How close are we to truly photorealistic, real-time games?

[dlevere]

By Kyle Orland

Every graphical and technical advance the game industry has seen from Pong to Crysis has been a small step toward the end goal of a real-time, photorealistic 3D world that is truly indistinguishable from a real-world scene.

Speaking at the DICE Summit Thursday, Epic Games founder and programmer Tim Sweeney examined the speed and direction of computing improvements and determined that we "might expect, over the course of our lifetime, we'd get to amounts of computing power that come very close to simulating reality."

[BeyondTheStatic]

I'm sure there must be a better example than that Crysis 2 screen shot. It doesn't even look like they used any real global illumination (radiosity) scheme at all... just ambient occlusion baked to textures which is far from convincing, IMHO. The article mentions "two bounce" lighting, and to me that means radiosity, not AO. But that example image screams "I use ambient occlusion!" I could be wrong.

To look photorealistic, future games will definitely need good radiosity algorithms. The sampling must be tight and influenced by all objects, within reason. They'll also need subsurface scattering, truly reflective/refractive objects, reflected/refracted caustics (like the concentrated light you see above/under pools), and volumetrics (clouds, rays of sunlight through fog), among other things.

And of course you'll need realistic interactions with the world, such as fully destructible environments. If you blow a building up, it should produce not just a few chunks and some billboard smoke, but pieces ranging from large to medium to small to infinitesimal. Volumetrics could be used with a fluid sim to make the dustier parts of the exploding building approximate a true physical occurrence.

A physical simulation will not just be needed for blowing things up, but for things like walking through bushes, kicking around all the little pebbles as you move about, jumping into water and tons of other things.

Also, in addition to traditional polygons, future game engines will probably use voxel elements for many things like water, smoke, sand and who knows what else.

Many of these things are already being used, but not always well.

I expect the industry will get pretty close to real-time photorealistic scenes in my lifetime. And some day, in the far, far future, somebody will make a complete physical simulation (of limited size), down to the subatomic level, once we figure out why things are chaotic and appear random

[dlevere]

On March 1, 2010, a new video of the CryEngine 3 was released for the i3D 2010 symposium, which demonstrates 'Cascaded Light Propagation Volumes for Real Time Indirect Illumination'.

Crytek explains that "the LPV technique was extended to support secondary occlusion and multiple bounces. The propagation scheme is improved and compared to similar Discrete Ordinate Methods."

[BeyondTheStatic]

Flash is currently disabled, but I was just looking at some screenshots for CryEngine 3. Pretty impressive, though I have a feeling they're using AO to support multiple light bounces, not just using LPV to support AO and GI. Too much AO will make everything look unrealistically dirty and shadowed in the wrong way.

I noticed their SSAO implementation brings out the outside edges of surfaces, not just the inside ones. That can be useful for adding abrasion to things and sticking dirt in corners. But using it to enhance the lighting is just wrong in the long run.

[Pyriel]

I love the caption on the picture: "While this scene from Crysis 2 looks pretty good, in a few decades it's going to look like outdated crap." Really, decades? The only way almost any technology from today won't look outdated in a few decades is if 2042 is a post-apocalyptic hellscape.

[dlevere]

Far Cry 3 Gameplay HD