ControlNet settings:

Preprocessor: none Model: MLSD.

Control mode: Balanced or My prompt is more important.

Resolution for txt2img: 512x768

You can use it with OpenPose (as a second ControlNet unit. In this case use 0.8 weight and "My prompt is more important" mode for MLSD) or generate background and character separately then use img2img to merge them together (this method works much better than 2 ControlNet units)