✨ Create Stunning Videos with AI!Access Kling 3.0, Seedance, Veo, Flux, Nano Banana and more in one platform. Start Creating

Gemini image editing / field notes

Gemini photo prompts.
Real edits, explained.

A practical Nano Banana prompt library built around actual outputs. Remove distractions, combine references, preserve a product, render a headline and refine a photo one step at a time.

Explore tested promptsEdit your photo
5 worked examples2 named modelsUpdated September 19, 2026
One referenceA new medium
Original fictional cafe portraitNano Banana 2 watercolor edit of the same fictional portrait

Photo to watercolor

Nano Banana 2

Not a secret phrase.
A clear editing brief.

Good Gemini photo editing prompts identify the source, the change and the parts that must stay recognizable. This guide goes deeper into assigning reference roles, resubmitting images between edits and checking text and product details.

Need the basic method for backgrounds, lighting, clothing and identity? Start with the general AI photo-editing guide. It includes three identical-prompt comparisons rather than repeating this library.

HighReach is an independent AI platform, not Google Gemini. These examples were generated through fal.ai, not through the Gemini consumer app.

Know your model

Nano Banana is a family, not one setting.

These are the two versions used in our worked examples. Selecting the named model makes the setup clearer; it does not guarantee the same output.

Nano Banana 2

Gemini 3.1 Flash Image

gemini-3.1-flash-image

Used here for the source images, object removal, watercolor and the follow-up lighting edit.

Nano Banana Pro

Gemini 3 Pro Image

gemini-3-pro-image

Used here for combining the portrait and product references, and composing the short-headline poster.

Google's image-generation documentation identifies these model names and supports reference-led generation and editing. The original Nano Banana is Gemini 2.5 Flash Image; it is not the model behind these examples. Provider controls and consumer-app labels may differ.

Start with evidence

Two source images. Clear reference roles.

The portrait and bottle are fictional AI-generated examples, both made with Nano Banana 2. We reuse them so you can check what survived each edit. Upload photos you own or have permission to process.

When you supply multiple images, label their jobs: image 1 supplies the person and scene; image 2 supplies only the product. Do not ask the model to vaguely combine everything.

Open an image to inspect it. The prompt buttons do not transfer these images into your account. Use your own inputs, or download these fictional references to repeat the exercise.

Person + cafe scene
Person + cafe sceneSource / Nano Banana 2
Bottle + label
Bottle + labelSource / Nano Banana 2

Nano Banana prompts / tested library

Copy the brief. Inspect the result.

Five original workflows with the exact submitted prompts, visible inputs and named model versions. The practical lesson is in both the requested change and the details that need checking.

Object removalMultiple referencesText + productCreative styleSequential editing
Method, not a leaderboard. Generated September 19, 2026. One square 1K PNG per prompt, no fixed seed, no masks, no rerolls and no manual retouching. Website images are WebP copies. The blue-hour intermediate comes from our background comparison. A small, selected set of tasks cannot establish overall model quality.

01 / Local editNano Banana 2
Remove an object without repainting the portrait

Input images, in order

Input 1: portrait reference1

Upload the original cafe portrait.

Remove an object without repainting the portrait

Gemini 3.1 Flash Image / 1K / square

The mug is gone and the table grain continues through its former position. The newly exposed area beside the hand is generated, not recovered from hidden photo data. Check that hand against the source before accepting the edit.

Exact tested prompt

Edit the uploaded photo. Remove only the yellow mug and its handle, liquid and shadow from the left side of the wooden table. Reconstruct uninterrupted pale oak grain in that small area. Preserve both hands, the woman's face, skin texture, clothing, pose, wall, window, plant, existing lighting and exact square crop. Do not change anything else.

Use this prompt

Opens Image Variations with Nano Banana 2 selected and the prompt filled in. Upload the listed inputs in order and choose square / 1K output. Nothing is generated automatically.

Generation details

September 19, 2026. fal.ai endpoint: fal-ai/nano-banana-2/edit. One PNG requested; no fixed seed or mask. Displayed as WebP. This is a single example, not a model benchmark.

02 / Two referencesNano Banana Pro
Give each reference image a separate job

Input images, in order

Input 1: portrait reference1Input 2: product reference2

Upload the cafe portrait first and the bottle reference second.

Give each reference image a separate job

Gemini 3 Pro Image / 1K / square

The output places a blue bottle where the mug stood, with a readable NOMA label. Its scale is the model's visual interpretation, not a measured 25 cm placement. Compare the cap and proportions with the source before using it as a product image.

Exact tested prompt

Use image 1 as the portrait and scene reference. Use image 2 only as the product reference. In image 1 replace the yellow mug with the cobalt-blue NOMA bottle from image 2 standing upright on the left side of the wooden table, about 25 cm tall relative to the woman's hands. Preserve the bottle's cylindrical shape, black cap, blue finish and white label with the exact text NOMA. Match its light and contact shadow to the cafe. Preserve the woman's face, age, skin, hair, rust-red top, pose and both hands, and preserve the wall, window, plant, table and square crop. No other changes.

Use this prompt

Opens Create Image for multiple references with Nano Banana Pro selected and the prompt filled in. Upload the listed inputs in order and choose square / 1K output. Nothing is generated automatically.

Generation details

September 19, 2026. fal.ai endpoint: fal-ai/nano-banana-pro/edit. One PNG requested; no fixed seed or mask. Displayed as WebP. This is a single example, not a model benchmark.

03 / Text + productNano Banana Pro
Keep the product, add a short headline

Input images, in order

Input 1: product reference1

Upload the bottle reference, not the portrait.

Keep the product, add a short headline

Gemini 3 Pro Image / 1K / square

This output spells TAKE A BREATH correctly on two lines and retains a readable NOMA label. That success with a short headline does not establish reliability for small print or long copy. Add legal text in a layout editor and verify the bottle proportions.

Exact tested prompt

Create one square minimalist product poster using the uploaded bottle as the product identity reference. Keep its cobalt-blue cylinder, black screw cap and white NOMA label intact and clearly readable. Bottle upright in the lower two thirds on a warm sand-colored stone plinth, pale sky-blue background and a soft realistic shadow. In the upper quarter, add the headline exactly 'TAKE A BREATH' in large dark navy uppercase sans-serif type, on two balanced lines, with generous margins. Preserve the exact label text NOMA. No other text, logos, people or decorative stickers.

Use this prompt

Opens Image Variations with Nano Banana Pro selected and the prompt filled in. Upload the listed inputs in order and choose square / 1K output. Nothing is generated automatically.

Generation details

September 19, 2026. fal.ai endpoint: fal-ai/nano-banana-pro/edit. One PNG requested; no fixed seed or mask. Displayed as WebP. This is a single example, not a model benchmark.

04 / Creative transformationNano Banana 2
Change the medium, keep the composition

Input images, in order

Input 1: portrait reference1

Upload the original cafe portrait.

Change the medium, keep the composition

Gemini 3.1 Flash Image / 1K / square

The watercolor keeps the rust-red shirt, yellow mug and overall arrangement. Pencil-like contours make some facial lines more pronounced, so this is a stylistic interpretation rather than neutral retouching. Compare the face before sharing.

Exact tested prompt

Transform the uploaded portrait into a hand-painted watercolor on warm ivory textured paper. Preserve the same woman's recognizable face, brown skin, dark wavy hair, rust-red shirt, seated pose and both hands on the table, with the yellow mug left and plant right. Use transparent washes, fine pencil contours around the eyes and hands, burnt sienna and muted sage colors, soft pigment blooms around the background. Keep the original square composition. Make it visibly an illustration, not a photograph with a filter. No lettering or new objects.

Use this prompt

Opens Image Variations with Nano Banana 2 selected and the prompt filled in. Upload the listed inputs in order and choose square / 1K output. Nothing is generated automatically.

Generation details

September 19, 2026. fal.ai endpoint: fal-ai/nano-banana-2/edit. One PNG requested; no fixed seed or mask. Displayed as WebP. This is a single example, not a model benchmark.

05 / Multi-step editNano Banana 2
Refine an accepted edit using the original identity

Input images, in order

Input 1: background nano banana 21Input 2: portrait reference2

Upload the Nano Banana 2 blue-hour result first and the original cafe portrait second.

Refine an accepted edit using the original identity

Gemini 3.1 Flash Image / 1K / square

The street-cafe background survives the second pass, while the lighting change is subtle rather than a dramatic blue rim. The first input supplies the scene and the second supplies the original person. Both were explicitly resubmitted; this does not rely on chat memory.

Exact tested prompt

Image 1 is the edited blue-hour cafe scene to refine. Image 2 is the original portrait and is only the identity reference, not the background to copy. Keep image 1's street cafe background, warm blurred lights, square crop, table, yellow mug, rust-red top, hairstyle, pose and hands. Change only the light on the woman: subtle warm soft fill on her face and a faint cool-blue edge on the hair to connect her to the evening scene. Preserve the original face and skin texture from image 2. Do not restore the original wall or window, alter her expression, smooth her skin or add objects.

Use this prompt

Opens Create Image for multiple references with Nano Banana 2 selected and the prompt filled in. Upload the listed inputs in order and choose square / 1K output. Nothing is generated automatically.

Generation details

September 19, 2026. fal.ai endpoint: fal-ai/nano-banana-2/edit. One PNG requested; no fixed seed or mask. Displayed as WebP. This is a single example, not a model benchmark.

Working across steps

Keep an edit history, not a chain of guesses.

01 / Save the source

Keep the original portrait and the exact first prompt. Approve the background before asking for new light. If the face drifts, return to the original instead of polishing the wrong person.

02 / Resubmit the right inputs

For the worked follow-up, image 1 is the accepted blue-hour edit and image 2 is the original identity reference. The prompt explicitly says not to restore the original wall.

03 / Check the unchanged parts

Review the face, hands, cup and framing after every step. If your tool has only one reference slot, use the accepted result and restate the keep list; do not assume a separate request remembers the source.

Adapt the pattern

More Gemini photo prompt templates.

Eight additional starting points for portraits, restoration, products and creative scenes. These templates are untested: no output is claimed for them. Links open Nano Banana 2 as a starting model, not a claim of testing. Adjust the people, reference order and exact text to your project.

Untested template / Reference edit

Natural portrait from a reference

Use the uploaded portrait as the identity reference. Preserve the person's face, apparent age, skin tone, hairstyle and distinctive marks. Change the background to warm off-white and the lighting to soft window light. Keep the original clothes, expression, pose and crop. Natural skin texture, no beauty retouching or new accessories.

Use this prompt

Untested template / Two references

Two adults, two identity references

Image 1 shows adult A; image 2 shows adult B. Use each only as that person's identity reference. Create one relaxed waist-up photograph of both adults sitting side by side on a park bench, all faces clearly visible. Preserve each person's distinct face, age, skin tone and hairstyle. Keep their respective reference outfits. Soft overcast daylight, natural expressions, separate hands resting on their own laps. Exactly two people, no facial blending or text.

Use this prompt

Untested template / Reference edit

Restore a faded family photograph

Restore the uploaded family photograph conservatively. Reduce visible scratches and uneven fading while preserving the original people, facial features, apparent ages, clothing and composition. Keep the original black-and-white treatment and authentic grain. Do not invent crisp facial details that are not supported by the source, add people, modernize outfits or smooth away age lines.

Use this prompt

Untested template / Reference edit

Give a product a new set

Use the uploaded product photo as the only product reference. Keep the same package shape, proportions, material, cap and exact label wording. Place it upright on a pale stone surface with a softly defocused green background and one soft light from the left. Match the contact shadow to the base. No added label claims, props covering the package, extra products or changed branding.

Use this prompt

Untested template / Reference edit

Change the setting of a travel portrait

Edit the uploaded adult portrait into an imagined quiet coastal scene. Preserve the person's face, hair, skin tone, age, clothes and standing pose. Replace only the background with a pale stone terrace overlooking the sea. Match the horizon to the existing eye-level camera and keep gentle natural daylight. No landmarks, signs, extra people or travel-date text.

Use this prompt

Untested template / Text to image

Create a typographic print

Design one square editorial typography print on ivory paper. The only text is 'MAKE ROOM' on the first line and 'FOR IDEAS' on the second line, spelled exactly. Large dark-green sans-serif letters, generous outside margins and a small abstract terracotta circle below the text. Flat front-facing composition, no mockup frame, signatures, extra words or decorative letter substitutions.

Use this prompt

Untested template / Two references

Combine a room and a furniture reference

Image 1 is the room to edit. Image 2 is the chair to insert. Keep the room's walls, windows, floor, perspective and daylight unchanged. Place one chair from image 2 beside the window at a believable scale, preserving its silhouette, material and color. Match its perspective and contact shadow to the floor. Do not copy image 2's background or replace other furniture.

Use this prompt

Untested template / Text to image

Generate a cinematic portrait from text

A fictional adult ceramic artist with short curly dark hair, standing in an uncluttered pottery studio in a plain blue work apron. Waist-up portrait, hands resting separately on the workbench, a single unfinished clay bowl at the side. Soft late-afternoon window light, muted earthy colors, realistic skin and fabric texture, calm direct gaze, square composition. No text, logos or celebrity resemblance.

Use this prompt

Before you publish

Check what the model was asked to preserve.

Faces and consent

Compare apparent age, eye shape, hairline and distinctive marks. Get permission to edit and publish portraits. Do not present generated scenes as documentary evidence.

Products and typography

Inspect labels, packaging dimensions, cap shape and exact wording. Use original product cutouts or a layout editor if commercial accuracy must be pixel-exact.

Lighting and contact

Check whether the inserted object sits on the surface, casts a plausible shadow and matches the scene's light. Good lettering does not compensate for an implausible object.

Cost and repeatability

A prompt is repeatable; an output is not guaranteed to be. Review the current credit estimate and avoid repeated paid attempts when a small manual correction would solve the problem.

Frequently asked questions

Using Gemini prompts with your photos.

01What is the difference between Gemini and Nano Banana prompts?

Nano Banana names refer to Gemini-family image models. The useful distinction is the specific model and task, not a special set of magic words. Our tested examples identify Nano Banana 2 or Nano Banana Pro, the reference images and the exact prompt.

02How do I use a Gemini AI photo prompt with my own photo?

Choose an image-editing tool, upload a clear photo you have permission to use, and describe both the change and the details to preserve. In HighReach, tested prompts open the named model in Image Variations for one reference, or Create Image for multiple references. Upload the listed images yourself in the specified order and choose square, 1K output.

03Are these real Nano Banana results?

Yes. The five worked examples and two source references were generated through the named fal.ai Nano Banana endpoints on September 19, 2026. The blue-hour input comes from our same-photo comparison. No examples were rerolled or manually retouched. Website files are compressed WebP copies. Additional templates are clearly labeled untested.

04Which Nano Banana version should I select?

The worked examples use Nano Banana 2 (Gemini 3.1 Flash Image) and Nano Banana Pro (Gemini 3 Pro Image). Choose the version labeled on the example to repeat its setup. These few examples do not establish a best model for every task, and availability and credit costs can change.

05Can Nano Banana preserve the exact same face or product?

A clear reference and explicit preservation instructions help, but exact reproduction is not guaranteed. Compare faces, hands, labels, proportions and small details against the source. Use conventional compositing when a product or logo must remain pixel-identical.

06Do I need a new prompt for each step?

Write each follow-up around the next change and identify which result to continue from. Our multi-step example supplies the accepted background edit as image 1 and the original portrait as image 2. A standalone generation request should not assume it remembers previous images.

07Are Nano Banana prompts free?

These prompts are free to read and copy. Running an image model uses your provider's allowance or credits. HighReach shows the generation cost before you submit; this guide does not promise free or unlimited generations.

General editing methodPhotoshoot ideasCouple photo promptsFamily photo prompts

HighReach AI Studio

Your photo. One deliberate edit.

Upload a reference, adapt a prompt and choose an image model. Review the cost before generating, then compare the result with your original.

Open Image VariationsLearn the editing method