The hard part of optimization is not finding a decimation control. It is deciding how much geometry the model actually needs. I wanted to make that judgment visible, so I used the same sculpt to compare progressively smaller exports and then tested a second strategy that starts low and adds detail only where it helps.
Why I made this
The test character began at roughly 194,000 triangles. That may be comfortable while sculpting, but it is an expensive starting point for a real-time engine—especially when several objects or animated characters share the scene.
A percentage alone does not answer the design question. The useful threshold depends on silhouette, painted detail, animation needs, and where the model will be seen. Developers and technical artists need a repeatable comparison process, not a magic number.
What I did
I begin by centering and orienting the sculpt around Medium's origin so it will arrive in Unity predictably. I create and paint a quick character, then use the export screen to establish the full-resolution count.
The first test exports at ten percent—about 19,000 triangles—and comes back into Medium for a side-by-side comparison. I repeat the process at one percent, roughly 2,000 triangles, and show that the silhouette still holds up surprisingly well even though the face becomes visibly blockier.
Then I test the opposite workflow: begin with a low-resolution layer, block out the major form, and increase resolution only when finer detail requires it. Separate higher-resolution layers can preserve local detail without doubling the entire sculpt, although merging those layers can increase the final count and introduce new tradeoffs.
The last comparison moves from an overly aggressive 1,600-triangle export to an 8,000-triangle version that preserves the important surface detail. The lesson is the loop itself: export, reimport, inspect, adjust, and repeat.
Chapters
Select a chapter to jump the video directly to that part of the workflow.
A note on the archive
This video uses Oculus Medium as it existed in 2019. The product later became Medium by Adobe, and Adobe still maintains documentation for the sculpting and layer workflow. Interfaces, export behavior, hardware support, and surrounding Adobe tools may differ now.
The transferable lesson is the measured comparison process, not the exact buttons or triangle targets. Current projects should validate their own platform budgets, shaders, rigging, animation, and level-of-detail requirements.