A problem that I run into with every single tool that does this is that any database that I would _like_ to have an ERD of, is inevitably too insane to be cleanly represented by an auto-generated ERD.
Best of all, you can filter the views down to specific classes, keywords, and web service routes, which solves the scaling problem ("too many nodes!") that one generally finds with UML or ERD on large projects.
This approach also provides metadata that can be used to figure out which tables are being used where in the code, and vice versa.
Context is king!