It's a stack of separate decisions. Turn a dial and the prompt rewrites itself. Copy it, upload your photo, paste.
This is style transfer, not generation. The model redraws an image you already have — same pose, same person, new rendering. Works in Nano Banana (Gemini image), Higgsfield, and most img2img tools.
Pick one photo and run it five times, changing a single dial each round. Change two at once and you won't know which one did it.
When a combination lands, paste that style block exactly the same way every time. Reword one sentence between images and a set stops looking like a set.