The facial features of the people are too similar, with almost no variation, making the probability of repetition too high.
The facial features of the people are too similar, with almost no variation, making the probability of repetition too high. A random seed is used。
A highly realistic portrait of a aged 18-25 American girl, her smile sweet and radiant, her gaze sharp and intelligent. Her features reflect the diverse complexion and varied facial characteristics typical of America's multicultural melting pot heritage. The lighting should evoke a uniquely American feel with bright, optimistic natural light and wide-open landscape freedom. The clothing is simple and modern, with a blurred background in red, white, and blue tones reminiscent of the American flag and iconic Statue of Liberty skyline. The portrait must perfectly capture the characteristics of a 20-year-old American woman. A solo shot, centered composition. The text "United States" is displayed in the lower right corner.
Yeah... thats the problem with distilled models. Maybe there is a LORA out there that can add facial variations? Lets see what the Base model can do when its released though.
have you heard of dynamic prompts or wildcards? add some more to your prompt. If you give a generic description, you get a generic result, sorry has been like that for a long time, with all models beyond just clip. The prompt is guiding the image, I have something like this: a {__age/older__|__age/young__} __species/human__ woman with a __body/height__ __body/shape__ . she has __colors/skin__ skin color {with __body/skin-feature__| | }. she has a __face/shape__ with __colors/eye__ eyes in a __face/eyes__, __face/eyebrows__, __face/lips__ with __face/lipstick__ __colors/lips__ lipstick and a __face/nose__. Her hair is __colors/hair/*__ in __hairstyle/female__
all those __words__ are wildcards (text files with lists of options) and you will not get the same face ever again unless you specifically prompt for it...