Damaged Facial Features While Generating Image

[mkafil]

Generated images

The above image are generated image

Input Images

Problem :
the Generated images get damaged eyes and not realistic

Model : SD1.5 Number of images used to train 10

Is there any thing should i change to get some better result while training lora model

Which method are the best ?
Lora or dreambooth or finetuning

And some image don't have much similarities . I know input images plays import role. is there any other things in should concentrate
like text encode and unet ... or something

this are the params i use to train

Thank you in advance ......