Text-based Editing of Talking-head Video

Supplemental materials

[back to index]

Blending synthesized results into real video

Our algorithm produces results which align well into the original (larger) video, both spacially and across time. On careful inspection the synthesized results have less details (e.g. beard stubs) compared to the input video. However, we found that casual observers do not detect these and preceive a seamless blend.