This article presents the progress in the development of a generative algorithm to support the synthesis of periodic metasurfaces. The proposed model corresponds to a conditional Deep Convolutional Generative Adversarial Network (cDCGAN) to solve the backward design problem for absorptive metasurfaces. To build the training dataset, three canonical shapes were simulated in the 50-80 GHz band. The algorithm was fed with absorption spectra and additional conditions to render an image with a proposed geometry. We demonstrate that the rendered designs produce absorption profiles similar to the ground truth from the validation set, in terms of shape and absorbance. Also, the conditioning of the DCGAN works with high accuracy level. These results are promising though there is room for improvement by extending our training dataset and by delivering an operational forward network and optimization pipeline.