In previous works [1,2], it has been shown that phase-spaces can be modeled with Generative Adversarial Network (GAN [3]) such that the generator approximates the initial distribution of particles, thereby being able to be used as a source for Monte Carlo simulation (MCS). This approach was applied for modeling photon beams from Linac head in radiation therapy treatment [1], and for gammas exiting phantoms or patient CT during simulation of SPECT systems [2]. Compared to conventional phase-space files, GAN’s generator was shown to be compact, few MB instead of few GB. Moreover, splitting MCS of SPECT imaging system in two parts, 1) tracking emitted particles within the phantom/patient CT, 2) tracking particles in the detector, and modeling the first part with the GAN generator was shown to lead to computational time gain. However, such a GASP (GAN-generated Source of Particles) approach cannot readily be used for the simulations of PET imaging systems because it requires particle time information and the knowledge of the initial emitting event to distinguish between Trues, Scatter and Random coincidences. In this work, we extend the GASP method to consider pairs of particles, timestamps, particle weights. We also propose a new parameterization that improves the training. [1] D. Sarrut, N. Krah, and J. M. Létang. Generative adversarial networks (GAN) for compact beam source modelling in Monte Carlo simulations.Physics in Medicine & Biology, 64(21):215004,76, 2019. [2] D. Sarrut, A. Etxebeste, N. Krah, and J.M. Létang. Modeling complex particles phase space with GAN for Monte Carlo SPECT simulations: A proof of concept.Physics in Medicine & Biology, 2021. [3] I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville and Y. Bengio. Generative adversarial nets. In Advances in Neural Information Processing Systems, vol 2, pages 2672–2680, 2014.