Nano Banana 2 supports JSON prompt specs for pose locks, mirror scenes, and brand boards

The clearest shift is from descriptive prompting to specification prompting. In the pose-edit example, the author breaks the request into hard fields for framing, body axis, upper-limb perspective, lower-limb perspective, depth hierarchy, and “hard locks,” including numeric ranges for camera angle, hand size, elbow bends, and subject scale. That turns a vague “dynamic peace-sign pose” into a constrained edit where identity and clothing stay fixed while only geometry changes pose lock template.

The mirror prompt uses the same logic in a lighter form: placeholders for subject, mirror type, impossible reflected scene, replacement room, color cast, and lens. The examples show the structure traveling well across genres, from a boxer reflected against a moonlit whale breach to a nurse facing a candlelit library with floating books

.

Brand work benefits because art direction can be split into clean buckets: brand inputs, visual style, composition rules, color language, typography, and “important” exclusions. The Glossier example asks for a layered campaign-world collage with packaging, typography, UI fragments, stickers, and merch, while explicitly banning grid layouts and minimalism brand board prompt.

UI-heavy image concepts get similar gains. The music mockup prompt defines a single variable for artist name, then specifies orbiting Spotify and Apple Music cards, glassmorphism materials, foreground occlusion, depth of field, and even automatic replacement of placeholder tracks with the artist’s top songs; a second post shows the template producing a Post Malone variant with consistent composition Post Malone example and links to a reusable prompt page prompt page.

The strongest evidence is in transformations where models usually drift. One creator’s MS Paint recipe preserves pose and facial likeness while forcing a Windows 95 look through fixed constraints on palette, aliasing, dithering, tools, and canvas logic

. Another workflow starts with a Niji 6 illustration and uses Nano Banana 2 for a controlled 3D conversion that keeps the character design intact 3D conversion. A mobile example shows the same principle in-app: draw over an image in Krea, describe the edit, then run Nano Banana on the masked area Krea mobile edit mobile edit demo.