It's hard to take anything Hideo Kojima says about Metal Gear at face value, but there's a chance he was speaking out of sincerity when he told the audience at PAX that the Metal Gear Solid: Ground Zeroes demo was running on a PC that parallels current-gen hardware. Of course, there were some elements that looked next-gen, but for every wrinkle, reflection, and raindrop, there were tells that yes, in fact, a PlayStation 3 or Xbox 360 can produce graphics that look this good.
Before delving into the finer technical points, the direction and composition of the demo at large should be our first point of examination. Hideo Kojima is a self-proclaimed film buff and is known to obsess over camera angles and lighting during cutscenes. These aspects alone aren't responsible for the visual quality of the Ground Zeroes demo, but when done right, lighting can significantly affect the perceived realism and accuracy of objects within a scene.
Global illumination is a lighting technique that has only gained a foothold within game engines over the last few years, but it dramatically affects the perceived realism of a scene. By attempting to re-create the way light "bounces" from one surface to another, scenes are increasingly looking less like collections of unrelated objects. A few months ago, Kojima Productions released images displaying the prowess of the Fox Engine, prominently featuring its ability to employ dynamic lighting and global illumination, as seen below.
Of course, where there's light, there's shadow. This is another tool in Kojima's kit that can be used to convey realism, as the counterpoint to light, but in the right hands, it can also be an effective resource management tool.
To elaborate, the majority of the background plane during the first half of the demo is poorly lit and clouded in mist. Spotlights are also used to great effect in the demo to create contrast, further drowning out the already obscure background by drawing our focus into their light. This limits the viewers' instincts to scrutinize details in the background, and gives the artist an excuse to omit details from the scene that would take up valuable RAM and processing cycles otherwise. It's not until the player begins to control Snake that we get a full view of the military base. At that point, however, there is very little in the fore- and mid-ground. By limiting the detail in the background, artists can use the freed-up overhead to increase the details of objects in the fore- and mid-ground, and vice verse.
It may surprise some people to find out that even simple models with modest polygon counts can achieve high levels of detail once the image completes its rendering passes. Based on its work on Metal Gear Solid 4, as detailed in a case study from 2008 published by Autodesk (now archived at MGSForums), Kojima Productions is truly adept at getting the most out of its models. Its techniques aren't unprecedented, but what we can surmise is its talent at scaling up quality without drastically impacting computational resources. This is primarily achieved through the creative use of textures by the studio's artists.
Most people are probably familiar with textures, or more specifically, texture maps: 2D images that are applied to the surface of a 3D model. The most commonly used texture maps, diffuse maps, define the default color of an object and its various details. Of course, texture maps come in a variety of flavors, characterized by their particular target attribute. Here are two of the primary suspects responsible for the perceived level of detail in the Ground Zeroes demo.
Specular maps: Artists use specular maps to define the way lighting affects 3D models. Altering the value (where white is "high" and black is "low") of the details in a specular map will increase or decrease the potential influence of light on the models' default color.
Normal maps: To put details into a model without increasing poly counts, normal maps dictate the slope of a surface, usually generated from a high-res model and applied to another version with simpler geometry. Rather than adding or displacing the geometry of the model in question, normal maps define the interaction of light with the model, and ultimately, the camera. Both of these map types are used to great effect in the XOF patches that are thrown out of the helicopter, just after the 7-minute mark of the demo. Of course, the same instance also reveals the inherent deficiencies of these techniques and the target hardware.
The great thing about normal maps is their ability to add relief to a polygon's face without adding extra geometry, but what they can't do is actually change the geometry proper. This is why you see crisp edges and angles on the XOF patch, rather than smooth curves. In order to transform visible edges into curves, the game must support tessellation. By the time Metal Gear makes its debut on the next generation of consoles, we can look forward to tessellation and other techniques that will have a significant impact on in-game graphics.
Tessellation: The process of dividing geometry, resulting in a mesh with increased geometry and vertices, allowing for greater detail potential and surface manipulation.
While the simple model sits in RAM, tessellation divides its geometry during processing. This method requires fewer resources than tracking a traditional, high-poly model.
Once tessellation is applied, artists can then move forward, replacing normal maps with displacement maps. Where normal maps dictate the angle at which light bounces off a surface, faking detail, displacement maps actually deform a model with enough vertices to deform, creating detail rather than implying it.
Displacement maps: A 2D image that translates surface elevation to a 3D model, deforming vertices and edges according to the maps' features.
Displacement maps are great, but they require a lot of geometry in order to properly displace details. In some cases, console games are already using displacement maps. It's highly likely that there are elements within the Ground Zeroes demo that feature displacement maps, but it's clear that objects with stiff edges don't have the right ingredients, such as tessellation, to take advantage of it.
Kojima did state that the demo was targeted at the current generation of consoles, so it's not news that we're inclined to believe him, but with such impressive visuals on display, discovering current-gen problems within the demo gives his comments credence. The final product will have to account for the biggest variable in any game, the player, which means Kojima and company won't be able to rely on camera and lighting tricks all the time. But if the demo is any indication of the potential for Ground Zeroes' graphics, then it looks like there is still room for other developers to squeeze a little more power out of our six-year-old consoles.