A new research paper argues 3D assets should be generated as code, not raw meshes.
The paper, titled "Code-native generation of highly programmable 3D assets" and posted to arXiv, describes an approach that produces 3D objects as structured, parametric code rather than fixed polygon meshes or point clouds. That framing matters because a mesh is a dead end - once generated, it is hard to edit without starting over. Code-native output, by contrast, can in principle be tweaked, parameterized, and reused the way a programmer edits a function rather than a finished drawing. The work surfaced on Hacker News this week, drawing a handful of votes and a single comment - a quiet reception typical of early research shares rather than product launches.
Most generative 3D tools today, from text-to-3D pipelines to game-asset generators, still output meshes that are effectively frozen once rendered. If a code-native representation holds up in practice, it would let developers programmatically adjust geometry, swap materials, or spin up variations without a full re-generation pass - closer to how procedural content generation already works inside game engines. That is a meaningful shift for anyone building 3D pipelines where assets need to stay editable downstream, not just look good in a demo reel.
It is one paper with modest traction so far, and the real test is whether "programmable" survives contact with an actual production pipeline instead of staying a research demo.