Nano Banana 2 adds structured prompt templates for JSON portraits, night-flash shoots, and icon sets

• Nano Banana creators are moving from one-off prompts to reusable spec sheets, with PromptsRef’s 400-plus library organizing hundreds of templates for portraits, ads, and scene-driven image work via the prompt library.
• Two widely shared Nano Banana 2 examples show the format: long JSON-style portrait prompts that lock in camera angle, lighting, styling, constraints, and negative prompts for consistent selfie outputs portrait JSON mirror selfie JSON.
• Other creators are adapting the same idea to style recipes, including Leonardo workflows built around a night-flash photography base prompt with fixed lens, exposure, and ambient-light variables night-flash thread base prompt.
• The pattern is spreading beyond portraits: ProperPrompter uses a repeatable “scene to turnaround” template for character sheets turnaround demo, while another Nano Banana prompt system generates Notion-style icon sets that can be vectorized into larger asset packs Notion icons.

The clearest shift is structural. Instead of prompting Nano Banana like a chatbot, creators are writing prompts like production briefs with fields you keep and fields you swap. PromptsRef’s prompt library packages that approach at scale, with 400-plus entries split across Nano Banana Pro and Nano Banana 2 and aimed at repeatable editorial, cinematic, and social-photo outputs.

That same library is being used as a reference bank for increasingly rigid prompt formats, including scene-heavy portrait sets such as the bathhouse-style collage in scene collage. The point is less “copy this look” than “reuse this scaffold.”

The Nano Banana 2 portrait templates read like shot lists. One JSON prompt specifies an upside-down close-up portrait with 3:4 framing, warm directional indoor light, shallow depth of field, messy hair as an authenticity cue, plus a negative prompt that excludes studio polish and airbrushed skin. Another locks a bedroom mirror selfie to direct flash, readable reflected text, realistic mirror dust, and anatomy safeguards such as no duplicated phones or extra fingers mirror selfie JSON.

A separate recipe from ProperPrompter applies the same structure to character design: extract a subject from one image, then rebuild it using the turnaround layout from another reference, producing front, side, back, and close-up views turnaround demo. The follow-up example extends that template to Ghibli-inspired fan characters without changing the underlying format anime turnaround.

MayorKingAI’s Nano Banana 2 workflow inside Leonardo turns a style into a reusable lighting system. The base prompt fixes the core look—compact-camera direct flash, 28mm–35mm lens feel, slight distortion, digital noise, underexposed background, and hard highlight clipping—then swaps only angle, outfit, light sources, and palette.

The same “keep the structure, change one variable” logic shows up in icon work. Amir Mushich’s Notion icons prompt produces black-and-white Notion-style portraits for decks and web assets, and he argues in replies that a good vectorizer turns the setup into a “content factory” rather than a single finished image vectorizer reply.