AI, machine learning and procedural generation in 3D art is the future, according to Andrew Price from Blender Conference 2018. During his excellent presentation, he explained how the rising cost of development could lead to a standard AAA game costing $200M+ to make in the coming years.
However, adapting AI, machine learning, and procedural generation can help bring down development costs. We are already seeing where rising development costs are taking us. Data shows (1995-2018) that every 10-years the cost of video games development goes up 10-times. At this rate, an average AAA game can cost $200M to make.
Each company tries to outdo the other which is partly the reason why we are seeing a rise in development budgets. A big chunk of this budget goes to the creation of assets, here is where most of the money is spent. Assets are unreasonably expensive to make. From modeling to texturing to final renders, the cost of significantly high. Assets are revisited to adjust various aspects depending on narrative and their placement in the game. At the cost of $60/hr average wage, creating a single four-story building in, let’s say, a game like The Division costs close to $4000. Thanks to Price and his estimates, we can see how one scene ended up costing Ubisoft $200,000.
There must be a smarter way to approach video games development to bring down costs. And Price believes procedural generation and AI is the answer. Games like No Man’s Sky have given procedural generation a bad name but if handled correctly, this method can actually help artists create cheaper assets and improve the workflow in organizations.
Publishers have a static workflow which means a 1-to-1 input to output ratio. If developers wish to create more than one building they usually start from scratch. This means each building in The Division is created from the ground up. Of course, one can use an existing model as a base for a separate asset but you need to make numerous changes to it, retexture it, so you are still getting a similar output.
The answer to this problem is procedural workflows. If developers are creating a building they can simply set parameters for it and let the AI produce various models. One can argue that procedural generation may remove the creative aspect but the results are quite contrary to this popular belief.
As Price explained, it can actually help artists understand what makes something look good. Often, a developer may simply be doing things without thinking about it. You can see the entire presentation which is enlightening.
