From Sketches to Renderings: A Comparison of Rapid Visualization Methods for Sketches Based on ControlNet (CN)

Abstract—The use of the ControlNet (CN)-controlled Stable Diffusion (SD) model increases the potential for artificial intelligence to enhance efficiency in architectural design. This study compared the results of using different CN processors (Canny, MLSD, Lineart, Scribble, Softedge) to process various styles of design sketches and generate architectural renderings. The aim was to explore the potential of each CN processor to visualize architectural design sketches. The results indicate that with appropriate CN control, architectural hand-drawn sketches can be quickly visualized, yielding satisfactory outcomes. The comparisons revealed that: (1) CN performs better with orderly hand-drawn sketches; (2) CN weights between 0.5 and 1.0 are more effective for image control; and (3) minimizing conceptual lines in hand-drawn sketches reduces the likelihood of rendering failures. This study provides a foundation for future research in architectural and planning image generation and offers a more efficient and cost-effective design method for architectural professionals.