Apply a different dense layer for each timestep in Keras

Question

I want to have different dense layer for each of the time steps of a sequential 3d input: (N, timesteps, features).

In my mind came TimeDistributed, but then recogni