Recent Project Insights of AI-assisted workflow

A recent AI-assisted workflow I used in a major project competition in northern China helped meet tight deadlines and produce high-quality visuals quickly.

I was pleased to be involved in a major competition for a large-scale project in a northern Chinese city. The timeline was incredibly tight, and we were up against top-tier international architectural firms. To meet the challenge of producing high-quality and competitive designs in a short period, I explored using AI-assisted rendering with Stable Diffusion for the final visualizations, without involving an external rendering consultant.

(image)

Currently, AI-assisted rendering is primarily used in the early design stages for internal discussions, as producing precise architectural images requires significant time for fine-tuning. This workflow involved combining traditional rendering with outputs from Stable Diffusion, aiming for a synergy where 1+1 equals more than 2 in terms of quality.

Thanks to Semantic Segmentation, the accuracy of generating architectural details improved significantly. By producing layered images representing different architectural and natural elements, the generated images reached a satisfying level of precision. Finally, I overlaid my most satisfactory AI-generated images with traditionally rendered ones from the same angles, and further processed them to create the final visualizations.

(image)

(image)

(image)

This attempt greatly shortened the time required to produce high-quality renderings. More attempts to further refine this workflow will be introduced in this column in the future.