Games that look like CGI
- Thread starter Shifty Geezer
- Start date
- Status
Seen the progress realtime graphics has made in the last few years, and the hazy idea of what exactly is CGI, shouldn't the distinction really be between Realtime CGI and Renderfarm CGI?
This is essentially what is being discussed here, no?
This is essentially what is being discussed here, no?
D
Deleted member 11852
Guest
Finally, John 117 is revealed!
Shifty, Sure I know of all the shortcuts and aproximations both Game engines and CGI animation use. Again, coming up with a perfect description of what should be considered "CGI" and what shouldn't would be just a boring semantics discussion.
I think that's not necessary. I believe everybody understood that by CGI people mean something like the great movies Laa-Yosh produces, or a Pixar feature.
Now my point still stands, there is not much fun in using exemples that take advantage of extremely constrained designs, otherwise we could just point to any FMV based game and be done with it.
I think that's not necessary. I believe everybody understood that by CGI people mean something like the great movies Laa-Yosh produces, or a Pixar feature.
Now my point still stands, there is not much fun in using exemples that take advantage of extremely constrained designs, otherwise we could just point to any FMV based game and be done with it.
L. Scofield
Veteran
Anything with high quality object-based motion blur and good AA.
I can only second Milk's opinion, we should really try to draw a line somewhere and at least broadly define what we consider to be "CGI". Most people seem to refer to movie visual effects when they talk about CGI, so one major topic should be realism; but another one could be the look of CG animated features from the Disney/Dreamworks school. So I believe we should focus on games that are trying to achive such results.Then the discussion could be about identifying the artistic and technical aspects of games that do manage to look better than the rest, and also about exploring where there's still room for improvement even on today's hardware.
Another interesting case is of course creating convincing looking digital characters, but I think we already have a thread for that. However there are other similarly important elements in a game world, from special effects (water, fire, smoke and so on) through vegetation to more interactive environments (like destruction and such).
Another interesting case is of course creating convincing looking digital characters, but I think we already have a thread for that. However there are other similarly important elements in a game world, from special effects (water, fire, smoke and so on) through vegetation to more interactive environments (like destruction and such).
The original goals of the development of Renderman have included those, obviously; but another important and sw architecture defining element was real-world scene complexity.
This is actually a really important point, because all of these require very powerful hardware to achieve - and even offline rendering had to develop complex technologies to achieve them on lots of render nodes. Aiming for similar results on a very limited hardware platform with real-time rendering times and full interactivity is a huge undertaking, it requires very complex and advanced technologies, and that would be a very interesting topic for discussion
L. Scofield
Veteran
How complex does a scene need to be? If you added OBMB to say, Captain Toad it would look CGI to me.
The original definition of scene complexity included both the possibly huge number of objects in a scene, and also their detailed geometry. Basically anything that can appear in a movie - landscapes, cityscapes, crowds and so on; and also close-ups of realistic characters and any other item.
The first versions of Renderman used two main techniques - also closely connected - to deal with this complexity.
First of all, every piece of geometry is tessellated into micropolygon grids with usually at least one vertex per pixel (but the "shading rate" parameter can increase or decrease this number anywhere), even completely flat surfaces, and only the vertices are shaded. This has also immediately made displacement mapping very cheap, as it only takes a little extra work to move those vertices before shading and rendering.
The other big thing was a major implementation of "divide and conquer". The final image is split into quadratic tiles (buckets in offline rendering) and ALL the scene geometry is sorted and split iteratively until every piece is small enough to fit into one such tile. This way the actual amount of data required for the rendering of the bucket can be kept to a manageable level that would fit into RAM.
Another major technical innovation was how the shaded grids were then sampled for the final rendering, this has allowed for high quality antialiasing, motion blur and depth of field (both of which are actually possible to render in 3D, instead of using 2D post-processing trickery).
If any of you are interested in more details, this is the paper to read:
http://www.researchgate.net/publication/265004726_Advanced_RenderMan_2_To_RI_INFINITY_and_Beyond
I think this should actually be sort of a pre-requisite for meaningful discussions here
Don't worry, it's neither too long nor too complicated, and in fact is a pretty fascinating read - especially considering that it's nearly 15 years old now!
I do agree, which is why I specified something that can show at Cannes or a similar film festival. Oh, and in 3D. there are plenty of 2D animations out there on computer, but CGI pretty much means 3D in vernacular even if the definition isn't so distinct.
I think that's too small a target, aiming for the big budget AAA creations. There are lots of indie animations and smaller studios and short works that can't afford the fidelity of the movie industry.
Here's a playlist of 200 of them, and I'd say any game that can fit on that list without looking out of place counts as looking like CGI :
By and large they seem to go for the Pixar/Dreamworks look, but there are others branching out, like number 7. 34 looks very akin to Source Filmaker, (and uses 2D painted clouds in the opening, not 3D raytraced volumetric clouds using an accurate water simulation
). There might be some turkeys on that list though - I haven't watched them all! Hence why I'd rather set the bar at something that's screened for quality like getting shown at Raindance. This channel seems a better selection criteria as the quality bar is clearly high.
https://www.youtube.com/channel/UC-1rx8j9Ggp8mp4uD0ZdEIA
That wasn't the intention! This isn't a technical thread and though your interpretation is a valid one, that's not what it was meant to achieve. It's just a celebration of games that have moved away from looking obviously like computer games and into a realm where it's much easier to forget they're games and appreciate them more like well made animations. Final Fantasy X looked great until we hit the real-time graphics and then the jaggies and shimmer caused brain-haemorrhaging - the game didn't look like CGI. We're getting to a point where games can, and that's a good thing! I think it worth making a note of games that make that subjective transition away from games into...art? TV?
My benchmark for "Games that look like CGI" in curent gaming market is anything that comes close to replicating this
Game does not need to have this level of asset/effect quality and density, but it needs to be spotless with its presentation with no obvious signs of rendering shortcuts [aliasing, low poly world/charachters, poor lightning, poor materials, jerky animations...].
So far, only Order 1886 and DriveClub [during car movement] qualify.
Game does not need to have this level of asset/effect quality and density, but it needs to be spotless with its presentation with no obvious signs of rendering shortcuts [aliasing, low poly world/charachters, poor lightning, poor materials, jerky animations...].
So far, only Order 1886 and DriveClub [during car movement] qualify.
Okay I'm totally humbled again That was our first job with the Arnold renderer, but we've moved to full raytracing a few years before that.
Another interesting issue is that nowadays there are many games that have assets with a much higher quality. For example Ezio's head model was somewhere along 5 thousand quad polygons or so, which is easily surpassed by many games like UC4 or Infamous 2. Yeah we've had subdivision and displacement on top of the base mesh, but the quality of the facial deformations was determined by the base model.
On the other hand, even at that time we've used global illumination, reflections, shadows, cloth sims and hair rendering that are far superior to anything that a realtime renderer can even dream about nowadays. Not to mention the AA, texture filtering, motion blur and such.
But funnily enough, today's most advanced games have practically nothing to be ashamed about when viewed next to that 5-year old movie. Games have come very very far in that time!
That was our first job with the Arnold renderer, but we've moved to full raytracing a few years before that.Another interesting issue is that nowadays there are many games that have assets with a much higher quality. For example Ezio's head model was somewhere along 5 thousand quad polygons or so, which is easily surpassed by many games like UC4 or Infamous 2. Yeah we've had subdivision and displacement on top of the base mesh, but the quality of the facial deformations was determined by the base model.
On the other hand, even at that time we've used global illumination, reflections, shadows, cloth sims and hair rendering that are far superior to anything that a realtime renderer can even dream about nowadays. Not to mention the AA, texture filtering, motion blur and such.
But funnily enough, today's most advanced games have practically nothing to be ashamed about when viewed next to that 5-year old movie. Games have come very very far in that time!
L. Scofield
Veteran
The original definition of scene complexity included both the possibly huge number of objects in a scene, and also their detailed geometry. Basically anything that can appear in a movie - landscapes, cityscapes, crowds and so on; and also close-ups of realistic characters and any other item.
The first versions of Renderman used two main techniques - also closely connected - to deal with this complexity.
First of all, every piece of geometry is tessellated into micropolygon grids with usually at least one vertex per pixel (but the "shading rate" parameter can increase or decrease this number anywhere), even completely flat surfaces, and only the vertices are shaded. This has also immediately made displacement mapping very cheap, as it only takes a little extra work to move those vertices before shading and rendering.
The other big thing was a major implementation of "divide and conquer". The final image is split into quadratic tiles (buckets in offline rendering) and ALL the scene geometry is sorted and split iteratively until every piece is small enough to fit into one such tile. This way the actual amount of data required for the rendering of the bucket can be kept to a manageable level that would fit into RAM.
Another major technical innovation was how the shaded grids were then sampled for the final rendering, this has allowed for high quality antialiasing, motion blur and depth of field (both of which are actually possible to render in 3D, instead of using 2D post-processing trickery).
If any of you are interested in more details, this is the paper to read:
http://www.researchgate.net/publication/265004726_Advanced_RenderMan_2_To_RI_INFINITY_and_Beyond
I think this should actually be sort of a pre-requisite for meaningful discussions here
Not too long? That's an entire book xD. I think in terms of this thread the definition of what "looks like CGI" is more dependent on the level of graphics one is used to rather than the actual technologies behind the graphics.
To one person, "looks like CGI" means "Jurassic Park level of fidelity"; to another, it would be "PS1 Final Fantasy FMV level of fidelity" and so on.
The one thing to me that always separated realtime vs cgi was the use of extra effects such as motion blur and a smooth IQ. Now we can have them in realtime too so YAY.
The Renderman architecture related chapter is only something like 22 pages
Not that the rest isn't interesting either - for example the Dinosaur movie stuff talks a lot about using displacement mapping and shadow maps, quite relevant for today's video games.
It's also a pretty interesting read because at the time it was written, computers were still either 32-bit Pentium 3 level PCs or now extinct 64-bit Silicon Graphics MIPS systems or DEC Alphas or such. Back then we didn't have things like affordable raytracing or SSS, most of the content was built with NURBS geometry instead of multi-million polygon meshes, and so on.
Thus, funnily enough, most of the theory there is now obsolete for offline CGI - however, it can be quite relevant for realtime rendering with its limited resources. It's not surprising that Tim Sweeney became quite interested in micropolygons a few years ago
Not that the rest isn't interesting either - for example the Dinosaur movie stuff talks a lot about using displacement mapping and shadow maps, quite relevant for today's video games.
It's also a pretty interesting read because at the time it was written, computers were still either 32-bit Pentium 3 level PCs or now extinct 64-bit Silicon Graphics MIPS systems or DEC Alphas or such. Back then we didn't have things like affordable raytracing or SSS, most of the content was built with NURBS geometry instead of multi-million polygon meshes, and so on.
Thus, funnily enough, most of the theory there is now obsolete for offline CGI - however, it can be quite relevant for realtime rendering with its limited resources. It's not surprising that Tim Sweeney became quite interested in micropolygons a few years ago
Fair enough Shifty, this was your thread idea after all, and I do think it is a very cool point of view for discussion. But as Yosh, I initially though it was meant as way to evaluate how close the more modern engines and top studio's art manages to get to the state of the art offline productions within the interactivity requirements of a game supposed to run at real time on consumer level hardware. With that kind of intent, games that take heavily abstracted art styles or design restrictions like fixed cameras, would do little to add to that kind of comparison.
But if your original idea was more broad, and more about finding games that manage to hide obvious game-visual-like traits through either tech or smart art design and aesthetical choices, I'd say that that's also an interesting topic. Resolution aside, I'd say a couple older games that chose heavily stylized looks like okami got somewhat close, though were still hindered by obviously low poly-counts or low-res textures way too often...
The latter Mario games on Wii-U are also, aside from cheap post-effects and little to no anti-aliasing, quite close to pixar-looking for the average non-technical viewer, and if we really push it, I'd say even galaxy on the original Wii was not that far. This has more to do with the target visual of those games being less ambitious than something like Assassin's Creed II, than with pure technical accomplishments, though those are definitely part of it too.
Laa-Yosh, that's a very cool book. Thanks for yet another educational piece. Appreciated as always!
But if your original idea was more broad, and more about finding games that manage to hide obvious game-visual-like traits through either tech or smart art design and aesthetical choices, I'd say that that's also an interesting topic. Resolution aside, I'd say a couple older games that chose heavily stylized looks like okami got somewhat close, though were still hindered by obviously low poly-counts or low-res textures way too often...
The latter Mario games on Wii-U are also, aside from cheap post-effects and little to no anti-aliasing, quite close to pixar-looking for the average non-technical viewer, and if we really push it, I'd say even galaxy on the original Wii was not that far. This has more to do with the target visual of those games being less ambitious than something like Assassin's Creed II, than with pure technical accomplishments, though those are definitely part of it too.
Laa-Yosh, that's a very cool book. Thanks for yet another educational piece. Appreciated as always!
I agree, which is why I think it noteworthy we've overcome those shortcomings. Rogue Galaxy looked like a cartoon on PS2, except it didn't due to the IQ, so it was obviously a game. Up until Drive Club, every racing game I ever saw a screenshot for (sans maybe photomodes) looked immediately like a computer game to me, no matter how pretty, but DC can sometimes shed that and look like one's watching a prerendered car scene.
I'd say a more objective analysis would put the requirements to 'look like CGI' as needing to hide the obvious shortcomings of realtime rendering. You'd have to eliminate in-your-face-geometry (or hide it in the art style!), jaggies and shimmer, pixelated shadow maps crawling across the screen, and maybe dirty up the output to remove that cold, hard look of games. I'd also say there's something else which I can't put my finger on, the lighting maybe, where high quality visuals still look like a computer game to me, but that may be content or something else. FPSes like BF4 look like games no matter how high the quality is cranked up to. Maybe it's attention to detail in the lighting?
I think you need to use another term than CGI for what you really want to say, because a Mandelbrot is clearly and undebatable CGI, but's not what you mean. CGI = Computer Generated Imagery, that is literally all stuff that the computer transforms from [nonvisual] data to something visual.
I'm using the vernacular application. Realtime in-game graphics are CGI too, by pure definition, but over the years to make the distinction between general computer graphics and 3D rendered graphics, the term CGI came to be used, so we have game graphics, computer animation (2D, flash animations etc.), computer art (static), and CGI. AFAIK there is no alternative. Does anyone know any other short-form for/meaning 'non-realtime computer generated animation'?
- Status
