Interactive Stroke-based Neural SDF Sculpting

A new neural SDF sculpting framework brings stroke-based editing to implicit 3D geometry, aiming to close one of the biggest usability gaps between neural representations and traditional mesh tools. Instead of relying on isolated spot edits or slow re-optimization, the system lets users drag continuous strokes across a surface and see the shape update interactively.

The core idea is to sample a tubular neighborhood around each stroke and apply a configurable brush profile along the curve. That gives artists and technical artists a more familiar sculpting workflow while preserving the smoothness that makes signed distance fields attractive in the first place. The practical win here is not just convenience; it is keeping edits local, controllable, and fast enough to stay in the loop.

The work was submitted on 5 Feb 2025, revised on 19 Aug 2026, and later connected to High-Performance Graphics 2025. For teams experimenting with neural assets, this is a meaningful step toward making implicit representations editable in production-like workflows rather than just renderable or reconstructable ones.

The broader implication is that neural geometry tools are starting to look less like research demos and more like actual content-creation interfaces. If stroke-based sculpting holds up outside the paper, it could reduce the need to bounce between neural and mesh representations during lookdev, prototyping, or procedural asset iteration.