Optimising art assets for gaming

Optimising art assets in gaming

    When creating a 3D asset for a game, many things have to be considered one of the most prominent features is how well optimised it is. Why is this important? Well even one scene or small section of a level can be comprised of multiple 3D meshes and in turn this is a finite resource within the engine that is shared with other systems such as UI, blueprints, etc... When talking on a larger scale a whole level could be made up of hundreds if not thousands of assets and so as a 3D artist we need to ensure that our models are optimised in a way to reduce the impact on the engine while still maintaining visual fidelity. 

How are models optimised?

    There are a variety of ways artists optimise objects they are preparing in games, from baking high poly mesh details onto a low poly base mesh to ensure the polygon count isn't extreme, to making texture sizes smaller in resolution, instead of having every asset being a full 4K high resolution some are only 512 pixels large. 

    These things are always on the artists mind when creating a new asset, what do they have to sacrifice in order to ensure the model isn't so resource intensive that it could potentially have a significant impact on the performance of the game when someone is playing it.

    Another step in the pipeline is creating LOD's or Level of Detail assets. The primary feature of an LOD is take an asset that is comprised of thousands even millions of polygons and triangles and then condense that number down depending on the distance the object is from the player's view while maintaining proper silhouette and texture definition for the distance.

    Sometimes models have to be highly detailed such as character models or important landmarks in a game or vehicles, this is something that is unavoidable at times but the intelligence of LOD's is that if the object is a certain distance away from a player then it doesn't need as much detail, in turn an LOD is there to replace the mesh with a lower poly and lower resolution version which will have less impact on performance, this is done multiple times as the player gets further and further away until it is completely out of view and the model is then removed entirely.

How are LOD's made?

    Again there are various methods of making an LOD, but the basics are create copies of the model and retopologise them down into lower poly versions of the previous instance, then within the game engine set these up in a draw chain that takes into account the player's draw distance (view distance) and set the correct LOD as required. This is a time consuming process however and if you consider it on a massive scale to be done on all models that require LOD's then this can be a daunting task in itself. 

    However there are software and systems that can automate this process for you, aiding the workflow for artists and removing the need for spending unnecessarily long hours on producing LOD's for models. One such option was developed by Simplygon who produced an automated system for optimising models and LOD's to get them game ready and drastically reduce the time spent on art assets.

    I came across a GDC talk which summarises how the Simplygon program works, this is from 2014 however so I'm sure they have made much progress and more improvements over the many years, but still as base concepts go it is still an informative and interesting talk to watch; 

https://www.gdcvault.com/play/1020666/Optimize-your-3D-Assets-with 

    The main speaker is Koshi Hamedi who works for Simplygon, with guest speaker Tramell Isaac from Sony Online Entertainment who worked on Planetside 2 and used Simplygon for the entire process with his team.