The article introduces AF_unmasked, a version of AlphaFold that integrates experimental information to predict large protein complexes more accurately. AF_unmasked improves upon AlphaFold by leveraging quaternary templates, which provide structural information about protein interactions, without requiring retraining. This approach allows for faster and more accurate predictions, even when little evolutionary information is available or when experimental structures are imperfect. AF_unmasked is particularly effective in cases where standard AlphaFold fails to build the correct complex, and it can also refine predictions based on experimental data. The method uses structural inpainting to fill in missing areas in the structure, and it demonstrates that templates can be used to inject experimental information into the prediction pipeline. AF_unmasked was tested on various cases, including challenging multimeric targets from CASP15 and cryo-EM datasets. The results show that AF_unmasked can generate high-quality structures with DockQ scores above 0.8, even when using imperfect templates. The method also allows for the prediction of large protein complexes, such as the 27-mer CASP15 target H1111, which was previously difficult to model. AF_unmasked was also tested on cases involving mutations and conformational changes, where it successfully predicted the correct structures. The study highlights the potential of AF_unmasked to enhance the integration of experimental data with predictions in structural biology.The article introduces AF_unmasked, a version of AlphaFold that integrates experimental information to predict large protein complexes more accurately. AF_unmasked improves upon AlphaFold by leveraging quaternary templates, which provide structural information about protein interactions, without requiring retraining. This approach allows for faster and more accurate predictions, even when little evolutionary information is available or when experimental structures are imperfect. AF_unmasked is particularly effective in cases where standard AlphaFold fails to build the correct complex, and it can also refine predictions based on experimental data. The method uses structural inpainting to fill in missing areas in the structure, and it demonstrates that templates can be used to inject experimental information into the prediction pipeline. AF_unmasked was tested on various cases, including challenging multimeric targets from CASP15 and cryo-EM datasets. The results show that AF_unmasked can generate high-quality structures with DockQ scores above 0.8, even when using imperfect templates. The method also allows for the prediction of large protein complexes, such as the 27-mer CASP15 target H1111, which was previously difficult to model. AF_unmasked was also tested on cases involving mutations and conformational changes, where it successfully predicted the correct structures. The study highlights the potential of AF_unmasked to enhance the integration of experimental data with predictions in structural biology.