Nano Banana Prompts: Edits, Headshots & Products

Nano Banana prompts work best when they read like a creative brief, not a bag of keywords. Define the subject first, then describe scene, lighting, composition, and the single change you need in this run. If you are editing an existing image, say what must stay fixed before you describe the transformation.

Nano Banana prompt guide: quick answer

• Use C Dance AI when you want one workspace for text-to-image and image editing.
• Open the Nano Banana workspace with image editing selected when you already have a reference image and need controlled changes.
• Start from the Nano Banana model page when you need the current C Dance AI feature and pricing context before generating.
• Use Nano Banana 2 for faster iteration, more reference images, and Google Search grounding.
• Use Nano Banana Pro for stricter polish, cleaner textures, and presentation-quality product or portrait output.

The rest of this guide turns those rules into reusable prompt templates, settings choices, and failure fixes. If you want working prompt examples before you write your own, open the current Nano Banana prompt route and then return here to adapt the structure instead of copying random social prompts.

What Nano Banana means in C Dance AI

As of Monday, August 24, 2026, the current C Dance AI workspace exposes both nano-banana-pro and nano-banana-2 in text-to-image and image-editor modes. The current repository contract also shows three important differences that affect prompt writing:

Those limits matter more than generic prompt advice. If the product contract allows only eight inputs, writing a 14-image workflow into a public guide would be wrong even if the broader Google documentation describes bigger model families.

Which Nano Banana model should you choose?

Choose the model by job, not by hype.

Use Nano Banana 2 when:

• You want faster iteration on several prompt variants.
• You need the wider current aspect-ratio set, including ultra-wide or tall variants beyond the Pro set.
• You want to try Google Search grounding for current objects, places, or reference context.
• You need more than eight reference images in one run.

Use Nano Banana Pro when:

• The image is client-facing and polish matters more than extra throughput.
• Texture detail, headshot finish, or product presentation quality matters more than speed.
• The task involves fewer, better references and stricter style control.
• You are finalizing an approved concept instead of exploring five loose directions.

That split matches both the current Google documentation and the live C Dance AI workspace. The official docs describe Nano Banana 2 as the generalist workhorse and Nano Banana Pro as the premium precision option, and the local workspace preserves that distinction through feature exposure and credit framing.

The Nano Banana prompt formula that holds up

For most tasks, this order is reliable:

1. Subject or fixed asset
2. Scene or environment
3. Style and visual goal
4. Lighting
5. Composition or camera framing
6. One controlled transformation
7. Constraints on what must not change

For text-to-image, write all seven parts. For image editing, the first and last parts matter more than the middle because the uploaded image already defines a lot of what the model can see.

Bad prompt:

make this look premium, cinematic, polished, viral, luxury, high-end, realistic

Better prompt logic:

keep the bottle shape and label unchanged, remove the hand, place it on a pure white studio background, use soft commercial lighting, add a natural contact shadow, keep reflections restrained, and preserve accurate product proportions

The second version gives the model a job. The first version gives it adjectives.

How text-to-image prompts differ from image editing prompts

Text-to-image has to invent the whole frame. Image editing has to protect an existing frame while changing part of it.

In text-to-image mode

Write the whole shot:

• who or what is visible;
• where it appears;
• what the lighting should feel like;
• how the frame is composed;
• what stylistic direction should dominate.

In image-editor mode

Start with preservation:

• what identity must stay fixed;
• what product shape or label must remain accurate;
• what composition must stay stable;
• what exactly should change after that.

This is the simplest way to reduce drift. Nano Banana is good at controlled edits, but only when the prompt makes the protected elements explicit.

Copy-ready Nano Banana prompt templates

The templates below are starting structures, not guaranteed outcomes. Replace the bracketed parts with your own details and change one variable per run.

1. Professional headshot from a casual source photo

Use the uploaded image as the identity reference. Keep facial features, skin tone, hairstyle direction, and eye shape consistent.

Create a half-length professional headshot. Dress the subject in [navy blazer / charcoal suit / black crew neck], with a clean studio backdrop in [light gray / soft blue / white]. Use soft three-point portrait lighting, natural skin texture, realistic catchlights, and restrained retouching. Frame from chest up, camera at eye level, professional expression, polished but believable finish.

Do not change age, ethnicity, facial proportions, or smile shape. Do not add extra jewelry, text, or background objects.

2. E-commerce product cleanup from a handheld photo

Identify the main product in the uploaded photo and keep its shape, label, material finish, and brand colors accurate.

Remove the hand and all background clutter. Recreate the product as a premium e-commerce studio shot on a pure white background with a soft natural contact shadow. Use even commercial lighting, clean reflections, corrected perspective, realistic sharpness, and true-to-product color.

Do not redesign the package. Do not invent accessories. Do not stylize the label.

3. Reference-led poster or campaign image

Use the uploaded image as the subject reference. Keep the same face, hair, and outfit silhouette.

Place the subject in a [fashion poster / product campaign / editorial cover] scene with [specific environment]. Use [lighting style], [camera framing], and [color direction]. The result should feel like a finished campaign visual, not a casual snapshot.

Keep the subject recognizable. Preserve the main pose direction unless otherwise stated. No watermarks. No unrelated background people.

4. Room redesign from one source image

Use the uploaded room photo as the exact layout reference. Keep wall positions, window locations, and room proportions consistent.

Redesign the space in [minimal Japandi / warm modern / boutique hotel / soft industrial] style. Replace furniture, textiles, and decor to match that direction. Use natural daylight from the existing windows. Keep the camera angle and room geometry unchanged.

Do not move structural elements. Do not add impossible windows or doors. Do not change the room into a different floor plan.

5. Product lifestyle shot from a clean product reference

Use the uploaded product image as the exact product reference. Preserve shape, logo placement, cap details, label layout, and material finish.

Create a lifestyle product photo in [bathroom shelf / coffee table / kitchen counter / gym bag] context. Use soft editorial lighting, realistic shadows, and a believable environment that supports the product category. The product remains the hero object, sharp and readable, with surrounding props kept secondary.

Do not distort the product shape. Do not replace the label. Do not crop the product out of the hero position.

What good Nano Banana edits look like in practice

This is the simplest pattern to copy: keep one reference strong, define one visible change, and avoid stacking five style goals into the same run.

Use the example as a target workflow, not a promise of a first-run match. The prompt has one clear business task: isolate the product, remove the hand, keep the packaging accurate, and restage it under studio lighting.

The short demo above uses owned before-and-after examples already published on the C Dance AI Nano Banana model page. Use that same mindset when you review your own run: judge whether the model preserved the protected object or identity first, then judge whether the styling change succeeded.

When should you use Google Search?

Use Google Search on Nano Banana 2 only when the task actually needs current external context. In the current C Dance AI billing logic, that adds 2 credits per run, so it should solve a real ambiguity.

Reasonable use cases:

• travel or place-specific scene grounding;
• current packaging or location context;
• culturally specific reference details that the source image does not already carry;
• educational or infographic tasks that depend on fresher real-world information.

Bad reasons to turn it on:

• "maybe it will make everything better";
• a plain headshot run with a strong source image;
• basic white-background product cleanup;
• poster styling where the scene is already fully described.

Google Search is not a substitute for prompt clarity. It is a narrow grounding tool.

Why Nano Banana prompts fail

Most failures come from one of four causes.

1. Too many competing instructions

You asked for luxury, cinematic, minimal, colorful, moody, viral, soft, realistic, and stylized in one pass. The model has no dominant direction.

2. Weak preservation rules

In editing mode, you never said what had to stay fixed. So the model changed face shape, product proportions, or label details because it assumed those were flexible.

3. Too many reference images with no roles

More images do not equal more control. If one image defines outfit, another defines lighting, and a third defines pose, say that in the prompt logic. Otherwise the model has to guess which one wins.

4. No revision discipline

You changed scene, lighting, aspect ratio, style, and model at the same time. Now you do not know which variable helped.

The fastest repair is usually smaller, not bigger: one strong reference, one strong visual goal, one revision at a time.

A repeatable Nano Banana review loop

Use a four-pass loop instead of treating every generation like a final attempt.

Pass 1: structure

Check the core layout. Did the right subject, product, or room survive the transformation?

Pass 2: preservation

Check identity, product shape, logo, and composition drift. If these failed, do not waste time polishing style yet.

Pass 3: style

Once preservation holds, tune lighting, mood, and finish.

Pass 4: delivery

Check whether the image is actually usable in the target context: ecommerce listing, LinkedIn headshot, landing page hero, or internal concept review.

That workflow is more reliable than rewriting the whole prompt from scratch after every weak result.

When should you use GPT Image 2 instead?

Nano Banana is not the answer to every image job. If your main problem is text rendering, compositing precision, or a more exact text-heavy design task, compare the workflow with GPT Image 2 and browse the current GPT Image 2 prompt library.

In practical terms:

• choose Nano Banana when you want reference-heavy iteration, quick product cleanup, or strong image editing control;
• choose GPT Image 2 when the output needs cleaner text blocks, more exact compositing, or a stronger production default for design-heavy image generation.

The right decision is the model that reduces retries for your actual task, not the one with the loudest reputation.

How Nano Banana fits a broader creative workflow

Nano Banana often works best as the still-image stage inside a larger production system. You might clean a product shot, lock a brand-safe portrait, or build a campaign still first, then move that approved asset into another workflow.

Two existing C Dance AI posts cover the next stages of that workflow.

If the still image is already working but you need more advertising hooks, product-scene angles, or commercial concepts, continue with Seedance 2.0 product ad prompts. That article solves a different job: expanding campaign ideas and scene hooks after the still-image direction is already clear. Use Nano Banana first when the product packshot or portrait identity must survive. Use the product-ad prompt guide when the creative concept itself needs more directions.

If the still image is good and the next problem is motion rather than polish, compare the logic with the Kling 3 prompt guide. Kling 3 is a video prompt problem, not an image-editing prompt problem, but the same discipline still matters: protect the important subject, define one dominant action, and avoid changing every variable at once. The difference is that video prompts also have to describe timing and camera movement, while Nano Banana prompts usually succeed or fail on preservation, styling, and composition.

That workflow split is useful because it keeps each model doing the job it is actually good at:

• Nano Banana for controlled still-image generation and editing
• GPT Image 2 for text-heavy or compositing-sensitive image work
• Seedance 2.5 or Kling when the approved still now needs motion

Readers who keep those jobs separate usually waste fewer credits than readers who ask one model to behave like three different tools.

Final workflow to copy

If you need a safe default, do this:

1. Open C Dance AI and decide whether the job is text-to-image or editing first.
2. Launch the Nano Banana workspace with image editing selected if you already have a source image, or switch to text-to-image if the whole frame must be invented.
3. Keep the first run narrow: one subject, one scene, one style, one protected constraint.
4. Review preservation before polish.
5. Check Pricing before you scale a batch or enable extra grounded runs.

Keep a plain-text log of the exact prompt, model, aspect ratio, and one variable you changed. That tiny habit makes later fixes much faster.

It also gives you a clean comparison when you revisit the same campaign several days later.

That is the quickest path to usable output. Return to C Dance AI, run the Nano Banana workspace with image editing selected, and keep your revisions small enough that each run teaches you something.