Observational and experimental studies demonstrate that voters discriminate against candidates from ethnic minorities. This study draws on social psychological theories of automatic social categorization and recognition effects to argue that familiarity can help reduce voters' bias against minority candidates. The persistent underrepresentation of ethnic minorities in political office, I argue, prevents voters from automatically ascribing relevant traits to minority candidates. To test this hypothesis, I experimentally manipulate voters' subliminal familiarity with candidates by exposing them to a series of synthetic images created by General Adversarial Networks (GANs) that show faces of fictional political candidates. Half of these photorealistic, high-quality images contain subtle visual features of known, real-world minority politicians, evoking subconscious feelings of familiarity in voters. The other half, instead, contains features of unknown, hypothetical minority candidates. Drawing on a large set of such synthetic images that offer rich variance in other relevant candidate features, I propose two experimental studies. Study 1 examines the effects of subliminal familiarity on voting decisions, while Study 2 aims to understand to what extent subliminal familiarity affects the inferences voters make about candidate characteristics. In both studies, I examine whether some voters are more susceptible to subliminal recognition effects than others and whether their partisan identity and level of political sophistication can explain any such potential variance. Not only do the results of this study contribute to our understanding of subconscious processes that influence voters' decision-making, they also have important implications for policies aimed at improving the quality of descriptive representation in modern democracies.