Author - James Dargie

In general, there are fundamental differences between Movie and Game generated assets.  A primary concern is polygon count and efficiency.  Currently the only way to model in video games is by using polygons, which can require a denser mesh to emulate smoother or more natural looking models such as humans and animals.  NURBS models can be created, but need to be converted and optimized to polygons for use in the game.  In pre-rendered movies, any technique is allowed to create your models.

Movie models can be generated up to millions of polygons using several different techniques at once.  A model consisting of NURBS and polygons as well as subdivision surface models is normal and completely acceptable.

Gaming models have to be more efficient in their use of modeled details to maintain a manageable data set to render.  The reasoning here is that an efficient streamlined environment composed of the lower poly assets will render more smoothly and give better frame to frame renders during gameplay.  What your gaming system is in essence, is a renderer that constantly has the task of rendering each frame of gameplay at 30 frames per second.  Some games hit the magic number of 60 frames a second.  If this rate drops during the game the result is a poor experience and hampered gameplay.  This applies to PC games as well, although they will typically have more processing power to run higher resolution models.

With constant innovations and improvement in next-gen consoles and technology, development of more advanced techniques and processes give us more detailed looking models at a lower cost.  One of these advances is the use of normal mapping.   A normal map acts like a bump map, in that is adds surface detail without adding polygons.  Normal maps go a step further because they actually replace the surface normal with new multi-channel data to represent an X, Y, Z coordinate system.  What this means is that we can create a high resolution model of 2 or 3 million polygons and bake the high resolution detail down to a normal map that retains the component space data of that high resolution model.  It is then a process to create a streamlined model that emulates the general proportions of the high density model, but at a much more efficient poly count of 2500, for example.  Once the normal map data is applied to this low-res rendition of our high-res monster, the model immediately looks more complex geometrically but at an affordable rendering cost.  Movie productions also use Normal Mapping techniques, but the asset that they use the Normal Map on is typically a more detailed model than the one used in games.

In this game example, the scene calls for a wine barrel to be created.  It will be duplicated many times in the scene, so it needs to be efficient.

We start by building a low-res model of a few hundred polygons:


Lo-res model*

Then a higher resolution asset is created that has more detail and visual interest.  This model has many more polygons and modeled details, running up into the 10’s of thousands.

From this higher resolution asset we bake out our normal map and apply it to the lower density asset.  The result is seen here:

Low-res model with Normal Map*

The final in-game asset, completed with a color and specular map, can be seen here:

Final game render

What we have is a high resolution looking in-game asset that renders like a low resolution asset. 
           
Another difference between movie and game modeling is the fact that not everything needs to be built for a movie or pre-rendered model.  It is common practice for film to only build those elements in the scene that you can actually see on the screen.  In a game environment, it is necessary to make most things viewable from 360 degrees.  Can you imagine walking around your favorite game level and not seeing the back side of 3D car you just walked up to?  Or not being able to see the back of the character you just spoke to?  It wouldn’t keep you immersed in the game very long.  Well in a movie if the camera never travels to the rear of that set or never moves around the corner, it doesn’t need to be built.  This is certainly true for aspects of the gaming world, like the far off detail of the mountains, or implied buildings that you as a player can’t actually get to in the game.

A common practice among the two disciplines is that of creating LOD models, or Level Of Detail models.  In a game, when a character carrying a machine gun walks up to you from the far end of a long hallway, chances are it is not a consistent model the entire journey for the character or gun.  When it is far away, a lower resolution model, with lower resolution textures is used.   The reasoning for this is that the details cannot be discerned at that length so there is no need to use CPU time to render those higher resolution elements.  As the character approaches, there may be 2 or 3 changes that swap the model and textures out with higher and higher resolution assets, until it has walked right up to you in camera.  If done properly, these “swap outs” go unnoticed for the most part.  Here are three LOD models that range from the lowest, to the highest resolution and density:

LOD examples (all seen up close)

Now that you can see it up close and appreciate the details, the highest resolution asset is required here.  In a game, these types of LOD model are used constantly and in varied ways to allow the best frame rate at all times.  Seen up close, the lower resolution models look out of place.

Looking at them in the context of how far away they would be seen, however, makes their existence much more understandable:

LOD examples (seen at appropriate distances)

This is a primary scene management tool employed constantly in games.

Movie modeling might use aspects of LOD’s too.  There are close up models and models built for distance shots too.  The main difference for film models is that rarely do the various LOD’s have to seamlessly blend.  Much of this decision making process lies in the story or action that needs to be conveyed for that shot.  For the very next shot, it may require a completely different set of assets and details that didn’t apply to the first shot.  Typically there are three levels of modeling that occur for movie models: Block, Medium and Detailed.  Each stage identifies and solves different problems for the production.

 At the block stage, the overall proportions are identified with a simple low detail model.  This helps to define the silhouette of the model and have a low resolution asset useful for animatics or test renders.  Medium level models take the next step and begin by adding other details onto the Block model that help to define the finished look of the model.  Additions like antennae, guns, rear view mirrors or other details that are not defining the general shape of the model qualify.  This stage helps to identify moving parts and areas that may require special attention from a technical artist.  Finally there is the Detailed model, which contains all of the detailed parts and pieces on a higher resolution chassis.

An example utilizing these ideas is a space-ship model that flies past the screen as it speeds towards its destination.  Because we only see the one side of the ship, this is the only part that needs to be built.  This close fly by model needs to have a high amount of detail and geometry to look convincing.

Highly detailed geometry for highly visible areas

The other side of the model, does not need this level of detail:

 
Lower level of detail for low visibility areas

For the next shot in this sequence, though, the ship may have to turn towards the camera at a far distance and deploy a docking claw that attaches to a space station.  Here we would have a different model of the ship that has an animating robotic arm component, one which the close fly-by model didn’t have.  Likewise, extreme close up shots would have a special model made for those elements.

There are no concerns for efficiently, really, in the movie created asset.  As long as the model can render, it is considered to be acceptable.  For a pre-rendered sequence, render time can be extensive, but typically there are large render farms that can tackle the job.  There is also the safety factor for these models that any render anomaly can be fixed in Post, where the game model must work all the time at every frame it is rendered in.  Other stipulations sometime burden the game model such as the fact that at times the game asset must be “water tight”.  What this means is that all of the vertices on the model need to be welded or merged.  Render times for real-time shadows and advanced lighting can be complicated if a model is not sealed at the vertex level, and therefore they take longer to compute.

It is a common expression that there is a time and place for everything.  Nothing could be more true when discussing modeling for Movies or Games.  There are certainly similarities between the two mediums and many different approaches to solve the task at hand.  As game systems become more and more advanced, these two approaches may become more and more alike.  Perhaps one day there may be no distinction in the modeling process between the two.  Until then we need to rely on these various techniques and build our models accordingly.

*Special Thanks to Eben Cook for supplying the Winery Barrel