What did you learn?
Over this month I finally implemented a paper for dendritic noise patterns, as well as starting a gpu implementation. The gpu part was surprisingly easy, I needed to implement it for a work project at the same time. The original Dendry noise leaves a lot of options unexplored for ways to alter the noise in order to create more or less eroded terrains, but the parameter space was too large to properly explore in this time.

How did you improve the terrain?
Visually I'm not happy with the rivers that result from this Dendry implementation. They're highly dependent on the size of the world being created, and tuning that relationship was more time consuming than I planned for. I'll definitely be revisiting it in future as there is a lot more to try. In terms of performance, the cupy version of the code ran extremely fast, bringing my render time for a 4k heightmap down to less than 10 seconds for my previously existing code.