Heidelberg University has launched ConTorlNetXS, an optimized version of the ControlNet model, significantly enhancing performance. ConTorlNetXS has only 1% of the original model's parameters, resulting in doubled inference and training times, yet it produces higher quality images with more precise control. The new architecture is named ControlNet-XS and excels in FID scores. By optimizing the ControlNet architecture, it addresses the issues of large models and reduced image quality.