Urban block layout design presents a critical and challenging task for urban planners, as block layout directly influences the physical structure of the urban area, the interaction between people and their environment, and the microclimate in urban areas. Thus, developing appropriate design strategies plays a decisive role in this process. This research uses the generative adversarial networks (GAN) technique to explore block layout design strategies. The GAN technique can effectively generate real and diverse urban blocks based on learning existing morphological properties. Therefore, it can be a powerful approach to urban block layout research. Genetic algorithms not only help to identify the ideal design solution but also enable the summary of key design strategies by numerous evolutionary periods. The study presents a CFD-based optimization framework that utilizes a genetic algorithm and 3D block models generated by the GAN technique to improve the urban wind conditions at the block scale. By comparing solutions, the study successfully optimized three objectives by 68.36%, 51.74%, and 41.83%. Ridge regression models examined the relationship between objective functions and urban morphological indices. The maximum R2 value of the ridge regression models reached 0.801, indicating that the models can effectively predict wind conditions by morphological indicators. Design strategies for wind-friendly blocks were developed by regression models and validated by case studies. The design strategies suggest that architects should prioritize the building layout, the building height at boundary areas, and the building shape, as these factors significantly affect the urban outdoor wind conditions.