First test which looks kind of nice. The seeds are instances. Each seed particle (positioned at the tip of the seed) also has a counterpart at the end of each seed. These "end" particles get turbulized in the simulation stack. The vector between the two point positions (tip and end) is the new direction of the seed.
On another empty polygon mesh I copy the topology for the area where the fibres are attached to. There is also a copy of this mesh which gets deformed like the seeds. While the topology gets copied (one copy for each seed particle) it receives a copy index. This index is then used via an "ID to location" compound, to receive the current position and orientation of the seed particle which is resonsible for these polygons.
There is a static and animated version of the copied topology in order to use the "reinterpret location on new mesh" compound. Otherwise it's difficult to keep the fibres attached to the seed.
Next step is to animate and deform the stems via strands, attach the "flower" heads with all the seeds and fur strands to the stem and then also have seeds flying away from their initial positions.