We present Faceform NanoSkin, our approach for increasing the resolution of actor-specific skin microdetail by a factor of 4 to 16. NanoSkin is a neural network trained on our proprietary skin dataset to generate displacement textures up to 16K.
Skin microdetail is one of the key factors that contributes to the realism of digital characters. However, one of the most common production approaches is still to use pre-made skin detail maps captured from other people. We will give a high-level overview of the approach we use to enhance the details of a specific actor, rather than inventing them from scratch or transferring them from another person.
We will also discuss our work on lookdev automation and automatic material map acquisition, where texture maps are estimated in a way that allows the rendered character to match the actor’s reference photographs as closely as possible.
