Unfortunately, you may have to prompt the faces yourself if you're using a free web-based image generator. When I make generate images using a fork of Stable Diffusion on my own system, I also get the same face syndrome unless I prompt specific features or use wildcards to randomly assign different features. You don't have to be that detailed, but prompting the shape of a character's face, nose, eyes, cheeks and lips will help. There are plenty of charts on google images to tell you the difference between facial features for your prompts. But again, you'll probably have to enter the features in manually, especially if you're using a free service.