The best results are from even aspect ratios such as 768x768. I made this model to assist in recreating extreme close ups of faces. Even with decent prompts and reference images the AI severely struggles with facial dimensions. Without a LORA I found it to be extremely difficult to get a result even close to what I wanted.
If you have a control net image of lips or a mouth the LORA can help you generate more accurately. If you have no control image you can also get a good result by toying with the prompt and cfg scale. Good luck and I hope you can get a good result!
