Nintendo isn't big on generative AI, and with Zelda: Tears of the Kingdom was careful with automated art – "The game could potentially end up losing something valuable"

Some elements of The Legend of Zelda: Tears of the Kingdom's art and design processes were automated for efficiency, but Nintendo was careful about offloading too much to computers out of fear of "losing something valuable" to the gameplay experience.

As Automaton reports, Tears of the Kingdom lead QA engineer Takuma Oiso discussed the game's development pipeline at Japan's recent Computer Entertainment Development Conference. Oiso dug into the implementation of the game-altering Ascend ability which lets Link leap through virtually any roof or ceiling, and the way this impacted Tears of the Kingdom's many cave systems. 

With hundreds of caves to make for Tears of the Kingdom's layered world, Oiso says Nintendo sought ways to optimize development by automating some elements. "However, if we were to pursue this blindly, the game could potentially end up losing something valuable," he adds. 

To that end, the team set clear boundaries on anything gameplay-facing, instead relegating purely "decorative" aspects to a procedural system that could spit out caves for artists to retouch. Level design, enemy and resource placement, and other similar variables which affect the way players explore and interact with caves were all done by hand. 

This method ended up working so well that it was later applied to Tears of the Kingdom's sky islands, Automaton reports. Oiso says artists were eager to find ways to pragmatically automate other processes since it freed up time that could then be invested in creation and polish elsewhere. My mind always goes to the kitchen with subjects like this; no sense whipping cream by hand when you could dump it in a stand mixer and do something else. Machines, baby. 

This Tears of the Kingdom anecdote functions as a practical example of Nintendo's view of procedural content in the context of AI, an increasingly divisive subject in the games industry. The methodology differs from full-fat generative AI, but the application is similar: let a computer