A symmetry-guided AI model narrowed 110,000 generated crystal structures to four stable antiferromagnetic candidates, pointing to a faster computational route for discovering materials for future spintronic memory technologies. The study, published in the journal npj Computational Materials, introduced a generative deep learning framework called “Space Group Crystal Diffusion Variational Autoencoders (SG-CDVAE)”, which incorporates crystallographic space-group information into its latent representation. From an initial pool of approximately 110,000 generated crystal structures, the SG-CDVAE generation and high-throughput screening workflow identified 80 high-symmetry antiferromagnetic (AFM) candidates. Antiferromagnets present a promising alternative to ferromagnetic materials, as their adjacent atomic spins align in opposite directions, producing no net macroscopic magnetization and making them resistant to magnetic interference, with terahertz-scale spin dynamics suitable for ultrafast information processing.