Probable Uncanniness
by covert.c. on Dec.14, 2005, under computer graphics
Wow, this was a hot topic.Literally, ten thousand people clicked through the source articles in various places as it spread across the ‘net, and almost a thousand individuals read my journal entry from yesterday. Its not surprising, considering what a great piece of work that 3D image is.
Here is a nice backgrounder that Max Kor (the artist) prepared. Its a really interesting read, even if you just look at the pretty pictures. 
The big argument? Its this : how far away are we from seeing this in a game?
Although my knowledge of 3D graphics is somewhat limited to intermediate OpenGL coding, I have a very strong belief that this will not only be possible, but it will be done within five years! Other people, disagree strongly. 20 years, they say!
I ttotally understand the industry professionals who contend that such things are impossibly difficult on a games budget and timeline. I mean, they know what they’re talking about, right?
Yet, I still cannot help but think where we were 12 years ago, in the literal ‘genesis’ of computer graphics. You think this stuff is expensive now? I don’t recall the specifics, but I remember Lucasfilm saying that it took days or weeks to render that famous “Genesis flyby” scene in Star Trek II. Do you remember that?
ftp://ftp.luth.se/pub/misc/anim/tv/genesisp.mpg
It was awesome… at the time.
I can only think of those early-eighties CG experts neighing in disbelief at Doom3 today. “Because of X Y & Z! It will never work! Neverrrrr!” The Incredibles? “Not in my lifetime!”. And this proves how conventional thinking works. To think I had arguments in 1994 with a telecomms engineer about whether ADSL was possible. The engineers know their biz, they’ve done their homework, and life is good. Right?
Wrong.
Costs will drop. Hardware capabilities will increase. Software will get smarter. Libraries will expand. Refinement will increase. And everything will get cheaper and faster and all that. I mean, this is the tech business people!
I can’t predict the future exactly, but I know the folly in shrieking “impossible!” when faced with musings on future tech. When will people learn.
What do you think? Sooner or later?
20 Comments for this entry
Leave a Reply
now playing
Quake Live (CTF) by iD Software
December 14th, 2005 on 8:21 am
I think it'll be possible possibly within the next gen of systems, ie: ps4 and xbox3. After looking at the rendering of MGS4 I think anything is beggining to be possible.
December 14th, 2005 on 12:58 pm
I think making the big argument "How far away are we from seeing this in a game" is kind of selling yourself short. It's going to happen. We know it's going to happen. "When" is only relevant to the self: "Will I see this before I die?!"
I propose the big argument should read more like: What are the repercussions of having photo-realistic, augmented reality rendering at our fingertips?
Have you seen this?
I know you're focussing on video games, but seriously man. What is the cost of bringing this power to fruition? Think about it. All of a sudden that WTO protester is carrying a gun. All of a sudden the envoy tracks are in the sand. All of a sudden... you actually had to be there and see it with your own eyes to know what's real. Scary shit, man.
December 14th, 2005 on 5:12 pm
"How far away are we from seeing this in a game" is kind of selling yourself short.
I didn't invent what people were arguing about. I'm pointing out the controversy.
I propose the big argument...
Everything has a sinister side if you are looking for it. Even iPods could cause the destruction of society if you connect enough dots. "Everything is connected, mannnn"
I'd be interested in reading your take on it in your blog, should you care to explore it further...
December 14th, 2005 on 10:17 pm
One thing I never understood about the primitive computer graphics of movies in the 80s: how primitive they looked. At the time, I kept thinking "you're supposed to be 200 years in the future--you can assume that computer graphics are as cool as the phasers on the ship." Even then, I thought they should have just done traditional cel animation (which was cheaper, anyway) to look like the Enterprise isn't run by a bank of Vic 20s.
I suppose there was a certain mileage in being obviously computer generated, but if Tron could look both genuinely futuristic and obviously 'computery', why couldn't everything else?
December 14th, 2005 on 10:43 pm
Good point. My thought was that it was one part experimentation and one part marketing. Advertising the 'state of the art computer-generated effects' in those movies was a compelling reason to get you to go see them.
By the same token, we wouldn't be where we are today if it weren't for the earlier pioneers in movies and television. And that goes for the failures as well as the successes.
December 16th, 2005 on 8:48 am
It'll be a while before we see something like this in-game. Long before then there will be artwork and characters that look 'good enough' to make an impression and sell a story, but the hardware capability is only the final step in a long art-driven process.
I think a more important question is: Will game designers and artists be able to utilise this increase in processing power to create a compelling game experience?
December 16th, 2005 on 9:19 am
Oh, and to elaborate on my earlier statement regarding the time before we see high-quality images like this in games:
One has to look at what it actually took to make the image on the left. The model on the right is a control mesh for a subdivision surface, so the rendered geometry is many times more complex than the wireframe model would suggest. The surface shading model is also a great deal more complicated than what is currently used in realtime gaming engines.
Basically, take the time it took to render the final image in seconds and multiply it by 30, which is a passable framerate for a game. So, let's assume for the high quality image render that it took about an hour. That's 3600 seconds, or 108000 times slower than a 30 fps game. It's going to take a while for hardware to catch up to that.
Pixar's Tom Duff wrote a pretty entertaining post about this topic when NVidia released the GeForce 2 with claims that it was "a major step toward achieving that goal" (of Pixar-quality realtime graphics).
Tom Duff on the GeForce 2
December 16th, 2005 on 12:51 pm
Thanks for your comments. And that article was great.
Will game designers and artists be able to utilise this increase in processing power to create a compelling game experience?
I certainly hope so.
Or at least the marketing department will try and change what we consider is compelling in order to sell us more crud.
But seriously, I'm sure that there will be the usual good, mediocre and BAD offerings far into the future regardless of the tech.
Now back to the technology - I most certainly concede that you are more knowledgeable than I in this area, so I'll offer my musings most humbly. One might ask, however, is that the techniques used to model, shade and light this object (for example, SSS, or a poor man's version of it), did they even exist 5 years ago? How long did Luxo Jr. take Pixar to render back when it was new?
The heart of my argument, or more aptly described as a "rationale for my wild speculation", is that the tools, techniques and research propel the capabilities of the technology all the time, and sometimes faster than we realise. And far exceeding our ability to predict where it will go.
Predictions I've found that *are* safe : Never say impossible. And, that better technology always seems to arrive sooner than we expected.
December 16th, 2005 on 1:35 pm
I'll agree that the sort of image quality we're used to seeing today in hours-long renders is inevitable with realtime hardware. However, we're not even seeing Luxo Jr. quality graphics in games yet. I know this is an inflammatory statement, so allow me to qualify this.
Luxo Jr. looks as if it'd be no problem to reproduce now on current hardware. However, if one looks carefully at the finished film, you'll notice things like the absence of polygon edges, extremely smooth self-shadowing, anti-aliasing and motion blur that I have never seen in a videogame.
Much of the technology that is currently used in high-end rendering didn't exist 5 years ago. The same is true for realtime graphics. However, it's still not possible to get the same image quality on today's realtime hardware that was possible in a film made 19 years ago. Does this give you hope that things are going to improve dramatically in the next 5 years? 10 years?
December 16th, 2005 on 2:08 pm
I do see your wisdom, and I doubly thank you for sharing it. Not inflammatory at all, especially considering that you could grind me into digital pulp on the subject of computer graphics.
Yet your words actually do make me feel better about this level of realtime graphics.
Luxo Jr. was conceived and produced with the intention of it appearing on film. Thus, a massive projection screen where an extra level of care and attention to those features you've mentioned is of critical importance. But, whats "good enough" for film is quite entirely different to whats good enough in a video game. I'll concede that you sort of already said this when you stated "the good enough will come long before that level of quality".
The original "Warrior" image, though cunningly crafted and quite beautiful in its own right, was made in a can. Pre-rendered, nothing realtime. Perhaps one day enough corners could be cut to produce something that was almost as good (nay, good enough?) without us even realising it. And certainly not designed to be blown up 100 times in size.
Now put that character in motion, have him shout and shoot arrows at you, and then he's shown in an entirely different light. The hand-crafted quality of something that was produced frame by meticulous frame becomes somewhat less important in that context. Not quite the "Pixar level" of production, but good enough and good enough *sooner*.
December 16th, 2005 on 2:28 pm
Well, speaking in terms of what is "good enough" isn't the point you were making in your article. It's about "seeing this (quality of graphics) in a game."
I do believe that what is required to make a story point without being distracting is well within reach on today's hardware. Photorealism is not the magic solution that makes all story points believeable. One could make in-game graphics the quality of say, the robot in 'Iron Giant" quite easily and have it tell a decent enough story.
However, if we can compare apples to apples and hold up that rendered image as the bar that games should reach without rendering shortcuts, context or other graphic hand-waving, I'll put a stake in the ground and say that we won't see it in less than 20 years. Probably longer.
December 16th, 2005 on 2:31 pm
Oh, and the number of pixels rendered for a frame of Luxo Jr. is very close to the HD resolution that the XBox 360 supports. More food for thought.
December 16th, 2005 on 3:00 pm
Ah, but you are positing that the same techniques used to generate jaggie-free, motion-blurred characters with rich surface geometry will remain constant for the next 20 years, and its up to the silicon to 'catch up' to just crunch faster. Its plain from my inductive point of view that this will not, and cannot, be the case. It defies how the computer graphics industry (and technology as a whole) blossomed in the first place. If nothing advances, and the tools and technologies don't see any sort of innovation, then we're down to pure silicon and 20 years would be correct.
Its like the consistent claim in my industry where the strength of cryptographic algorithms are almost wholly judged by a combination of Moore's Law and distributed brute-force processing. It *is* foolhardy to judge things based on speculative breakthroughs in cryptanalysis, much as I'm doing in this, yes. But we don't ignore it on the basis of a vapourous promise of some unknown collision technique - we mitigate the threat by constantly producing and edifying the alternatives. DES to AES, SHA-1 to SHA-256, etc.
So while I fully admit that speculation doesn't give us better graphics, it does allow me to look at the past and measure the future. Again, your technical argument is totally sound : 20 years based on today's techniques is completely correct. But are you using the exact same techniques and software that you used even 5 years ago?
December 16th, 2005 on 3:08 pm
And also, you said that Luxo Jr. isn't close to being possible in realtime because of its inherent complexity (jagfree, etc). And I'm saying that the level you're citing isn't even on the same level as the image that I posted. So if the image above is the "good enough" that I'm talking about, and not necessarily an exact highdef, realtime rendition of Luxo Jr., then 5 years seems totally reasonable in the face of my other bits of arm-waving.
Do you still think I'm totally and completely wrong?
December 16th, 2005 on 3:48 pm
Actually, I spoke of nothing other than what is evident in the image itself. The spatial and temporal antialiasing, self shadowing, and surface smoothness between a frame of Luxo Jr. and any realtime game can be compared without any discussion of technical methods used and the differences catalogued. If one then compares Luxo Jr. with that warrior image, many of the same qualties exist in both images. Both surfaces appear to be smooth with no polygon edges, there isn't any spatial aliasing apparent, both self-shadow without any apparent artifacts. The Warrior has some subsurface scattering component in his shading that Luxo Jr. doesn't have, though it adds complexity and time to the final render.
I'm not just counting on the silicon to crunch faster. The code used to generate these images cannot be the same (though interestingly, the Reyes algorithm used in the Luxo Jr. images remains relatively unchanged to this day). If I can't find the same quality of an image rendered 19 years ago in a realtime image made today, I have no reason to think there's some algorithmic bomb lurking around the corner that will make images of Warrior quality possible within a few years.
If we can now talk of what would be "good enough", I stand by my assertion that in many cases we are capable of this now. It's just that "good enough" in every application isn't dependent on graphics, but context.
December 16th, 2005 on 4:45 pm
Point taken regarding the comparison of "Warrior" with Luxo. Although from my untrained eye, I do detect some artifacting in the chainlink, and some problems with the skin on the forehead. I would say, with all things being equal, Luxor and Warrior are of a different quality but not more or less than the other (for different reasons). Warrior might look positively awful on a 30 foot projection screen. Luxo doesn't have the texture detail.
So, would you say that a single frame of Luxo took an equivalent amount of time then as a single frame of Warrior today? I would guess that Warrior was produced much, much faster since the tools and techniques are so much more known today. Thats obviously progress in the static space, although the techniques that afford us realtime are obviously completely different.
If I can't find the same quality of an image rendered 19 years ago in a realtime image made today...
There are certainly plenty of 19-year-old CG effects that are done in realtime now, and look even better. The original Genesis flyby (circa 1982?) that I cited in the original post is one such example. Young Sherlock Holmes, Tron, The Last Starfighter, etc. Is that what you're talking about, or are there better examples from that time period that cannot be done? I confess that I don't really remember.
And when Reyes was introduced, whenever that was, was it not the "algorithmic bomb" of its time? And I am sort of saying that something that is better or equally important will come into fruition over the next few years, only because its the nature of the beast. Not explicitly that, obviously, but one leap or another in different problem spaces. Even at the rate of merely incremental improvements in tools and tech between now and in 5 years, I'd say it almost guarantees that we'll get the close enough.
And if you say that "close enough" is already here now, then I will say "closer enough" is the Warrior image, and set that as the goal, versus what we see today in Zelda and MGS. I guess we'll see what happens. And of course the gaming context will justify our attention to it (unless its an EA game, whereas thats all that would really matter).
December 17th, 2005 on 11:56 am
One of the earlier points Headcrash was making was lost in the technical side of this argument, which is that the economics of video game making don't favor the warrior appearing anytime soon, even if some lab somewhere demonstrates its feasibility.
As EA demonstrates, games are already moving to a factory/studio approach where the overall production values are managed within a multi-million dollar budget. The warrior has to be not only technically feasible, but creatable under those conditions. Until it's not just possible but relatively easy for a team of programmers/artists to put that level of quality in, we won't see it.
WoW had a staff of hundreds of artists, and that was only to achieve a sort of cute, budgetarily sane aesthetic.
December 19th, 2005 on 1:41 pm
The agenda here is clearly on the technicalities of current and future rendering techniques to achieve photo-realistic gaming experiences. So you are very right that I have no right cloud the thread. I would maintain however that the social exploits on real time, augmented reality are there. I will stick to my own blog with these social engineering issues and not intrude with non-game points!
I am appreciating the very cordial language used in this thread and am looking forward to more insightful information.
December 19th, 2005 on 2:09 pm
I agree that it's unfair to compare full rendered film stock to game renders. (Mr. Analogy alert) It's like comparing records to mp3s! Removing the most amount of non-essential information has never been a real rendering issue as is always has been with 3D game environments, has it?
I've played an auto-crash something game on a console recently that seemed quite clever in the way it simulated speed blur and even slightly changed the FOV as you accelerated to trick the mind into thinking you were travelling faster. Maybe the solution doesn't lay in hardware horsepower but instead in tricking the brain? Doom II looked 3D but it wasn't... maybe you don't actually need photorealism?
Have you heard of the technology where feeding real-time audio representations of an environment to blind people can actually make them "see"? (I can't for the life of me find a link). The brain is amazing at adapting - Perhaps another area of investigation for you is to look at the grey-matter side to see what we need as humans to feel completely immersed in a 3D environment?
December 24th, 2005 on 5:20 pm
The comparison between film and in-game renders is invariably made by the proponents of real-time rendering, never the other way around. I agree that it is unfair, but it's a game that realtime developers are continually asking people to play.