We visualize the best performing related methods. FreeNeRF struggles to reconstruct a valid head geometry and generates floaters. Preface is able to reconstruct the overall geometry but contains artifacts, in particular for strong expressions.

We compare with the best performing related work (Preface) on in-the-wild captures. While Preface performs well for neutral faces, strong facial expressions lead to an incorrect geometry. Note the blob-like artifacts in the mouth and below the chin.