Magnetic Resonance Fingerprinting (MRF) acquisitions with balanced Steady State Free Precession (bSSFP) and spiral trajectories are prone to off-resonance artifacts. Thus, those artifacts hinder the reconstruction of the tissue maps (T1 and T2). In this work, we propose a model based on Generative Adversarial Networks (GANs) mixed with transformer blocks to decrease these artifacts. Our method improved the NMSE for both quantitative maps T1 and T2. Heavily reducing the effects of the off-resonance in comparison to classical bSSFP-MRF.