video, watched it, and... It's not the same impression.
Now I see many kinds are wrong. The flat objects, the shadows and shading... Aliasing...
I'm pretty sure this is not a downgrade. It's the same video file. But the impression I get its downgraded.
Do our expectations of what is looks real and what not real is a 'moving goalposts'? Simply because we consume more and more photorealistic stuff.
Here's the story. When ubi showed the division gameplay, I thought it was awesome photorealistic. But now I just stumbled the same video, watched it, and... It's not the same impression.
Now I see many kinds are wrong. The flat objects, the shadows and shading... Aliasing...
I'm pretty sure this is not a downgrade. It's the same video file. But the impression I get its downgraded.
yeah thats what i suspected, thus making me feels the division looks "downgraded". Ugh.... i wonder if i play TLOU again it will look worse
btw your name @Inuhanyou ...
O_O in know Inu = dog right? But hanyou?
btw your name @Inuhanyou ...
O_O in know Inu = dog right? But hanyou?
Jurassic Park was on last night. At the time of its release, the dinosaurs looked real. Now they look fake. As I was saying to VFX_Veteran, humans develop sensitivities. The more you experience something, the better you're able to sense the fine details.
L. Scofield
Veteran
The night scenes still look pretty good. And the gallimimus scene as well.
My gosh, you tell me she is a real person and I believe she is. Outwordly work form that artist.Have to mention though that those images aren't from FF TSW, but from an amazingly talented guy called Luc Bégin:
http://forums.cgsociety.org/showthread.php?f=43&t=1192739
The head model's worth a look as well
Toy Story hasn't been matched yet, in realtime, I mean. But FF Spirits Within looks very fakey.
In fact, Sunset Overdrive (pics taken with the upcoming screen capture feature on X1) looks much more real in these shots than anything in that movie:
Last edited:
holly molley... that looks marvelous! it seems the way the light works is like the one on Fable blahblahblah (for X1 and PC, forgot the name).
SO simply uses artist placed lights to simulate GI. Don't remember where I've read it, but I did.
There is another pretty interesting issue related to this that I've wanted to post about.
On some recent projects we've had to work with game assets because of various circumstances; the main difference to our previous jobs was that this time the assets weren't to be used in the background (like for example crowd characters) but right next to the camera. Also, I'm not talking about average stuff but "hero" quality assets from a leading developer.
Now I was aware from previous experience that things were not likely to be smooth, but the results from just simply rendering the asset were below even my pessimistic expectations.
Proper supersampled antialiasing has completely ruined silhouettes and thin objects without actual thickness. Normal map based surface detail was almost invisible, especially on skin with proper SSS. Raytraced self shadows even from area lights were so sharp that you could've seen the poly edges. Transparency mapped polygons for hair just failed.
Mind you, we're using a proper path tracer (Arnold) for rendering, and I believe this is the main reason why all the assets were exposed as cheats and tricks and such. It was still surprising to see how fine the balance is between the assets' technical side and the renderer's unique solutions.
So, we ended up reworking everything, to various levels - remodeled parts of the character and practically the entire set, added subdivision and displacements here and there, replaced hair with curve based stuff and so on. All in all I'd say it would've been more efficient and possibly better looking if we've just started from scratch.
I'm still thinking about the lessons to learn from this project. The most obvious conclusion has to be that the entire pipeline should be engineered around the assets first; we should probably be able to refit our toolset to be better at hiding the flaws in lowres assets and such.
The more interesting consequences are about the future developments for game renderers and assets. A lot of existing tech and asset creation methods would probably fail if the fidelity of the rendering algorithms should move a big step forward, so perhaps the next step in hardware resources for new consoles might not be enough to support both. Or maybe it could be overcome with a lot of innovation? I've always talked here about how game engines are using less complex approximations of offline technologies, but I feel like there's a middle ground between the fast way and the proper way that hasn't been explored yet but it might be a lot more challenging than we'd think. Or it could be super simple - after all, PRMan has been using cheats and tricks for decades to avoid raytracing, and it has worked pretty well. Maybe realtime renderers can also increase scene complexity radically and still rely on approximations for the rendering, and delay the move to more advanced tech for another generation of consoles?...
If people are looking at computer games now and saying, "that looks like CGI," then I think that's a given!
Another interesting test would be like-for-like comparisons of the same scene rendered realtime and offline and compare the look jsut on rendering quality. So no true SSS on the CGI version if the game doesn't have it, but approximate the SSS in the CGI version in the same way. So basically use realtime shaders and offline lighting. Then have a best case CGI version using the same assets (or similar quality, so same triangle counts etc.) but improve the shaders. Then an idealised CGI version I guess. It'd help examine how the shaders influence the final experience, and the lighting. Ooo, and a CGI version without AA and post effects!
Here you can find some celebrity no makeup selfies.
http://www.msn.com/en-ie/entertainment/celebrity/celebrity-no-make-up-selfies/
VFX_Veteran
Regular
We could use an SSS approximation for sure. Henyey-Greenstein is one that Disney uses without having to bake to point clouds. I think that you'd still run into issues with framerate with like-for-like shaders though. Some of our shaders do a lot of complex things (including evaluating several procedural noise functions) that I don't think would translate to real-time at playable framerates. I've always stated that the big limitation I see is lack of geometry in the scene. We still need real hair curves and good displacements and several million instances of things like foliage, trees, etc. I think the shaders have moved a lot further than geometry bandwidth.
I'm told we can't use a lightmap based fake SSS ion Arnold and I'm not an expert so I believed the guy...
Also, the point is not just about using "cheaper" rendering methods, the point is that as far as I can tell a path tracer will never be able to render a lowrez, normal mapped, realtime shaded game asset the way a GPU does. We just can't match the output no matter how low we set the dials.
Also, the point is not just about using "cheaper" rendering methods, the point is that as far as I can tell a path tracer will never be able to render a lowrez, normal mapped, realtime shaded game asset the way a GPU does. We just can't match the output no matter how low we set the dials.
I am a console gamer but I think this GDC conference will be interesting
http://schedule.gdconf.com/session/...-directx-11-12-visual-effects-in-star-citizen
And Next Week Epic will show an Open world demo.
http://schedule.gdconf.com/session/...epics-open-world-demo-presented-by-epic-games
http://schedule.gdconf.com/session/...-directx-11-12-visual-effects-in-star-citizen
And Next Week Epic will show an Open world demo.
http://schedule.gdconf.com/session/...epics-open-world-demo-presented-by-epic-games
Last edited:
