G.A.N Possibilities created from the recursive networks intensify the human relationship. RNN have an identifiable source. The result is a fuzzy image. RNN represent a recursive time. The given time is conditioned. Its previous state. It allows the model to store complex signals over long periods of time. Recursive neural networks use both the past and the present to create possible futures. We note that it is possible to know from which image the machine started to create its possible. These are almost layers of time that accumulate, translating by a phase of multiplication of the image and of the memory. This liquidity resembles an observation under psychotropic drug. The possible is a complex and problematic notion. It is a construction which presupposes, from the existence, that we imagine the possibility of it while projecting it backwards. The possible is more than its realization. We assist finally to a process of imagination which lets foresee still new relational possibilities: machines eating images? Human realities?