How to handle gender with machine learning is a controversial topic. A growing critical body of research brought attention to the numerous issues transgender communities face with the adoption of current automatic gender recognition (AGR) systems. In contrast, we explore how such technologies could potentially be appropriated to support transgender practices and needs, especially in non-Western contexts like Japan. We designed a virtual makeup probe to assist transgender individuals with passing, that is to be perceived as the gender they identify as. To understand how such an application might support expressing transgender individuals gender identity or not, we interviewed 15 individuals in Tokyo and found that in the right context and under strict conditions, AGR based systems could assist transgender passing.