One reference, and your character stays consistent

Cut-and-paste photo editors lift a face from one photo and blend it onto another — which is why the light never quite matches and the edges smear on the first sharp turn. BananaBanana does it differently: the character travels as a reference, not as pixels. You attach one or more portraits of the person, describe the scene you want in plain language, and the model — Nano Banana 2 or Nano Banana Pro — generates the entire image from scratch with that character woven in. Lighting, skin tone, perspective and style all belong to the new scene, so the result reads as a photograph of the person, not a collage. Up to 14 references per generation let you lock several characters and even products at once, and multi-turn editing lets you change the outfit or background afterwards while the character stays recognizable.

The biggest use case is AI influencer and brand-persona content: one character sheet, then an endless feed of consistent posts — beach today, café tomorrow, product-in-hand on Friday. E-commerce teams put the same model into every product shot without booking a studio twice. Creators render themselves into art styles — oil painting, cyberpunk, film noir — for avatars and covers. Agencies keep a client's spokesperson consistent across a whole campaign, mixing face references with product references in a single prompt. And because a still is only the first frame, you can hand the render to image-to-video and get the same person moving in a Veo 3.1 clip. Each image costs from $0.03, so a hundred-post content calendar is a few dollars, not a retainer.