Why Researching Workflow Matters
Autodesk has a fun quiz on their site called the Fake or Foto test. In the quiz, you have to look at each image, and guess whether it is real or a render. Surprisingly, I aced it the very first time!
After I took the quiz, I went back and documented all the things that made me think a particular image was real or fake. That way I could see what it was that clued me so that I might improve my own work. In almost all cases, it wasn’t that the CG images were obviously wrong in any way, it was simply that the real images had more “detail” in them.
For years, the limitation has been what the software could do, but I’m not so sure that’s true anymore. In games especially, production costs have gone up as graphics have become more realistic simply because making detailed images is very time consuming. The relationship between time, money and graphics detail is linear. The problem, of course, is that neither time nor money are infinite. Limited budget means limited production cycles, which means limited time to spend adding things like scratches, scrapes and nose hairs. Eventually, we’ll reach a point where we can’t add more detail not because we’re out of RAM–we can’t add more detail because we’re out of time! Truthfully, I think we’re on the horizon of that already.
While I’m certain that there are still numerous improvements to be made in traditionally researched areas such as lighting, I think technological improvements to workflow have more potential to improve the quality of computer graphics in the near future than virtually any other area. Automated procedural geometry, better methodologies and tools that allow for less restrictive workflow patterns–such as recent advancements in re-topology interfaces–can allow us to complete the same job in less time and with far less effort. We know the artists and software can do it; now we need to make it practical.
For the curious, here are my impressions from the Autodesk quiz:
** WARNING ** Spoilers follow. You might want to take the quiz yourself before you read my analysis of the images.
The Car: Fake
What Gave it Away: Ok, I have to admit I have an unfair advantage on this one. I’ve worked on a half dozen racing games in my career, and I’ve seen a heck of a lot of CG cars. This is a really good image. Interestingly, my first tip-off that it wasn’t real wasn’t anything with the rendering itself: it was the composition! I don’t know why, but over the years car-on-a-cobblestone street has become nearly as ubiquitous as chrome sphere on a checkerboard. The real give-away though was the distance between the fence and the background elements. I’d bet dollars to donuts the buildings behind the fence are a 2D photo.
The Coffee Mug: Fake
What Gave it Away: This image was the easiest of the bunch to discerne. It suffers from a very common problem with CG images: the “too shiny” effect. Reflections can add a lot of realism to your scenes, but its easy to over-do it. Fewer reflections and a tighter lighting setup would improve this image significantly.
The Corkscrew: Real
What Gave it Away: On the far right edge of the main cylinder, there is a raised imperfection in the metal. In addition, the brushed metal behind the corkscrew is very difficult to do well in CG because of the unusual way the tiny lines distribute the light. I figured this one had to be real.
In recent years artists and developers have made great strides in making things look more real by making them less perfect, but we still haven’t reached the point where most images have intentional defects on this scale. Again, it’s less a technology issue, and more a time and workflow issue… in a production environment, there just isn’t time to add scuffs to every screw. I wouldn’t be surprised however if we start seeing major advancements to add exactly those kind of details in a completely automated fashion in the very near future.
Diamonds: Fake
What Gave it Away: The lighting is too uniform for real diamonds. Real diamonds have very complex refractions. These only seemed to have one or two levels of internal lighting on the gemstones.
Gemstones are an absolutely fascinating subject when it comes to lighting techniques… real gemstones are all about light! If you’d like a deeper exploration of the math, Richard Hughes has written an interesting article about the phenomena of brilliance, window and extinction in gemstones.
Bolt: Real
What Gave it Away: This image was very hard to tell. The graininess of the focal blur implies CG, and there’s an almost identical render done by Jaime Vives Piqueres in POV-Ray, so I knew it could be done. But, something about the pitting on the head suggested this one was probably real. The top of the bolt is smooth on the edges and pitted in the middle. Again, while there’s no technical reason this kind of detail can’t be added to a CG image, the time requirement means it usually isn’t, which is why I figured this one must be real.
Egg Beaters: Fake
What Gave it Away: The “whiteness” of the reflections. I have this problem with my own images sometimes, aargh. 🙂
Car Hood: Real
What Gave it Away: Like the coffee mug, this is another one I just knew from my gut immediately. The micro-distortions in the reflections, the angle of the shot, and the material qualities of plastic-chrome all made this image seem unlikely to be CG. (So I guess I know now what I need to add to my images!)
Wood Monkeys: Fake
What Gave it Away: The wood grain seems unnatural, something is out of scale about it. The monkeys are too similar to each other.
Texture scale is a very important visual cue that a lot of artists miss. An otherwise flawless texture will look far less realistic if it is too large or small. This is a common problem in video games, particularly, where textures may be reused on different objects to save on memory. If you can’t scale the texture because of system requirements, a handy trick that sometimes works is to scale the UVs instead.
Glass Chips: Real
What Gave it Away: Wow, bubble mania! Even with macros, this shot would be difficult. The diffusion on the glass (gel?) also seems to have some sub-surface scattering and other complex lighting that most renderers would be hard-pressed to replicate with precision (yet).
Green Peppers: Real
What Gave it Away: This one was tricky, because I think a render of this quality could be achieved. I knew most CG artists would make at least some of the peppers identical, or at least morphs of each other, so I immediately started looking for repeats. After a moment of study, it was clear they were all unique, and even the bottoms had a lot of fine detail. I figured the image must be real. 🙂
Thoughts? Comments? Tips and tricks for adding realism? Leave a comment!
Tags: auto desk, cg, cgi, computer graphics, photo, photorealistic, procedurals, realism, render
December 1st, 2007 at 9:47 am
I took the quiz and got 8/10… I blew it on the two car images, totally had them flipped. Doh!