AIGC technology is playing an increasingly crucial role in generative design. At present, many AIGC tools are used to generate images in the architectural design, such as Midjourney, Stable Diffusion. Although the single image generated by AIGC tools might have high quality, it is difficult to maintain the consistency of the ontology when representing the same building from different angles due to the inevitable randomness of the generation process, which means different pictures contradict each other in details. This article discusses the possible solutions of ontology inconsistency problem with the guidance of different AIGC tools, and points out that generating 3D model is the most effective way to solve the problem, but the current generation tools still have various defects. With the contrast of different approaches, the technical path of Sketch-Multiple Pictures-Contradiction Elimination-3D Model is considered to be more appropriate for architectural design process. This path also puts forward technical requirements for the future specialization of current AIGC tools.