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.