Nano Banana 2: 10 Prompt Formulas That Actually Work (With Examples)

Nano Banana 2 — Google’s AI image generator built on Gemini Flash Image — has been making rounds for good reason. It handles subject consistency across multiple characters, renders text with surprising accuracy, and produces images sharp enough that you’ll stop apologizing before you share them. But like every AI image generator, it rewards people who know how to talk to it.

The difference between a generic output and something genuinely usable almost always comes down to prompt structure. Not magic words. Not secret handshakes. Just understanding what information the model needs, in what order, and how specific to get. These 10 prompt formulas cover the use cases that actually come up: portraits, product shots, editorial visuals, social media assets, and more.

Each formula below follows a proven structure, comes with a ready-to-copy example, and explains what’s doing the heavy lifting. Use them as templates, then swap in your own subject matter.

The Core Formula: Subject → Style → Lighting → Camera → Mood

Before the specific formulas, here’s the skeleton that underlies most of them. Every strong Nano Banana 2 prompt answers five questions: what is in the frame, what visual style applies, how is it lit, what camera/lens perspective, and what emotional register should the image carry. You don’t always need all five, but when a prompt fails, it’s usually because one of these is missing or contradictory.

Pro tip ✅

Put your most important element first. Nano Banana 2 reads prompts front-to-back and weights earlier tokens more heavily. Lead with subject, not with style adjectives.

Formula 1: The Editorial Portrait

This structure works for magazine-style single-subject portraits. It anchors on a specific person description, calls out a lighting setup, and specifies a publication aesthetic so the model has a clear visual reference category to pull from.

Portrait of a 40-year-old South Asian woman in a tailored charcoal blazer, sitting at a minimal concrete desk, Rembrandt lighting from camera left, shallow depth of field, shot on 85mm f/1.4, New Yorker magazine editorial style, neutral background, 4K

The phrase “New Yorker magazine editorial style” is doing real work here. It tells Nano Banana 2 the tonal register — restrained, intelligent, not flashy — without you having to spell out every visual parameter. Swap in “Wired magazine editorial” for a more tech-forward look, or “Vogue Italia editorial” for high-fashion contrast. The lighting callout (Rembrandt, camera left) prevents the flat, centered lighting that AI generators default to when you don’t specify.

Pro tip ✅

When specifying ethnicity or age in portraits, be concrete. “South Asian woman, 40s” generates more consistent, realistic results than vague descriptors like “mature professional.” Nano Banana 2 handles demographic specificity well — use it.

Formula 2: The Product Shot on White

Clean product photography is one of Nano Banana 2’s strongest suits. The key is telling it exactly what the product is, what material it’s made of, and that you want a commercial studio finish — not an “artistic interpretation.”

Studio product photography of a matte black stainless steel water bottle with a brushed aluminum cap, placed on a pure white surface, soft box lighting from above and right, sharp focus across entire product, no shadows, commercial photography style, isolated background, 4K resolution

Adding “no shadows” and “isolated background” are the two parameters that keep product shots clean. Without them, Nano Banana 2 tends to add atmospheric ground shadows that look fine artistically but kill the image for e-commerce use. “Sharp focus across entire product” prevents the shallow-depth-of-field instinct the model has when it sees “studio photography.”

Lifestyle product photography of a ceramic pour-over coffee dripper on a reclaimed wood kitchen counter, morning light from a window at camera right, steam rising from the coffee, warm color grading, slightly shallow depth of field, cozy hygge aesthetic, food photography style

The lifestyle variant trades sterility for context. “Hygge aesthetic” pulls the color palette warm and desaturated simultaneously — it’s a compact instruction that does the work of five separate color directives. Steam rising adds life without requiring any animation parameters.

Formula 3: The Cinematic Scene

This is where Nano Banana 2’s 4K output and subject consistency features start to matter. If you’re building a multi-shot sequence or need a cinematic still, this formula anchors on a specific film grammar.

Cinematic wide shot, two astronauts in worn white spacesuits standing on a red Martian ridge at dusk, one pointing toward the horizon, backlit by a deep orange sunset, lens flare on camera right, grain overlay, anamorphic widescreen format, color grade like Interstellar, atmospheric haze in distance

“Anamorphic widescreen format” pushes the model toward the letterboxed aspect ratio and characteristic oval lens bokeh of cinema lenses. “Color grade like [film title]” is one of the most efficient instructions you can give — it packages color temperature, saturation, contrast, and shadow detail into a single reference. “Grain overlay” tells it this should feel like celluloid, not digital render.

Pro tip ✅

For subject consistency across multiple images — one of Nano Banana 2’s headline features — describe your characters with a fixed physical description block and reuse it verbatim across every prompt in the sequence. The model tracks consistent subjects across up to five characters when you hold the descriptions steady.

Formula 4: The Social Media Hero Image

Social assets have specific constraints: they need to read at thumbnail size, they usually need space for text overlay, and they need enough visual pop to stop a scroll. This formula builds those requirements in from the start.

Bold flat-design illustration of a neon green plant growing through a cracked concrete wall, dark navy background, high contrast, negative space on left third for text overlay, social media thumbnail composition, graphic design aesthetic, no gradients, vector art style

The “left third for text overlay” instruction is practical and specific — it tells the model to leave compositional breathing room rather than fill every pixel with content. “No gradients, vector art style” keeps the output usable at multiple sizes without compression artifacts eating the detail.

Minimalist Instagram carousel cover image, overhead flat lay of a desk with a notebook, coffee cup, and small succulent plant, all objects in white and sage green tones, soft natural light, lots of negative space, clean lifestyle aesthetic, square crop composition

“Square crop composition” is worth including explicitly for Instagram output. The model defaults to landscape unless you specify otherwise. “Flat lay” plus “overhead” is redundant in theory, but in practice doubling up on the perspective instruction locks it in reliably.

Formula 5: Precise Text Rendering

Nano Banana 2 handles in-image text better than most AI image generators, but it still needs careful instruction. The formula: tell it exactly what the text says, how it should be styled, and where it sits in the frame.

Clean typographic poster design, large bold sans-serif text reading "STAY CURIOUS" centered at top third, white text on deep indigo background, minimal geometric decoration, single thin horizontal rule below text, poster design style, high contrast, print-ready quality

Putting the exact text in quotes (“STAY CURIOUS”) is non-negotiable. Short text strings of one to three words have significantly higher accuracy rates than full sentences. If you need a sentence, break it into a separate text element instruction and keep each string under four words. “Print-ready quality” nudges the model toward crisp edges on letterforms rather than the slight soft-focus it sometimes applies.

Warning ⚠️

Text rendering accuracy drops sharply for strings longer than six words. For anything beyond a short headline, use Nano Banana 2 for the background and visual treatment, then add text in a design tool. Don’t fight the model on long-form type.

Formula 6: The Architectural / Interior Shot

Interior and architectural prompts reward specificity about materials, time of day, and design era. Generic prompts produce generic interiors.

Architectural interior photography of a mid-century modern living room, low teak credenza against a wall of floor-to-ceiling windows, afternoon golden light casting long shadows on polished concrete floor, Eames lounge chair in foreground, no people, clean and quiet atmosphere, shot with 24mm wide angle, symmetrical composition, 4K

Naming specific furniture pieces (Eames lounge chair, teak credenza) gives the model precise object references rather than making it guess what “mid-century modern” looks like to you specifically. “No people” is always worth stating explicitly in interior shots — the model assumes human presence unless told otherwise.

Formula 7: The Brand Character Sheet

This is the subject consistency formula in action. Use it when you need a character shown from multiple angles or in multiple contexts — for a children’s book, game asset, brand mascot, or illustration series.

Character design sheet, cute round orange tabby cat mascot wearing a small blue postal worker uniform and cap, shown front view, side view, and three-quarter view, white background, flat illustration style, clean lines, consistent design across all three views, brand mascot style, vector art

Asking for multiple views in a single prompt is the most reliable way to get consistency — the model renders all views from the same character definition rather than regenerating independently. “Consistent design across all three views” seems obvious to state, but including it measurably improves output. This is also where Nano Banana 2’s character consistency architecture earns its keep.

Note 💡

All images generated by Nano Banana 2 carry SynthID watermarks — Google’s invisible watermarking system. These persist through screenshots and most editing operations. For commercial use, verify the licensing terms of your access tier before publishing.

Formula 8: The News / Editorial Concept Image

This formula is useful for blog headers, article thumbnails, and editorial features where you need a conceptual image that illustrates an idea without being literal.

Editorial conceptual photograph, a single chess king piece casting a long shadow shaped like a skyscraper on a marble surface, dramatic side lighting, dark background, metaphorical business concept image, New York Times Magazine style, medium format aesthetic, 4K

Metaphorical editorial images live or die on a strong central concept. The chess piece and skyscraper shadow is a classic power/business visual shorthand — the model knows this visual language well. “New York Times Magazine style” signals: serious, high contrast, slightly desaturated, not decorative. You can swap the publication reference to shift the visual register significantly.

Formula 9: The Nature / Landscape Hero

Landscape prompts need time of day, weather condition, specific geography, and a camera perspective. Without those four anchors, you get stock photography clichés.

Landscape photography of a Scottish Highland valley just after dawn, low mist in the glen, golden hour light breaking through storm clouds on the left horizon, dramatic crepuscular rays, a single stone cottage with lit windows in the middle distance, shot on 50mm, f/8 for deep focus, moody and vast atmosphere, no filters, photorealistic, 4K

“Crepuscular rays” — the beams of light through cloud gaps — is a term the model recognizes precisely and renders well. “No filters” tells it to skip the Instagram-era over-processing. The lit cottage window in the middle distance adds a human element that gives the scale of the landscape context without making it a portrait.

Formula 10: The Abstract / Hero Art

Abstract art prompts need visual anchors despite the lack of literal subject. Give the model a color palette, a texture reference, a movement quality, and an art context.

Abstract expressionist painting, swirling deep cobalt blue and burnt sienna oil paint, thick impasto texture with visible brushstrokes, influenced by Franz Kline and Mark Rothko, emotional and turbulent energy, gallery art print quality, no geometric shapes, no text, 4K resolution

Referencing specific artists (Franz Kline, Mark Rothko) works better in abstract prompts than in representational ones, because the model maps the aesthetic vocabulary of abstract artists more cleanly than it replicates a realistic photographer’s exact style. “Thick impasto texture” gives the model a physical material behavior to simulate, which results in much more convincing painterly outputs than “oil painting style” alone.

Pro tip ✅

Nano Banana 2 responds well to negative instructions in abstract prompts. “No geometric shapes, no text” actively prevents the model from defaulting to decorative patterns it likes to insert. In representational prompts, “no people” and “no watermarks” are worth including by default.

Pro tip ✅

When using Nano Banana 2 via AI Studio or the Gemini API, you can iterate prompts programmatically — send batch variations of the same base prompt with single parameter changes and compare outputs side by side. This is significantly faster than manual iteration in the Gemini app and is the workflow for anyone doing serious production work.

Take These and Run

These 10 formulas cover the vast majority of real production scenarios: portraits, products, editorial, social, text-in-image, architecture, characters, concepts, landscapes, and abstract art. The underlying logic is always the same — give the model enough specificity that it stops guessing, but leave enough creative room that it can do what it’s actually good at. Over-specified prompts produce flat, labored outputs. Under-specified ones produce generic noise. The formulas above live in the productive middle ground.

Nano Banana 2 is genuinely capable of producing images that would have required a day of Midjourney V7 iteration six months ago. The subject consistency feature alone makes it the right tool for anyone building visual series or character-driven content. Start with any of these formulas, change one variable at a time, and you’ll have a library of working prompts customized to your use case within an hour.

promptyze

See Full Bio