Blog / 9 min read · 2026-09-29

Nano Banana 2 Prompt Template: Copy-Paste Formats for Better Images

Copy a Nano Banana 2 prompt template for image generation or editing, then fill subject, composition, lighting, and fidelity constraints with confidence.

Here are copy-paste Nano Banana 2 prompt templates for creating and editing images, with clear places to specify the subject, composition, lighting, and what must stay unchanged. Nano Banana 2 is Google’s name for Gemini 3.1 Flash Image, which supports image generation, editing, multiple reference images, character consistency, and output up to 4K in the Gemini API. Google AI for Developers Use the template that matches your job, then replace every bracketed instruction before sending it.

The all-purpose Nano Banana 2 prompt template

Create a [image type] of [main subject].

Subject:
- [Describe the subject’s identity, materials, colors, clothing, expression, or defining features.]
- [Describe the action or pose.]
- [Add essential secondary objects only.]

Composition:
- [Camera angle or viewpoint.]
- [Framing: close-up, medium shot, full scene, overhead, etc.]
- [Where the subject sits in the frame.]
- [Aspect ratio: 1:1, 4:5, 16:9, 9:16, or another available format.]

Lighting and atmosphere:
- [Lighting direction and quality.]
- [Time of day, mood, weather, or background treatment.]
- [Style references described as visual qualities, not artist names.]

Fidelity constraints:
- Keep [specific person, product, logo, garment, object, or layout] recognizable.
- Preserve [colors, proportions, text, facial features, materials, or other details].
- Do not add [unwanted elements].
- No text except: “[exact text, if needed].”

Output:
- [Final use, such as editorial photo, product listing image, poster, or social post.]

The order is deliberate. Put the non-negotiable visual identity first, then establish the frame, then the atmosphere. Finish with preservation rules because they tell the model what not to reinterpret.

Nano Banana 2 is a useful fit for this format because Google documents support for up to four character references for consistency and up to 10 object references with high fidelity in one workflow. Google AI for Developers That makes “preserve this” instructions more practical when you attach relevant source images instead of relying on a vague text description.

How to fill each part

Subject: describe what identifies the thing

Write details that would let a photographer recognize the subject, not broad labels.

Weak:

A woman holding coffee.

Better:

A woman in her early thirties with a blunt black bob, wearing a camel wool coat and small gold hoop earrings, holding a white ceramic cup with both hands.

For an uploaded reference, say what the file represents and what must remain true:

Use the uploaded portrait as the identity reference. Keep the same face shape, hairstyle, skin tone, and gold hoop earrings.

Do not pile on decorative facts that do not matter. If the coat color is irrelevant, leave it out. Every extra detail is another chance to create a conflict.

Composition: make the frame measurable

“Cinematic” is mood, not composition. Name the view and placement.

Eye-level medium portrait, from mid-torso upward. Subject placed on the right third of the frame, looking toward open negative space on the left.

For a storefront image, you might use:

Straight-on exterior view from sidewalk height. Center the storefront, leave clear space above the entrance for a headline, and show the full door and both display windows.

If you use the API, aspect ratio and image size can be set in response_format; Google’s documentation lists 1K, 2K, and 4K output options for Gemini 3.1 Flash Image. Google AI for Developers In a chat interface, put the intended aspect ratio in the prompt, but check the final crop before treating it as production-ready.

Lighting: choose a source, not an adjective pile

Specify where light comes from and how it behaves.

Soft window light enters from camera left. Gentle shadow falloff on the right side of the face. Warm late-afternoon color temperature. Background is two stops darker than the subject.

A product alternative:

Large diffused key light from front-left, subtle rim light on the bottle’s right edge, soft shadow directly beneath the product, clean pale-gray seamless background.

Avoid mixing incompatible directions, such as “hard noon sunlight” and “soft candlelit studio lighting,” unless you genuinely want a composite image.

Fidelity constraints: name what cannot move

Fidelity constraints are most important for edits, products, people, and branded material. Be concrete about both the protected elements and the allowed change.

Keep the bottle shape, cap, label text, logo placement, and label colors unchanged. Replace only the background with a sunlit limestone bathroom. Do not add props or change the camera angle.

That is more reliable than “make it luxurious but keep it the same.” For person-focused work, see these Gemini photo-editing prompts that preserve the face for additional preservation phrasing.

Copy-paste templates by use case

1. Photorealistic portrait template

Create a photorealistic [vertical or horizontal] portrait of [person].

Subject:
- [Age range, hairstyle, clothing, expression, and pose.]
- Keep the uploaded reference person recognizable: preserve face shape, skin tone, hairstyle, and defining features.

Composition:
- [Close-up or medium portrait], [camera angle].
- [Subject placement] with [background description].
- [Aspect ratio].

Lighting:
- [Main light source and direction].
- [Shadow quality and color mood].

Fidelity constraints:
- Keep the person’s identity consistent.
- Do not alter facial proportions, add extra accessories, or change the hairstyle.
- No text or watermark.

Example:

Create a photorealistic 4:5 editorial portrait of the uploaded reference person.

Subject:
- Keep the same face, short curly brown hair, and green eyes.
- Dress them in a charcoal overshirt over a cream T-shirt, seated casually with one elbow on a café table.

Composition:
- Eye-level medium portrait, subject on the left third, softly blurred café interior behind them.

Lighting:
- Soft overcast daylight through a window from camera right, natural skin texture, subdued warm-brown palette.

Fidelity constraints:
- Preserve identity and facial proportions.
- Do not add glasses, facial hair, text, or additional people.

2. Product photography template

Create a premium product photograph of [product].

Subject:
- Show [product details that must remain exact].
- Include [one or two props] only if they support the product story.

Composition:
- [Angle], [crop], and [placement].
- Leave [location] as uncluttered negative space.

Lighting:
- [Key light], [fill or rim light], and [shadow style].

Fidelity constraints:
- Keep the product’s shape, materials, logo, label text, and colors unchanged.
- Do not invent extra buttons, packaging, claims, or logos.
- No text except: “[exact text].”

For a broader bank of product shot ideas, use these Nano Banana product-photography templates.

3. Precise image-editing template

Edit the uploaded image.

Change only:
- [Specific change.]

Keep unchanged:
- [Person or product identity.]
- [Pose, camera angle, crop, lighting, clothing, objects, and background elements that must remain.]

New visual details:
- [Material, color, environment, or style needed for the changed area.]

Boundaries:
- Do not add, remove, move, or redesign anything outside the requested change.
- Preserve all existing text exactly.

Example:

Edit the uploaded kitchen image.

Change only:
- Replace the cabinet doors with matte sage-green shaker cabinets and brushed brass handles.

Keep unchanged:
- Preserve the room layout, countertop, backsplash, appliances, camera angle, window light, floor, and all objects on the counter.

New visual details:
- Natural matte paint finish, narrow recessed panels, understated brass hardware.

Boundaries:
- Do not alter the cabinet dimensions or add decor.

4. Text-heavy infographic template

Nano Banana 2 supports legible text rendering and localization according to Google’s documentation, but important published text still needs a human proofread. Google AI for Developers

Create a [aspect ratio] infographic about [topic] for [audience].

Content:
- Headline: “[exact headline]”
- Section 1: “[exact label]” — “[exact supporting text]”
- Section 2: “[exact label]” — “[exact supporting text]”
- Section 3: “[exact label]” — “[exact supporting text]”

Composition:
- Use a clear visual reading order from [top to bottom / left to right].
- Reserve space for each text block; do not overlap text and illustrations.

Style and lighting:
- [Palette, illustration style, texture, and background.]

Fidelity constraints:
- Render every quoted phrase exactly.
- Do not add facts, numbers, logos, or claims not supplied above.

A simple revision method

Do not rewrite a prompt from scratch after each result. Keep the successful sections and issue one bounded change:

Keep the previous image unchanged except for the following:
- Move the subject slightly farther from the camera.
- Replace the background with a muted olive studio backdrop.
- Preserve facial identity, pose, clothing, lighting direction, and crop.

Nano Banana 2 supports multi-turn image modification in the Gemini API, so a controlled revision is a documented workflow rather than a workaround. Google AI for Developers If several things are wrong, fix identity or layout first. Fine styling comes last.

FAQ

Is Nano Banana 2 the same as Nano Banana?

No. Google identifies Nano Banana 2 as Gemini 3.1 Flash Image, while the original Nano Banana refers to Gemini 2.5 Flash Image. Google AI for Developers Use the templates in this article for Nano Banana 2, especially when working with multiple references or asking for higher-resolution output.

How many reference images can I use with Nano Banana 2?

Google documents support for up to 10 object reference images with high fidelity and up to four character reference images for consistency in a Gemini 3.1 Flash Image workflow. Google AI for Developers Use only the references that serve a clear role, such as identity, product, garment, or environment.

How do I stop an edit from changing the face or product?

Attach the source image, explicitly list the protected traits, state the one permitted change, and add a boundary such as “do not alter anything outside the requested change.” A targeted edit prompt is usually stronger than a long request for a completely new scene.

Should I put an aspect ratio in the prompt?

Yes, state the desired ratio in the composition section. API users should also set the aspect ratio in response_format, which is the documented control for output formatting. Google AI for Developers

Before generating, check that your prompt answers four questions: what is the subject, where does it sit in the frame, where does the light come from, and what must remain unchanged. If one answer is missing, add it before you regenerate.