Flux knows how to create such images, but this lora greatly reduces the chance that a face will be visible

Attempting to train flux lora locally