multiple conformations?

As has been pointed out, AlphaFold will generally only give one conformation of a protein or complex. Is this simply because it tries to maximize contacts of coevolving residue pairs? In cases where we have prior structural knowledge, it might be helpful to have the option to suppress a predicted contact, for instance if we would like to visualize a ligand-activated complex and we know that a contact is present only in the unactivated/ligand-free state. I may see if simply replacing residues of this sort with U gives the desired results. Of course it may be easier to simply use templates in this case. Are you considering any better ways to increase the number of predicted conformations?

The issue discusses AlphaFold predictions, suppressing contacts, and templates, but names no files, tests, or entry points. First clarify the desired mechanism and acceptance criteria; no completion condition or concrete implementation path is specified.