The global energy crisis challenges us to develop more efficient strategies for the sustainable production of energy. Given the excellent efficiency of the natural photosynthetic apparatus, biohybrid photovoltaic devices present an attractive solution for solar energy conversion. However, their composition, stability, and complexity can limit their inclusion into photovoltaic devices. Here, we combined computational design and directed evolution to overcome these limitations and create tailor-made photoenzymes. Photo-biocatalysts were designed by introducing photosensitizer binding sites into heme-containing helical bundle proteins. The designed binding sites were specific for the target photosensitizer and readily transplanted into other helical bundles. The best design was highly evolvable and reached nanomolar ligand affinity after mutagenesis and screening. The evolved enzyme generated 2.6 times higher photocurrents than the photosensitizer alone, primarily driven by increased photostability. Evolvability is a unique advantage of our protein-based approach over abiological photovoltaic and will be critical to developing efficient biohybrid systems.