AbstractWe propose a new efficient algorithm to simulate magnetic structures numerically. It contains a generative model using a complex-valued neural network to generate k-space information. The output information is hermitized and transformed into real-space spin configurations through an inverse fast Fourier transform. The Adam version of stochastic gradient descent is used to minimize the magnetic energy, which is the cost of our algorithm. The algorithm provides the proper ground spin configurations with outstanding performance. In model cases, the algorithm was successfully applied to solve the spin configurations of magnetic chiral structures. The results also showed that a magnetic long-range order could be obtained regardless of the total simulation system size.