This study introduces AF_unmasked, a modified version of AlphaFold designed to integrate experimental data and improve the prediction of large protein complexes. AF_unmasked leverages quaternary templates to bias the neural network, enhancing the prediction of complex structures. The method is effective even with imperfect templates and can handle large protein complexes, generating high-quality structures (DockQ score > 0.8) without retraining the neural network. Key features include the ability to use quaternary templates for improved integrative modeling, structural inpainting to fill gaps in experimental structures, and the estimation of mutation effects. The approach is demonstrated through various use cases, including challenging multimeric targets from CASP15 and cryo-EM datasets, showing significant improvements in prediction quality and efficiency. AF_unmasked is open-source and easy to use, making it a valuable tool for structural biologists.This study introduces AF_unmasked, a modified version of AlphaFold designed to integrate experimental data and improve the prediction of large protein complexes. AF_unmasked leverages quaternary templates to bias the neural network, enhancing the prediction of complex structures. The method is effective even with imperfect templates and can handle large protein complexes, generating high-quality structures (DockQ score > 0.8) without retraining the neural network. Key features include the ability to use quaternary templates for improved integrative modeling, structural inpainting to fill gaps in experimental structures, and the estimation of mutation effects. The approach is demonstrated through various use cases, including challenging multimeric targets from CASP15 and cryo-EM datasets, showing significant improvements in prediction quality and efficiency. AF_unmasked is open-source and easy to use, making it a valuable tool for structural biologists.