The long-awaited ControlNet for SDXL is now available for download, thanks to Xinsir who has released three specially trained ControlNet for SDXL models. The three ControlNet’s released are named Canny, Openpose and Scribble. It’s recommended to rename the downloaded files, as all are identical in size (2.5GB) and bear the same name, which could cause confusion when identified.
Initial testing was conducted on the Canny model using ComfyUI by altering the standard workflow. The results obtained with Canny are impressive, and when coupled with a high-quality model like SDXL, superior outcomes are achievable. The testing process involved using a primary workflow with no upscaling at the initial phase. It’s only the inception of the investigation with SDXL Controlnet and it’s believed that further exciting possibilities will be explored with this technology.
The Canny model’s training was carried out on over 10 million images as mentioned in their readme, a process that is both commendable and intensive in terms of GPU power and time requirement. This reveals the detailed and extensive process required to confirm the functionality and efficiency of the model.
Many individuals have eagerly anticipated the release of these ControlNets for SDXL, constantly asking when they would become available. Now they’ve been unveiled, and although currently these are not accessible in Automatic1111, it’s expected they will be added in the not-too-distant future.
While three ControlNet for SDXL models are now available, hopes are there will be more shortly. This encouraging progression will awaken creativity and possibilities for using the SDXL Controlnet in diverse ways. The clear goal is to optimize the high-quality output that the SDXL model is capable of producing, leading to an expansion of the applications that users can explore.
This release marks the beginning of a series of experiments and research into the application of the SDXL Controlnet. As progress continues, users look forward to continuous improvement on the models, more detailed outputs, and the availability of more ControlNet for the SXDL. The excitement is clear from users who are keen to leverage these ControlNet models for improved quality results and application of the SDXL model for their unique tasks.
