I know perfectly well that Guids are supposed to be globally unique. This is a special case.
We have tools that create configuration files where Guids are part of the data. Creating new Guids every time you run the tool is very easy. However, in a couple of test scenarios, we are faced with the problem that independent tools should create objects with the same Guid where some underlying data match. To maintain and access a centralized repository of Guids in use is not practical.
By "seeding", I mean that having two different seeds is guaranteed to produce different Guids and that the same seed will always produce the same Guid.