If there is a template available with fairly high sequence similarity you should probably just use the one template. Otherwise it may be advisable to use a few different ones and compare the results. I've typically tried a few different templates and qualitatively compared the final output models just to make sure there were no major differences, especially around known catalytic or active sites, binding sites, or other important domains. Most of the time you probably aren;t going to see huge differences, and it also really depends what you are using the homology model for in terms of how accurate you think it needs to be.

Use I-Tasser server.