I suggest you generate image and use inpaint to make different face expressions for each character.

Resolution for txt2img: 512x768/768x512

Preprocessor: none Model: segmentation.

Control mode: Test which one works better. Usually "My prompt is more important" works best.