Watermarking is a common way to claim ownership of a content, determine its origin and track its distribution. However, traditional watermarking techniques have their applicability limitations: they are visible, obstructive and negatively impact UX. In this talk we'll discuss how watermarking can be adapted to scenarios we deal with here at Reface. We will discuss how to infuse digital watermarks, or all sorts of additional meta-data in general, directly into video content using deep learning in a robust and non-obstructive way. We'll go through a complete process of designing a watermarking pipeline, including real-life use cases, problems, algorithms and their limitations