In the current era of information explosion, how to realize the efficient connection between advanced technology and art, and how to decentralize visual design, has been widely concerned by many scholars. Research a method of human-machine collaboration using generation technology as an extension of the design process, combining convolutional neural networks with generating adversarial network models, and introducing conditional methods to connect conditional information at each layer of the generator structure. Optimize the above model by using spectral normalization and group normalization in conjunction, Finally, a conditional depth convolution to generate an adversarial network model (CDCGAN) is constructed. The research results show that the average accuracy of the CDCGAN model is 97.28%, and its extensibility and learning ability in the intelligent visual recognition design platform are excellent. In conclusion, CDCGAN model has good performance and accuracy, and can be well applied to intelligent visual recognition design platform.