How to Transform Any Photo’s Style with Nano Banana 2

Photo style transfer used to mean either spending hours in Photoshop or getting results that looked like someone smeared petroleum jelly on your screen. Nano Banana 2, Google’s AI image generator built on Gemini 3.1 Flash Image and launched February 26, 2026, changes that equation considerably. The model handles subject consistency across up to five characters, renders text with actual precision, and outputs at 4K resolution — which means your style-transferred images don’t just look good in a thumbnail, they hold up when you zoom in.

This tutorial walks you through the full workflow: uploading a source photo, writing prompts that actually control the style transfer, avoiding the common mistakes that produce muddy results, and getting clean outputs whether you’re working in the Gemini app, AI Studio, or via the API. By the end, you’ll have a repeatable process and eight copy-paste prompts you can use immediately.

What You’ll Achieve

By the end of this tutorial, you’ll know how to take any photo — a portrait, a product shot, a street scene — and apply a specific visual style while keeping the subject intact. You’ll understand how Nano Banana 2’s subject consistency feature works in practice, how to write prompts that control style without destroying structure, and how to use real-time web grounding to reference specific artistic movements or photographers.

What You Need Before You Start

Access to Nano Banana 2 comes through several routes. The Gemini app (gemini.google.com) is the most accessible — sign in with a Google account, and image generation via Gemini 3.1 Flash Image is available directly in the chat interface. For more control over parameters and API access, Google AI Studio (aistudio.google.com) gives you a proper workspace. Developers working at scale can hit the Gemini API directly or use Vertex AI for enterprise-grade deployment. Antigravity, Google’s experimental interface for creative workflows, also surfaces Nano Banana 2 for users with access. Whichever route you take, the prompting logic in this tutorial applies across all of them.

Note 💡

Nano Banana 2 outputs carry SynthID watermarks — Google’s invisible digital watermarking technology embedded at the pixel level. They don’t affect visual quality and aren’t visible to the eye, but they’re there. If you’re using outputs commercially, factor this into your workflow and check the current terms of service for your access tier.

How Style Transfer Works in Nano Banana 2

Unlike older style-transfer models that essentially ran a texture filter over your image, Nano Banana 2 interprets style prompts semantically. You describe the aesthetic you want — lighting conditions, color palette, rendering technique, artistic era — and the model reconstructs your subject inside that visual context rather than just layering a filter on top. The subject consistency feature means that if you have a person, product, or scene in your source image, the model anchors to it and rebuilds it in the target style rather than hallucinating a different subject entirely. This is the difference between a result that looks like your photo got styled and one that looks like the model got bored and drew something else.

Real-time web grounding lets you reference specific artists, photographers, film stocks, or design movements and have the model actually understand what you mean — it can pull contextual knowledge about visual styles rather than working purely from training data. Prompts like “shot on Kodak Portra 400” or “in the style of Hiroshi Sugimoto” land with genuine accuracy as a result.

The Core Prompting Structure for Style Transfer

Every effective style transfer prompt in Nano Banana 2 has three components: a description of what to preserve from the source, a clear style target, and technical parameters for the output. Skipping any one of these tends to produce results where either the subject drifts, the style is vague, or the resolution and aspect ratio don’t match your use case. Here’s the structure: [preserve instruction] + [subject description] + [style target] + [technical parameters]. The prompts below follow this logic — dissect any of them and you’ll see all four components at work.

Pro tip ✅

Always describe your subject explicitly in the prompt even when you’re uploading a reference image. Nano Banana 2 uses both the image and the text prompt to anchor subject consistency — if your text prompt doesn’t mention the subject, the model weighs the style instruction more heavily and the subject can drift. Say “a woman in her 30s” or “a ceramic coffee mug” rather than leaving it implicit.

8 Copy-Paste Prompts for Photo Style Transfer

The following prompts are written to drop directly into Nano Banana 2 — either in the Gemini app or AI Studio. Each one targets a different use case: portrait photography, product shots, editorial, social media, and more. Swap in your own subject descriptions where noted.

1. Portrait → Cinematic Film Photography

Transform this portrait into a cinematic 35mm film photograph. Preserve the subject's facial features, expression, and pose exactly. Apply the visual aesthetic of 1970s Italian cinema: warm shadows, slight grain, overexposed highlights, muted earth tones, shallow depth of field. Shot on Kodak Vision3 500T film stock. 4K resolution, 2:3 aspect ratio.

This prompt works because it gives the model a specific decade, a specific film stock, and a specific regional cinema aesthetic — three anchors that triangulate a precise visual language rather than a vague “cinematic look.” The explicit preservation instruction up front tells the model to treat the subject as fixed and rebuild the environment around it.

2. Street Photo → Woodblock Print

Re-render this street photograph as a Japanese woodblock print (ukiyo-e style). Keep the scene composition and all architectural elements faithful to the source. Use flat color fills, bold black outlines, visible wood grain texture in the background areas, and a limited palette of indigo, vermilion, and ochre. 4K resolution, 16:9 aspect ratio.

Specifying the color palette prevents the model from defaulting to generic “illustration colors.” Ukiyo-e is a well-documented artistic movement, so web grounding here gives the model accurate stylistic reference rather than a vague approximation.

3. Product Shot → Bauhaus Editorial

Apply a Bauhaus design aesthetic to this product photograph. Preserve the product shape and proportions exactly. Surround the product with geometric primary color shapes (red, yellow, blue, black) arranged asymmetrically. Clean white background, sans-serif typography elements integrated as design features, bold graphic shadows. Flat lay perspective. 4K resolution, 1:1 aspect ratio for Instagram.

The 1:1 aspect ratio instruction directly serves the social media use case — Nano Banana 2 respects aspect ratio instructions reliably, which saves you the crop in post. The “typography elements integrated as design features” line gives the model license to add graphic text shapes without requiring you to specify exact copy.

4. Portrait → Studio Oil Painting

Paint this portrait in the style of Dutch Golden Age oil painting. Maintain subject likeness and lighting direction from the source image. Apply thick impasto brushwork in the background, smooth glazed technique on the face and hands, dark vignetting at edges, warm amber candlelight as primary light source. Chiaroscuro contrast. Museum-quality 4K resolution, 3:4 portrait format.

“Museum-quality 4K” is a useful phrase — it cues the model toward fine detail rendering rather than stylized looseness. The instruction to maintain the lighting direction from the source image is key here: it preserves the three-dimensional reading of the face while transforming the medium.

5. Landscape → Brutalist Architecture Poster

Transform this landscape photo into a Soviet-era brutalist travel poster. Preserve the main geographical features of the scene. Apply the visual language of 1960s USSR propaganda posters: simplified flat shapes, stark color contrast (deep red, cream, black), bold sans-serif Cyrillic-inspired lettering integrated into the composition, dramatic upward perspective. 4K resolution, 2:3 poster format.

This prompt demonstrates how style transfer works beyond portraits. The “upward perspective” instruction nudges the model to reframe the composition in a way that feels authentic to the reference era — not just applying colors but adjusting the compositional logic.

6. Any Photo → Wes Anderson Film Still

Recreate this image as a frame from a Wes Anderson film. Preserve all subjects and their positions. Apply precise symmetrical composition, pastel color palette (dusty pink, mint green, warm yellow), flat two-point perspective, period-accurate props from the 1960s–70s. Soft diffused lighting with no harsh shadows. Cinematic 2.39:1 aspect ratio, 4K resolution.

Wes Anderson’s visual language is distinctive enough that Nano Banana 2’s web grounding handles it with high accuracy. The symmetry instruction is the critical one — without it, you get Anderson’s color palette but lose the compositional signature that makes the style recognizable.

7. Headshot → Neon Noir Illustration

Convert this headshot into a neon noir digital illustration. Keep the subject's silhouette and facial structure intact. Apply deep black shadows, sharp neon accent lighting in electric blue and magenta, rain-slicked reflective surfaces in the background, urban nighttime setting implied through light leaks. Semi-realistic rendering with visible digital brushwork. 4K resolution, 9:16 format for Stories.

The 9:16 format instruction makes this prompt immediately usable for Instagram or TikTok Stories without any additional cropping. “Light leaks” is a specific enough visual instruction that the model knows to place environmental light sources at the frame edges rather than as the primary illumination.

8. Group Photo → Art Deco Travel Poster (Subject Consistency Demo)

Transform this group photo into a 1930s Art Deco travel poster. Preserve all five subjects' likenesses, clothing, and relative positions. Apply the visual language of Cassandre-era poster design: streamlined geometric shapes, metallic gold and navy color scheme, elongated elegant figures, bold decorative borders, sunburst motifs. All subjects rendered with equal stylistic weight. 4K resolution, 2:3 poster format.

This prompt pushes Nano Banana 2’s subject consistency feature with multiple people — explicitly stating “all five subjects” and “equal stylistic weight” instructs the model not to prioritize the central figure at the expense of background subjects, which is a common failure mode in multi-subject style transfers. Referencing Cassandre by name rather than just “Art Deco” gets a more precise stylistic result thanks to web grounding.

Pro tip ✅

When working with group photos, name the number of subjects explicitly in your prompt (“five people,” “three characters”). Nano Banana 2’s subject consistency handles up to five characters, but it needs the explicit count to allocate attention correctly. Without it, the model sometimes merges or drops peripheral subjects when the style transformation is heavy.

Prompt Variants: One Subject, Three Styles

Understanding how a single subject shifts across different style instructions is the fastest way to develop prompt intuition. Take a standard product shot of a glass bottle. Here’s how three prompt variants produce completely different outputs while keeping the object recognizable.

VARIANT A — Hyperrealist Product Photography:
Transform this bottle photograph into a hyperrealist studio product shot. Ultra-sharp focus, dramatic single-source rim lighting from camera left, deep black background, subtle surface condensation if applicable. Shot on medium format digital, 100mm lens equivalent. 4K resolution, 1:1.

VARIANT B — Vintage Botanical Illustration:
Render this bottle as a Victorian-era botanical illustration plate. Fine ink linework for the bottle outline, watercolor washes for color fill in muted greens and ambers, white paper texture visible, handwritten label annotations in period-accurate typography. 4K resolution, 3:4.

VARIANT C — Memphis Design Pop Art:
Reimagine this bottle in the Memphis Group design aesthetic (1980s). Bold primary colors, squiggly line patterns in the background, geometric shapes floating around the object, flat two-dimensional rendering with no photorealistic lighting, deliberately kitschy color clashes. 4K resolution, 1:1 for social media.

Same object, three completely different commercial applications — editorial, luxury packaging reference, social media content. Swapping the style target while keeping the preserve/subject/technical-parameters structure consistent is how you build a repeatable system rather than a one-off experiment.

Warning ⚠️

Heavy style transfers on faces can sometimes drift enough to change the subject’s apparent ethnicity, age, or gender — especially with illustrative styles that have strong conventions around how faces look. If subject fidelity on faces matters for your use case, add “preserve subject’s ethnic features, age, and gender exactly” to your prompt. It’s an explicit instruction the model respects and one that most people forget to include.

Working Across Platforms: Gemini App vs. AI Studio vs. API

The Gemini app is the right starting point for quick experimentation — you upload an image, type your prompt, and iterate conversationally. The chat-based interface lets you refine in natural language: “make the shadows warmer” or “increase the grain” without rewriting the entire prompt. For production workflows, AI Studio gives you a proper parameter panel, the ability to save prompt templates, and cleaner access to the 4K output without compression. The Gemini API and Vertex AI route is for anyone building this into an application or batch-processing large volumes of images — the same prompting logic applies, you’re just hitting the endpoint programmatically. Antigravity surfaces additional creative controls for users with access, including style intensity sliders that let you blend the source image and target style rather than doing a full transfer.

Pro tip ✅

In AI Studio, save your best-performing style transfer prompts as reusable templates using the “System Instructions” field. You can encode the preserve/subject/technical-parameters structure as a default instruction, then use the main prompt field just for the style target. This cuts prompt writing time significantly when you’re processing multiple images in a single session.

Pro tip ✅

Nano Banana 2 handles text rendering with genuine accuracy — one of the consistent weak spots in earlier AI image generators. If your style transfer target involves posters, editorial layouts, or any design work with text, include the exact copy you want rendered in quotes within the prompt. The model will place and style it appropriately rather than generating plausible-looking gibberish.

What to Do When the Style Transfer Drifts

The most common failure mode is style overriding subject — you ask for a bold graphic illustration and the model treats the subject description as a suggestion rather than a constraint. The fix is almost always the same: move your preservation instruction to the very beginning of the prompt, make it more specific, and reduce the intensity language in your style description. “Completely reimagine as” is a stronger style transfer instruction than “in the style of” — and sometimes that’s too strong. Swapping to “in the style of” while adding “preserve subject structure exactly” at the start usually recovers subject fidelity without killing the style transformation.

The second common issue is a muddy middle — the image lands somewhere between the source style and the target style without committing to either. This usually means the style description is underspecified. Adding a specific artist name, decade, geographic school, or technical medium (oil, watercolor, linocut, etc.) gives the model enough anchors to commit to a direction.

Avoid 🚫

Don’t ask for multiple simultaneous style transfers in a single prompt — “combine the color palette of Monet with the composition of Mondrian and the lighting of Caravaggio” produces outputs that satisfy none of the three references. Pick one dominant style and use the others as secondary modifiers if needed. Nano Banana 2 is strong on specific styles; it’s not a style blender.

Where This Actually Gets Useful

The realistic use cases for Nano Banana 2 style transfer fall into three categories that are worth being concrete about. First, content production at scale — social media teams that need the same product or portrait in multiple visual styles for A/B testing or seasonal campaigns. Second, pre-visualization — art directors and designers using style transfers as fast mood board material before committing to an actual shoot or illustration brief. Third, editorial and publishing — taking stock photography and transforming it into illustration-adjacent content that doesn’t look like stock photography. The 4K output resolution makes all three categories production-viable rather than just concept-viable, which is where most previous AI image generators stopped short. Generate, refine the prompt once or twice, export — that’s a realistic workflow now, not an optimistic one.

promptyze

See Full Bio