Hello,
The key thing to know here is that hybridize works best when it has templates that cover all domains. When it starts with a template that only covers half the structure it starts out with the other half as basically an extended chain and then works to fold the protein from there
To get around this, try making a pdb file with both templates in a single pdb file (and with the two chains oriented approximately as you would expect them to be)
This is not a very scientifically sound way of doing things, but i have done it before to get starting points that I hope to work with more experimental data later.
As an alternative I would suggest trying the robetta server — it is quite good https://robetta.bakerlab.org/
.
