Since you will be training the ControlNet with the SD3 medium model that was released in a broken state and generates images with human deformities, would that not affect the overall quality of the SD3 ControlNet model? Why not wait for the improved version of SD3 to be released in a few weeks and train on that?
Since you will be training the ControlNet with the SD3 medium model that was released in a broken state and generates images with human deformities, would that not affect the overall quality of the SD3 ControlNet model? Why not wait for the improved version of SD3 to be released in a few weeks and train on that?