Machine-learning algorithms for protein structure prediction can now generate models directly from sequences. However, photosynthetic assemblies represent a challenge due to additional levels of complexity arising from their multi-protein nature and presence of cofactors.