Spectral Subsurface Scattering from RGB via Biophysical Skin Inversion

This paper tackles a long-standing pain point in skin rendering: getting plausible subsurface scattering without a lot of manual parameter tweaking. Instead of homogenizing skin into a single medium and hand-authoring scattering values, the authors infer spectral optical properties directly from one RGB diffuse albedo texture.

The core idea is a mixture-of-media representation that approximates skin as three uncorrelated media whose aggregated scattering is learned from the input albedo. A chained neural decoder predicts the mixture parameters, including anisotropy, scattering radius, and scattering albedo. The paper says this is meant to preserve the layered complexity of skin better than the usual albedo inversion workflow.

For developers, the practical appeal is workflow reduction: less art-skilled tuning, fewer guessed parameters, and a more physically grounded result for path tracing-based rendering. The authors also say the method can be used in a random-walk path tracer with minimal changes, basically by randomly selecting which medium to traverse during scattering.

This is still an arXiv submission, not a shipped tool, but it points toward a useful direction for character rendering pipelines. If it holds up in practice, it could make skin authoring faster for technical artists and more consistent across assets, especially in spectral or offline-quality rendering setups.