Gemini AI Photo Prompt Copy Paste: 64 Nano Banana Prompt Templates

Most Gemini AI photo prompt copy paste templates fail on the same line: the one the writer never wrote. Nothing in the prompt told the model which parts of the picture were not allowed to move.

That single omission is what turns a headshot refresh into a stranger’s face and a packaging mockup into a product that no longer matches the physical box. Every nano banana prompt below closes that gap before you paste anything into Gemini.

Sixty-four templates, and you can be pasting one inside thirty seconds. If you came here to copy a prompt and get on with your day, take the checklist below, pick your subject from the index, and go.

If you want to know why one prompt returns a better photo of you and another returns a stranger, everything after the templates explains it.

If you are still choosing a model rather than a prompt, the best AI image generators shortlist covers that decision first.

How These 64 Prompts Were Tested in Gemini

All 64 Gemini AI photo prompt copy paste templates were run in Gemini on Nano Banana Pro, formally Gemini 3 Pro Image, in July 2026. Each returned image was reviewed against the same six criteria: instruction compliance, identity preservation, edit boundaries, composition, text accuracy, and visible artefacts.

All 64 templates below include an output image from the corresponding test or illustration workflow.

Model names, resolutions, plan limits and watermark policy are separately sourced to Google’s official documentation and were checked on July 28, 2026.

Results vary with the source image, the active Gemini model, the account tier, and later model updates. A template that works on one portrait is a strong starting point on another, not a guarantee.

Gemini AI Photo Prompt Copy Paste by Subject

Three steps and you are done. Upload one clear, well-lit photo to Gemini, paste a template after replacing anything written in UPPERCASE, and send it.

Reply in the same thread if the result needs one correction.

All 64 templates are listed below, sorted by who or what is in your photo. Each template appears exactly once, and every entry links to the full prompt, its tested image, and its notes.

• A single portrait of one person: 2. Golden-Hour Portrait, 8. Indoor-to-Daylight Correction, 48. Podcast Portrait With Title Space, 49. Fashion Editorial With Negative Space
• A profile picture or avatar: 4. Clean Profile Avatar, 3. Black-and-White Founder Portrait
• LinkedIn and company profile assets: 1. Professional Headshot Refresh, 18. LinkedIn Thought-Leadership Banner
• A couple, a family, or a whole team: 39. Team Group Portrait, 59. Couple Cinematic Portrait, 60. Family Group Portrait, 64. Father and Child Portrait
• A festival, beach, birthday, or travel scene: 50. Travel Editorial Portrait, 61. Festival Traditional Wear Portrait, 62. Beach Golden Hour Portrait, 63. Birthday Studio Portrait
• A hairstyle, outfit, or background you want changed: 5. Remove One Distracting Object, 6. Replace the Background Only, 7. Change Clothing Color Only, 9. Preview a New Hairstyle
• A damaged, faded, cropped, or old photo: 29. Old Photo Restoration, 30. Extend the Canvas, 31. Repair a Compressed Product Photo, 58. Then and Now Composite
• A product on white or in a studio: 13. White-Background Product Hero, 21. Luxury Product Spotlight, 22. Product Detail Macro, 44. Packaging Localization Mockup
• A product in a real-life scene: 17. Instagram Product Launch, 23. Seasonal Product Campaign, 35. Lifestyle Product Scene, 36. Ingredient Flat Lay
• A room, kitchen, office, or building: 25. Living Room Restyle, 26. Kitchen Renovation Concept, 27. Home Office Moodboard, 28. Architectural Exterior Concept, 37. Furniture Replacement in a Room, 38. Floor Layout to Room View
• Anything with readable text or labels on it: 11. Quote Graphic, 12. Event Poster, 24. Ecommerce Feature Graphic, 45. Source-Checked Infographic, 46. Labeled Educational Diagram
• A social post, banner, cover, or thumbnail: 10. Blog Featured Image, 19. SaaS Feature Announcement, 20. Social Carousel Cover, 40. YouTube Thumbnail
• An illustrated or fictional character: 14. Sticker and Icon Sheet, 15. Children’s Book Illustration, 32. Anime Character Adaptation, 33. Fantasy Hero Portrait, 34. Boxed Collectible Figurine, 41. Consistent Character Sheet, 42. Character in a New Scene, 43. Four-Panel Comic Page
• A concept, storyboard, or brand metaphor: 16. Storyboard Keyframe, 47. Surreal Brand Metaphor
• A trending 2026 social style: 51. Imperfect Film Portrait, 52. Instant Film Frame, 53. Era Swap Portrait, 54. Key Art Poster, 55. Y2K Camcorder Still, 56. Motion Without Blur, 57. Scrapbook Layout

Two rules cover every row that contains a person. Preserve the face before you describe the style, and never run an edit on a face you do not have permission to use.

Gemini AI Photo Prompts for Portraits, Edits, and Simple Graphics

Sixteen Gemini AI photo prompt copy paste templates that need at most one uploaded image. Replace the uppercase variables in each one before sending.

Templates 1 to 9 are portrait and single-photo edits. Templates 10 to 16 generate from nothing: blog images, text graphics, posters, stickers, illustration and a storyboard frame.

1. Professional Headshot Refresh

Job: restyle. Input: one portrait. Crop: 4:5.

Tested Gemini output for the Professional Headshot Refresh prompt, showing a navy blazer, soft office background, and natural portrait lighting.

Using the uploaded portrait as the identity reference, create a polished professional headshot. Preserve the recognizable facial structure, skin tone, hairstyle, eye color, and natural proportions. Change only the clothing to a tailored navy blazer over a neutral shirt, and replace the background with a softly blurred modern office. Use flattering softbox lighting, an eye-level camera, an 85mm portrait-lens look, natural skin texture, and a confident relaxed expression. Crop from chest up in 4:5. Do not add text, logos, jewelry, or beauty-filter smoothing.

Customize: the blazer and the office for your own wardrobe and setting, and leave every preserve clause alone.

Run a whole team from one identical wording rather than tuning each portrait. Deleting the wardrobe clause is what turns four matching headshots into four people photographed in four different rooms.

2. Golden-Hour Portrait

Job: restyle. Input: one portrait. Crop: 4:5.

Tested Gemini output for the Golden-Hour Portrait prompt, showing warm sunset rim light, a rooftop setting, and a soft cinematic background.

Use the uploaded photo as the identity source. Preserve the same person, face shape, hairstyle, skin tone, and apparent age. Place the subject on a quiet city rooftop at golden hour with warm rim light, a soft breeze, and distant buildings in creamy bokeh. Create a cinematic medium close-up with an 85mm lens look, shallow depth of field, realistic pores, subtle film grain, and warm amber highlights balanced by cool shadows. Keep the pose natural and the wardrobe simple. Output 4:5.

Customize: the rooftop for a beach or open field. Keep the realistic pores clause, because it is what stops the skin going plastic.

3. Black-and-White Founder Portrait

Job: restyle. Input: one portrait. Crop: 3:4.

Tested Gemini output for the Black-and-White Founder Portrait prompt, showing one side light for a calm and credible studio look.

Create a high-contrast black-and-white founder portrait from the uploaded image. Preserve identity, expression, facial proportions, and hairstyle. Change the setting to a minimal studio with a charcoal backdrop. Use a single large side light, gentle shadow falloff, crisp eyes, realistic skin texture, and a waist-up composition. The mood should read calm, credible, and decisive rather than dramatic. Output 3:4 with no text or props.

Customize: the mood line if the subject reads too severe. The lighting stays; the adjectives are what move the result.

4. Clean Profile Avatar

Job: restyle. Input: one portrait. Crop: 1:1.

Tested Gemini output for the Clean Profile Avatar prompt, showing a refined illustrated portrait with a simple geometric background and clear facial recognition.

Turn the uploaded portrait into a clean professional avatar while preserving the person's recognizable features, hairstyle, skin tone, and expression. Use a centered head-and-shoulders composition, a simple geometric background in two brand colors, soft even lighting, and a refined semi-realistic illustration style with crisp edges. Keep the face natural and avoid exaggerated features. Output a 1:1 square with generous safe space around the head and no text.

Customize: name your two brand colors explicitly. Left unnamed, the colour pair changes between runs and the avatars stop matching.

5. Remove One Distracting Object

Job: remove. Input: one photo. Crop: source ratio.

Tested Gemini output for the Remove One Distracting Object prompt, showing one object removed while the surrounding composition, lighting, and textures stay intact.

Using the uploaded image, remove only OBJECT_VALUE. Reconstruct the hidden background naturally using nearby texture, lighting, perspective, and shadows. Preserve every other person, object, color, crop, camera angle, and detail exactly as in the original. Do not beautify, relight, sharpen, or redesign the image. Keep the original aspect ratio.

Customize: one object per attempt. Naming two removals in one prompt is where this template starts returning one correct edit and one invented one.

6. Replace the Background Only

Job: replace. Input: one photo. Crop: source ratio.

Tested Gemini output for the Replace the Background Only prompt, showing the subject preserved and the new environment matched through lighting, perspective, and shadow direction.

Use the uploaded image as the subject and identity reference. Preserve the subject's face, hair, body proportions, clothing, pose, edges, and expression. Replace only the background with SETTING_VALUE. Match the new environment's light direction, color temperature, depth of field, contact shadows, and perspective to the subject. Keep the original crop and aspect ratio.

Customize: the setting freely. The match clause is what stops the pasted-on look, so it never changes.

7. Change Clothing Color Only

Job: replace. Input: one portrait. Crop: source ratio.

Tested Gemini output for the Change Clothing Color Only prompt, showing the garment recolored while its texture, folds, highlights, and shadows remain intact.

Using the uploaded portrait, change only the clothing color from ORIGINAL_COLOR_VALUE to NEW_COLOR_VALUE. Preserve the person's face, skin tone, hair, body proportions, pose, fabric texture, garment design, folds, lighting, shadows, background, crop, and every other detail. Make the new color respond realistically to highlights and shadows. Keep the original aspect ratio.

Customize: a hex value instead of a color name when brand accuracy matters.

8. Indoor-to-Daylight Correction

Job: correct. Input: one indoor portrait. Crop: source ratio.

Tested Gemini output for the Indoor-to-Daylight Correction prompt, showing the warm color cast removed and natural skin detail preserved.

Correct the uploaded indoor portrait so it reads as if photographed in soft natural daylight. Preserve the same person, pose, expression, clothing, background layout, and crop. Remove the color cast, reduce sensor noise, recover realistic skin tone, soften harsh highlights, and keep pores and hair edges natural. Do not change facial features or apply beauty smoothing. Keep the source aspect ratio.

Customize: name the cast you have if the first pass overshoots. A wrong diagnosis produces a worse image than no correction.

9. Preview a New Hairstyle

Job: replace. Input: one portrait. Crop: 4:5.

Tested Gemini output for the Preview a New Hairstyle prompt, comparing six hairstyles while keeping the face, clothing, pose, and background consistent.

Use the uploaded portrait as the identity reference. Preserve the person's facial structure, skin tone, eye color, expression, apparent age, body proportions, clothing, background, and lighting. Change only the hairstyle to HAIRSTYLE_VALUE with a realistic hairline, density, strand direction, volume, and shadows. Keep the hair color COLOR_VALUE unless another color is explicitly approved. Output 4:5.

Customize: hair or wardrobe, not both in one attempt. Together they drift more than either alone.

10. Blog Featured Image

Job: generate. Input: none. Crop: 1.91:1.

Tested Gemini output for the Blog Featured Image prompt, showing a clear text-to-image metaphor that remains legible at thumbnail size.

Create a clean editorial featured image for an article about TOPIC_VALUE. Use one strong visual metaphor, refined three-dimensional styling, controlled depth, a simple background, and enough empty space for a content-management crop. Avoid screenshots, tiny icons, text, and clutter. The image should communicate the topic in one glance and stay legible at thumbnail size. Output an approximately 1.91:1 horizontal frame.

Customize: check the result at thumbnail size before you accept it. Legible at full size is not the test that matters here.

11. Quote Graphic

Job: generate. Input: none. Crop: 1:1.

Tested Gemini output for the Quote Graphic prompt, showing balanced line breaks, refined typography, and a minimal textured background.

Create an elegant social quote graphic. Render the exact quote "EXACT_TEXT" and the exact attribution "ATTRIBUTION_TEXT" in a readable serif and sans-serif pairing. Use a minimal background with subtle paper texture, balanced line breaks, generous margins, and one small decorative motif. Output 1:1. Do not alter punctuation, wording, or attribution.

Customize: shorten the quote before you change anything else. Long strings are where in-image text starts to break.

12. Event Poster

Job: generate. Input: none. Crop: 3:4.

Tested Gemini output for the Event Poster prompt, showing a clear information hierarchy, restrained colors, and one strong architectural motif.

Create a modern event poster for EVENT_NAME. Render the exact approved wording "EXACT_TEXT" with a clear hierarchy covering title, date, time, location, and call to action. Use STYLE_VALUE, a restrained color palette, one central visual motif, generous margins, and strong contrast. Output 3:4. Do not invent speakers, sponsors, dates, addresses, or logos.

Customize: the style and the motif. The do-not-invent clause is what keeps a fake sponsor logo out of a printed poster.

13. White-Background Product Hero

Job: restyle. Input: one product photo. Crop: 1:1.

Tested Gemini output for the White-Background Product Hero prompt, showing clean reflections, a realistic shadow, and the full product silhouette preserved.

Use the uploaded product as the shape, material, color, logo, and packaging reference. Preserve the product geometry, proportions, label placement, brand colors, and all approved wording exactly. Create a high-resolution studio hero image on a seamless white background with soft three-point lighting, a realistic contact shadow, clean reflections, and sharp focus on the product. Use a three-quarter camera angle that reveals the front and side. Output 1:1. Do not invent claims, ingredients, accessories, or wording.

Customize: a light grey backdrop when the product itself is white.

Read the returned label character by character before the file leaves the folder. A hero image that alters a certification mark or a volume figure stops being a photography problem and becomes a claims problem for whoever signs off the listing.

14. Sticker and Icon Sheet

Job: generate. Input: none. Crop: 1:1.

Tested Gemini output for the Sticker and Icon Sheet prompt, showing a cohesive set of tactile 3D icons with consistent lighting, spacing, and visual style.

Create a cohesive sheet of twelve original stickers representing ICON_LIST_VALUE. Use one consistent tactile three-dimensional style, bold silhouettes, clean edges, simple lighting, and a white background. Keep each icon separate with equal spacing and no text. Output 1:1. If any requested icon is missing or duplicated, generate the missing icons individually in the accepted style.

Customize: the icon list. The recovery clause at the end is the part that saves the sheet when one icon comes back twice.

15. Children’s Book Illustration

Job: generate. Input: none. Crop: 4:3.

Tested Gemini output for the Children’s Book Illustration prompt, showing friendly characters, textured gouache, and a clear area reserved for typesetting.

Create a warm children's-book illustration for SCENE_VALUE aimed at AGE_RANGE_VALUE. Use friendly shapes, expressive but non-frightening characters, textured gouache or watercolor, a clear focal action, and a calm background. Leave a clean area for text to be added later. Avoid branded characters, logos, and imitation of a living artist. Output 4:3 with no text.

Customize: write the character description once and reuse the exact wording in every scene.

16. Storyboard Keyframe

Job: generate. Input: none. Crop: 16:9.

Tested Gemini output for the Storyboard Keyframe prompt, showing clear silhouettes and strong foreground, midground, and background separation.

Create one cinematic storyboard keyframe for SCENE_VALUE. Show SUBJECT_VALUE performing ACTION_VALUE in SETTING_VALUE with clear staging, directional movement, foreground, midground and background separation, and a CAMERA_MOVE_VALUE composition. Use a realistic previsualization style with controlled color and readable silhouettes. Output 16:9 and include no text, arrows, or shot numbers.

Customize: one keyframe per attempt. A whole board in one prompt returns four versions of the same shot.

Gemini Prompts for Products, Interiors, Restoration, and Character Styles

Eighteen Nano Banana photo prompts for work where consistency with something that already exists is the point.

Templates 17 to 24 cover products and brand assets, and 25 to 28 cover rooms and buildings. Templates 29 to 31 cover restoration and repair, and 32 to 34 recast a person into an illustrated or collectible style.

17. Instagram Product Launch

Job: generate from a reference. Input: one product photo. Crop: 4:5.

Tested Gemini output for the Instagram Product Launch prompt, showing a clear product hero, one supporting prop, realistic shadows, and open space for campaign copy.

Create a premium Instagram launch image for BRAND_NAME featuring PRODUCT_VALUE as the hero. Use a clean studio composition, a bold brand-color background, soft directional lighting, realistic shadows, and one distinctive prop that communicates BENEFIT_VALUE without text. Leave clear empty space at the top for copy added later. Output 4:5 and do not render logos or claims unless supplied as an approved reference.

Customize: the prop, never the palette. Palette drift is what makes a feed look assembled by three people.

18. LinkedIn Thought-Leadership Banner

Job: generate. Input: none. Crop: 4:1.

Tested Gemini output for the LinkedIn Thought-Leadership Banner prompt, showing a simple growth metaphor, calm negative space, and a wide professional layout.

Create a professional LinkedIn banner illustrating TOPIC_VALUE through a simple visual metaphor. Use a wide editorial composition, a restrained brand palette, subtle depth, clean geometric forms, and a credible business aesthetic. Keep the left third visually calm for a profile-photo overlap and the center-right open for a headline added later. Output 4:1 with no text.

Customize: the metaphor. The quiet left third is a layout requirement, not a style choice.

19. SaaS Feature Announcement

Job: generate. Input: none. Crop: 16:9.

Tested Gemini output for the SaaS Feature Announcement prompt, showing one clear interface card, subtle gradients, and open space for approved launch copy.

Create a polished software feature-announcement graphic for FEATURE_NAME. Show a simplified generic interface card representing the feature without copying any competitor screen. Use a clean light background, subtle gradients, one focal screen, small supporting icons, and a premium product-marketing aesthetic. Leave room for an approved headline and call to action. Output 16:9 with no rendered text and no invented metrics.

Customize: nothing in the last sentence. Invented UI numbers in a product announcement are a credibility problem, not a design one.

20. Social Carousel Cover

Job: generate. Input: none. Crop: 4:5.

Tested Gemini output for the Social Carousel Cover prompt, showing a bold title, one central productivity concept, and a consistent blue-and-orange palette.

Create the cover image for a three-slide social carousel about TOPIC_VALUE. Design one bold central visual, a consistent brand palette, a clear focal point, and visual cues that can continue across later slides. Render only the exact cover title "EXACT_TEXT" in large readable type. Output 4:5. Generate the cover first, then continue one slide per turn in the accepted style.

Customize: the slide count. One slide per turn is the instruction that keeps the series visually consistent.

21. Luxury Product Spotlight

Job: restyle. Input: one product photo. Crop: 4:5.

Tested Gemini output for the Luxury Product Spotlight prompt, showing controlled highlights, a low three-quarter angle, and a dark premium studio setting.

Preserve the uploaded product's exact shape, materials, color, logo position, and package wording. Place it on a dark stone pedestal in a premium studio with a narrow spotlight, subtle atmospheric haze, controlled specular highlights, and a soft reflected glow behind it. Use a low three-quarter angle and a 70mm product-photography look. Keep the product fully legible and physically plausible. Output 4:5.

Customize: the pedestal to a lighter stone when the product is dark. A black product on a dark set disappears.

22. Product Detail Macro

Job: restyle. Input: one product photo. Crop: 4:5.

Tested Gemini output for the Product Detail Macro prompt, showing stitching, material texture, and craftsmanship in sharp close-up.

Create a macro product detail image from the uploaded reference. Preserve the exact material, surface pattern, stitching, finish, logo, and color. Frame the most important craftsmanship detail in extreme close-up with a 100mm macro-lens look, shallow depth of field, controlled side lighting, and realistic micro-texture. Do not alter or beautify the material beyond the reference. Output 4:5.

Customize: name one detail explicitly. A macro prompt with two subjects returns neither.

23. Seasonal Product Campaign

Job: generate from a reference. Input: one product photo. Crop: 4:5.

Tested Gemini output for the Seasonal Product Campaign prompt, showing restrained autumn props, a dominant product, and open space for campaign copy.

Preserve the uploaded product exactly, including geometry, materials, brand colors, logo, and package wording. Create a seasonal campaign scene for SEASON_VALUE using restrained props, a coordinated color palette, realistic lighting, and clear visual hierarchy. Keep the product unobstructed and dominant. Use a premium advertising style rather than a busy holiday collage. Output 4:5 and leave a clean area at the top for copy.

Customize: the season. The restrained-props clause is what stops the product becoming set dressing.

24. Ecommerce Feature Graphic

Job: generate from a reference. Input: one product photo. Crop: 16:9.

Tested Gemini output for the Ecommerce Feature Graphic prompt, showing a clear product hero with three concise feature callouts and simple supporting icons.

Create a clean ecommerce feature graphic using the uploaded product. Preserve the exact product design, logo, color, dimensions, and approved packaging wording. Position the product on the right against a light neutral gradient. On the left, render the exact headline "EXACT_TEXT" in a bold modern sans-serif font, plus three short approved feature labels supplied below. Use simple line icons, strong spacing, and high contrast. Output 16:9. Do not invent specifications, certifications, or performance claims.

Customize: the headline and the three labels, and supply all four as approved copy. Nothing here is a field the model gets to fill.

25. Living Room Restyle

Job: restyle. Input: one room photo. Crop: source ratio.

Tested Gemini output for the Living Room Restyle prompt, showing the room architecture preserved while furniture, textiles, and decor are updated.

Use the uploaded room as the exact architecture and layout reference. Preserve walls, windows, doors, ceiling height, floor arrangement, camera position, and fixed fixtures. Change only the furniture, lighting, textiles, and decor to STYLE_VALUE. Keep scale, perspective, shadows, and walking routes realistic. Output the same aspect ratio as the source. Do not move structural elements or enlarge the room.

Customize: the style word only. Two styles in one prompt return a compromise of both.

26. Kitchen Renovation Concept

Job: restyle. Input: one kitchen photo. Crop: source ratio.

Tested Gemini output for the Kitchen Renovation Concept prompt, showing the room layout preserved while cabinetry, surfaces, fixtures, and finishes are updated.

Preserve the uploaded kitchen's room dimensions, windows, doors, plumbing locations, appliance positions, and camera viewpoint. Restyle the cabinetry, counters, backsplash, lighting, and finishes to STYLE_VALUE using MATERIAL_VALUE and COLOR_PALETTE_VALUE. Keep the renovation physically plausible and maintain realistic clearances. Output the same ratio as the source.

Customize: finishes freely. Moving plumbing in a concept image is how a client budget gets set wrong.

27. Home Office Moodboard

Job: generate. Input: none. Crop: 16:9.

Tested Gemini output for the Home Office Moodboard prompt, showing furniture, materials, colors, and finished-room references in one editorial layout.

Create a coherent interior-design moodboard for a home office in STYLE_VALUE. Include a desk, ergonomic chair, task lighting, storage, material swatches, a restrained color palette, and two small reference vignettes showing the finished atmosphere. Use a clean editorial grid with no brand logos and no text. Output 16:9. The board should stay practical for SPACE_SIZE_VALUE and WORKFLOW_VALUE.

Customize: the style and the space size. The element list is what keeps the board from turning into decoration.

28. Architectural Exterior Concept

Job: restyle. Input: one building photo or approved sketch. Crop: 16:9.

Tested Gemini output for the Architectural Exterior Concept prompt, showing preserved floor count, massing, window rhythm, roofline, and realistic street-level landscaping.

Use the uploaded building image or approved design sketch as the geometry reference. Preserve the massing, floor count, window rhythm, roofline, and site boundaries. Render the exterior with MATERIAL_VALUE at TIME_OF_DAY_VALUE, realistic greenery, physically plausible shadows, and a street-level 35mm architectural-photography perspective. Do not add floors, entrances, or structural elements that are not shown. Output 16:9.

Customize: materials and time of day. An added floor in a concept render is the kind of error a client only spots on site.

29. Old Photo Restoration

Job: restore. Input: one damaged photo. Crop: source ratio.

Tested Gemini output for the Old Photo Restoration prompt, showing surface damage removed while the family’s identities, poses, clothing, and original composition remain preserved.

Restore the uploaded old photograph while preserving every person's recognizable identity, apparent age, pose, clothing, and the original composition. Remove dust, scratches, stains, fading, and fold marks. Rebuild only damaged areas, recover natural tonal range, and keep authentic film texture. Do not modernize faces, change expressions, invent missing people, or alter the historical setting. Keep the original aspect ratio. If colorizing, use historically plausible restrained colors.

Customize: nothing. This is the one template where loosening a preserve clause does real harm.

On a badly degraded face the model reconstructs plausible detail that was never in the negative, and the result looks convincing enough that nobody questions it later. Keep the original file, and never let a restored face stand as identification evidence.

30. Extend the Canvas

Job: generate from a reference. Input: one photo. Crop: target ratio.

Tested Gemini output for the Extend the Canvas prompt, showing the original center preserved and the surrounding scenery expanded naturally on both sides.

Extend the uploaded image to ASPECT_RATIO while preserving the original content, subject identity, camera angle, lighting, and scale. Generate only the missing canvas areas so the extension continues the existing environment naturally with matching perspective, texture, depth, and shadows. Do not crop, resize, or alter the original central content. Keep important subjects inside the safe area for OUTPUT_USE.

Customize: the target ratio. Extend in one direction at a time when the original has strong perspective lines.

31. Repair a Compressed Product Photo

Job: restore. Input: one low-quality product photo. Crop: source ratio.

Tested Gemini output for the Repair a Compressed Product Photo prompt, showing compression artifacts reduced while the product shape, label layout, and framing remain consistent.

Restore the uploaded product photo without redesigning the product. Preserve exact geometry, proportions, material, color, logo, label wording, reflections, and camera angle. Remove compression artifacts and noise, recover clean edges, improve micro-contrast, and rebuild only damaged pixels. Keep the background and crop unchanged. Do not invent wording, alter claims, smooth away material texture, or create a new product.

Customize: nothing. If the label is unreadable in the source, reshoot rather than let the model guess it.

32. Anime Character Adaptation

Job: restyle. Input: one portrait. Crop: 3:4.

Tested Gemini output for the Anime Character Adaptation prompt, showing clean linework, controlled cel shading, detailed eyes, and an original cinematic setting.

Use the uploaded person as the identity reference. Preserve recognizable facial structure, hairstyle, skin tone, outfit silhouette, and expression. Reinterpret the portrait as an original high-quality anime illustration with clean linework, controlled cel shading, detailed eyes, and a cinematic background inspired by SETTING_VALUE. Do not imitate a named living artist or a copyrighted franchise. Output 3:4.

Customize: the setting. The originality clause stays, because a named-artist request is the fastest route to a refusal.

33. Fantasy Hero Portrait

Job: restyle. Input: one portrait. Crop: 4:5.

Tested Gemini output for the Fantasy Hero Portrait prompt, showing detailed armor, dramatic natural lighting, atmospheric depth, and a heroic three-quarter pose.

Preserve the uploaded person's recognizable face, hairstyle, apparent age, and body proportions. Recast the subject as an original fantasy hero wearing ARMOR_STYLE_VALUE in SETTING_VALUE. Use dramatic but plausible lighting, detailed materials, atmospheric depth, and a heroic three-quarter pose. Keep the design original and avoid protected franchise symbols. Output 4:5.

Customize: armor and setting. Describing the armor in materials rather than by franchise name is what keeps it original.

34. Boxed Collectible Figurine

Job: restyle. Input: one portrait. Crop: 4:5.

Tested Gemini output for the Boxed Collectible Figurine prompt, showing a stylized character, coordinated accessories, clear-window packaging, and premium studio reflections.

Use the uploaded portrait as the identity reference and preserve recognizable facial features, hairstyle, skin tone, and key clothing details. Transform the subject into a stylized collectible three-dimensional figurine inside original toy packaging. Include accessories related to HOBBY_VALUE, realistic molded plastic, a clear window, studio reflections, and premium product photography. Use an original generic package design with no copyrighted brand or franchise. Output 4:5 with no text unless approved copy is supplied.

Customize: the hobby and its accessories. The generic-packaging clause is what keeps a real toy brand off the box.

Advanced Gemini Prompts: Multi-Reference and High-Stakes Jobs

Sixteen Gemini AI photo prompt copy paste templates that use more than one input or carry a cost when they go wrong. Expect a refinement turn on most of them.

Templates 35 to 43 combine two or more reference images. Templates 44 to 50 take one input but carry publishing risk, from translated packaging copy to sourced infographics and editorial frames.

35. Lifestyle Product Scene

Job: composite. Input: two images. Crop: 4:5.

Tested Gemini output for the Lifestyle Product Scene prompt, showing the product integrated naturally through matched perspective, lighting, reflections, and contact shadows.

Use IMAGE_1 as the exact product reference and IMAGE_2 as the environment and mood reference. Preserve the product's geometry, material, color, logo, and label wording. Place the product naturally in the referenced setting with matched perspective, lighting direction, reflections, and contact shadows. Add only minimal contextual props that support the use case. Keep the product as the focal point and do not redesign it. Output 4:5.

Customize: the environment reference. A mismatched light direction is the tell that gives a composite away.

36. Ingredient Flat Lay

Job: composite. Input: one product photo plus approved ingredient references. Crop: 1:1.

Tested Gemini output for the Ingredient Flat Lay prompt, showing the product centered and supporting ingredients arranged at realistic scale in soft natural light.

Use IMAGE_1 as the exact product reference and the remaining images only as approved ingredient references. Preserve the product's packaging, logo, label, and proportions. Arrange the product in the center of a top-down flat lay surrounded by the approved ingredients at realistic scale, with soft daylight, natural shadows, and a clean stone surface. Do not add any ingredient that was not provided. Output 1:1 and keep all packaging wording readable.

Customize: the ingredient references. Every one has to be supplied, because an unnamed ingredient is an invitation to invent one.

37. Furniture Replacement in a Room

Job: replace. Input: two images. Crop: source ratio of image 1.

Tested Gemini output for the Furniture Replacement in a Room prompt, showing the replacement sofa matched to the room’s scale, perspective, lighting, and contact shadows.

Use IMAGE_1 as the exact room and camera reference, and IMAGE_2 as the exact furniture reference. Preserve the room architecture, floor arrangement, windows, doors, lighting direction, and camera position. Replace only FURNITURE_TO_REMOVE with the referenced furniture, matching scale, perspective, contact shadows, and material reflections. Keep everything else unchanged. Output the same aspect ratio as IMAGE_1.

Customize: one item per attempt. Two swaps in one prompt usually returns one correct and one invented.

38. Floor Layout to Room View

Job: generate from references. Input: two images. Crop: 16:9.

Tested Gemini output for the Floor Layout to Room View prompt, showing a floor plan into a realistic interior while preserving the main room proportions and openings.

Use IMAGE_1 as the spatial layout reference and IMAGE_2 only as the visual-style reference. Create a realistic interior view from the designated camera location in ROOM_VALUE. Preserve the room proportions, door and window positions, and major walking routes shown in the drawing. Apply the style, palette, and materials from IMAGE_2 without copying its architecture. Output 16:9 with no labels.

Customize: the style reference. The without-copying-architecture clause is what keeps the plan authoritative.

39. Team Group Portrait

Job: composite. Input: several individual portraits. Crop: 16:9.

Tested Gemini output for the Team Group Portrait prompt, showing consistent lighting, perspective, camera height, and color temperature across the full group.

Combine the uploaded team portraits into one believable group photo. Treat each image as an identity reference and preserve each person's recognizable face, hairstyle, skin tone, body proportions, and clothing. Place everyone naturally around a wooden meeting table in a bright studio office with large windows. Match the lighting direction, camera height, perspective, shadows, and color temperature across all people. Use a 35mm group-photo look, relaxed professional expressions, and a 16:9 horizontal crop. The result must read as one photograph, not a collage.

Customize: the setting, never the identity clause.

Nano Banana Pro documents consistency across up to 14 images and up to 5 people, so a twelve-person company photo is several shorter runs rather than one job. Check every face against its source at full zoom, because a merged pair of colleagues is easy to miss on a phone screen and obvious in print.

40. YouTube Thumbnail

Job: generate from references. Input: one or two images. Crop: 16:9.

Tested Gemini output for the YouTube Thumbnail prompt, showing strong subject separation, a short bold headline, and a clear host-versus-product layout.

Use IMAGE_1 as the host identity reference and preserve the recognizable face, hairstyle, skin tone, and expression. Use IMAGE_2 only as the product or interface reference when provided. Create a high-contrast thumbnail with the host on the left, the product visual on the right, and strong lighting separation. Render the exact short headline "EXACT_TEXT" in large bold lettering with high readability. Output 16:9 and avoid tiny type.

Customize: the headline, and keep it to three words. Four starts to break at thumbnail scale.

41. Consistent Character Sheet

Job: generate from a reference. Input: one character image. Crop: 16:9.

Tested Gemini output for the Consistent Character Sheet prompt, showing four matching views of the same character at a consistent scale.

Use the uploaded character as the identity reference. Preserve the recognizable face, hairstyle, body proportions, clothing design, color palette, and signature accessories. Create a clean character sheet with front, three-quarter, side, and back views in neutral standing poses on a light background. Keep scale and design consistent across views. Add no labels or extra costumes. Output 16:9. If a view changes identity details, regenerate that view on its own.

Customize: nothing, and save the output as a file.

Build the sheet once and treat it as the source asset for everything downstream. Re-referencing the sheet each time costs an upload, while re-referencing the last output costs the character.

42. Character in a New Scene

Job: generate from a reference. Input: one character sheet. Crop: 16:9.

Tested Gemini output for the Character in a New Scene prompt, showing the established character in a new environment while preserving her design, proportions, clothing, and visual style.

Preserve the uploaded character's identity, face, hairstyle, proportions, clothing, and accessories. Place the character in SETTING_VALUE performing ACTION_VALUE. Match the scene lighting to the character, maintain plausible anatomy and perspective, and keep the same visual style as the reference. Use a medium-wide cinematic composition and output 16:9. Change only the environment, pose, and expression described here.

Customize: setting and action, one scene per attempt, always from the same sheet.

43. Four-Panel Comic Page

Job: generate from a reference. Input: one character sheet. Crop: 3:4.

Using the accepted character reference, create one four-panel comic page from SCRIPT_VALUE. Preserve the character's face, clothing, colors, and proportions across every panel. Use clear panel borders, readable visual sequencing, consistent perspective, and speech bubbles containing only the exact approved dialogue "EXACT_TEXT". Output 3:4. Verify each panel and regenerate a single panel if it drifts.

Customize: the script. Supply the dialogue as approved copy, because four panels of generated speech is four chances to misspell.

44. Packaging Localization Mockup

Job: replace. Input: one packaging photo. Crop: front-facing mockup.

Tested Gemini output for the Packaging Localization Mockup prompt, showing the package layout preserved and approved Spanish copy applied to the specified text areas.

Use the uploaded package as the exact design, geometry, color, logo, and layout reference. Replace only the specified text blocks with the human-approved translation "EXACT_TEXT". Preserve every other visual element, spacing system, icon, barcode area, and brand mark. Match the original font style and hierarchy as closely as possible. Output a front-facing high-resolution packaging mockup on a neutral background. Do not translate, paraphrase, or invent any wording.

Customize: one string per attempt, and have a native speaker read the output before it goes anywhere.

45. Source-Checked Infographic

Job: generate. Input: approved source data. Crop: 16:9.

Tested Gemini output for the Source-Checked Infographic prompt. Verify every figure and relationship against the supplied source before publishing.

Create a clean infographic about TOPIC_VALUE using only the approved source data SOURCE_DATA. Use a clear title "EXACT_TEXT", a simple hierarchy, restrained colors, readable labels, and no decorative statistic that was not supplied. Represent each number and relationship exactly as provided. Output 16:9. Do not calculate, estimate, translate, or invent figures.

Customize: the topic and the data. The do-not-calculate clause matters more than the design does.

46. Labeled Educational Diagram

Job: generate. Input: approved labels and relationships. Crop: 16:9.

Tested Gemini output for the Labeled Educational Diagram prompt. Treat the layout as a draft that still needs a subject-matter review.

Create a labeled educational diagram explaining PROCESS_VALUE. Use only the approved labels "EXACT_TEXT" and the approved relationships RELATIONSHIP_DATA. Arrange the process from left to right with clear arrows, distinct stages, consistent icons, and high-contrast text. Output 16:9. Do not add steps, mechanisms, figures, or terminology that was not supplied.

Customize: the process. Keep the label count low, and split the diagram rather than crowding it.

47. Surreal Brand Metaphor

Job: generate. Input: none. Crop: 16:9.

Tested Gemini output for the Surreal Brand Metaphor prompt. Check that the scene still obeys gravity, because the eye rejects it as a mistake when it does not.

Create an original surreal visual metaphor for BRAND_CONCEPT_VALUE. Show SUBJECT_VALUE interacting with SYMBOL_VALUE in an unexpected but physically coherent scene. Use cinematic lighting, refined materials, controlled color, a strong focal point, and enough empty space for campaign copy added later. Avoid existing logos, copyrighted characters, and named-artist imitation. Output 16:9 with no text.

Customize: the subject and the symbol. Physically coherent is the word doing the work, because it keeps surreal from becoming incoherent.

48. Podcast Portrait With Title Space

Job: restyle. Input: one portrait. Crop: 16:9.

Tested Gemini output for the Podcast Portrait prompt. Check the hands and the microphone edges at full resolution before publishing.

Use the uploaded portrait as the exact identity reference. Preserve face, hairstyle, apparent age, and skin tone. Place the subject in a premium podcast studio wearing the same clothing, seated beside a professional microphone with warm practical lights and acoustic panels behind. Create a three-quarter medium shot, natural hand placement, shallow depth of field, and realistic low-light photography. Output 16:9 with empty space on the right for a title.

Customize: which side the empty space sits on, to match your cover template. Watch the hands near the microphone, which is where this template needs the most checking.

49. Fashion Editorial With Negative Space

Job: restyle. Input: one portrait. Crop: 3:4.

Tested Gemini output for the Fashion Editorial With Negative Space prompt. Reserve the typography area at generation time rather than in post.

Use the uploaded person as the identity and body-proportion reference. Preserve the face, hairstyle, body proportions, and outfit structure. Create a high-fashion editorial photo in a minimalist concrete gallery with directional window light and long clean shadows. Use a full-body low-angle composition, a controlled pose, premium fabric detail, a muted neutral palette, and magazine-quality realism. Leave empty space in the upper left for typography. Output 3:4 and render no text.

Customize: the setting. The negative-space instruction is what makes the frame usable for layout.

50. Travel Editorial Portrait

Job: restyle. Input: one portrait. Crop: 4:5.

Tested Gemini output for the Travel Editorial Portrait prompt. Check the ankles and hands on the walking pose at full resolution before publishing.

Preserve the uploaded person's recognizable identity, face, body proportions, and outfit silhouette. Place the subject on a sunlit Mediterranean street with pale stone walls, blue shutters, and soft morning shadows. Create a candid walking pose, natural wind in the clothing, and a documentary travel-photography feel. Use a 35mm lens look, an eye-level perspective, balanced colors, realistic fabric texture, and a vertical 4:5 crop.

Customize: the location. Keep the documentary framing, because it is what stops the result reading as a stock advertisement.

Trending Gemini AI Photo Prompts for 2026

Fourteen Gemini portrait prompts beyond the core fifty, covering the styles and subjects that get searched by name.

Templates 51 to 58 are style and era treatments.

Templates 59 to 64 are the people categories: couple, family, festival wear, beach, birthday, and father with child. Same rules apply, and the consent check in Step 5 applies harder here because every one of these edits a real face.

51. Imperfect Film Portrait

Job: restyle. Input: one portrait. Crop: 4:5.

Restyle this photo as an imperfect 35mm film photograph. Add fine grain, one soft light leak in a corner, mild halation on highlights, faded blacks, and slightly warm desaturated color. Preserve the face, bone structure, skin tone, expression, and hair. Change only the color grade, grain, and background softness. Output 4:5. Do not smooth skin or remove blemishes.

Customize: 35mm for FILM_STOCK_VALUE such as a disposable camera or Super 8. The do-not-smooth clause stays in every version.

52. Instant Film Frame

Job: restyle. Input: one portrait. Crop: 1:1.

Place the uploaded photo inside an instant film frame with a wide white border and a slightly off-center image window. Apply a soft direct flash look with a hard shadow behind the subject, a mild cyan shift in the shadows, and a gentle vignette. Preserve the face and identity exactly. Change only the framing, grade, and lighting. Output 1:1 with no text, handwriting, or date stamp.

Customize: nothing involving a second person. The celebrity version of this trend is the one Gemini refuses, and the one with the clearest consent problem.

53. Era Swap Portrait

Job: restyle. Input: one portrait. Crop: 4:5.

Restyle the uploaded portrait as if photographed in DECADE_VALUE. Use period-correct wardrobe, hairstyle, film stock characteristics, and lighting, with a background consistent with the era. Preserve the facial structure, skin tone, and identity. Change the wardrobe, hair styling, background, and grade. Output 4:5, photorealistic rather than illustrated.

Customize: one decade per attempt. A range returns a blend that belongs to no era.

54. Key Art Poster

Job: generate from a reference. Input: one portrait. Crop: 2:3.

Build a cinematic poster around the uploaded person. Use dramatic low-key lighting, a strong rim light, and a muted teal and amber grade. Render the exact title "EXACT_TEXT" in large condensed type across the lower third, with a small credits block beneath in a clean sans-serif. Preserve the person's face and identity. Output 2:3 with sharp, fully legible text.

Customize: the title, and keep it under six words. Shorten it before changing anything else when the type comes back broken.

55. Y2K Camcorder Still

Job: restyle. Input: one portrait or group photo. Crop: 4:3.

Restyle the uploaded photo as an early-2000s camcorder still. Use a direct on-camera flash look, blown highlights, slight interlacing artifacts, and low saturation, with a timestamp in the lower right reading "DATE_VALUE". Preserve the faces, identities, and positions of everyone in frame. Change only the grade, lighting, and artifacts. Output 4:3 with no other text.

Customize: the date, or drop the timestamp entirely if it renders wrong twice.

56. Motion Without Blur

Job: restyle. Input: one action shot. Crop: 16:9.

Add directional motion to the uploaded photo. Streak the background with horizontal motion blur and add subtle light trails in the direction of travel. Preserve the subject completely sharp, with the face, clothing detail, and edges unblurred and unchanged. Change the background only. Output 16:9.

Customize: nothing. The split between the preserve and change clauses is the whole template.

57. Scrapbook Layout

Job: generate from a reference. Input: one to three photos. Crop: 4:5.

Arrange the uploaded photos as a scrapbook page with torn paper edges, tape at the corners, and a plain cream paper background. Add handwritten-style captions reading "EXACT_TEXT". Preserve every face and the content of every photo, unchanged. Change only the layout, background, and added paper elements. Output 4:5.

Customize: the captions, two at most. Handwriting style plus long strings is where the text breaks first.

58. Then and Now Composite

Job: composite. Input: one current photo and one childhood photo. Crop: 4:5.

Combine the two uploaded photos into a single frame. Place the adult and the child in the same scene with the same lighting and the same color grade, against a plain soft background. Preserve both faces exactly as they appear in their source photos. Change the background and lighting only, and do not age either face up or down. Output 4:5, photorealistic.

Customize: nothing. This is the highest-variance template on the page, and the failure mode is one face borrowing features from the other.

59. Couple Cinematic Portrait

Job: restyle. Input: one photo containing both people. Crop: 16:9.

Use the uploaded photo of two people as the identity reference. Preserve both faces, expressions, relative positions, and clothing. Restyle the frame as a cinematic film still with anamorphic widescreen framing, a teal and amber grade, shallow depth of field, and soft practical lighting in the background. Change only the lighting, background, and color grade. Do not add or remove people. Output 16:9, photorealistic.

Customize: the background setting. The do-not-add-or-remove clause stays, because two-person edits drift more than solo portraits.

60. Family Group Portrait

Job: restyle. Input: one family photo. Crop: 4:5.

Use the uploaded family photo as the identity reference for every person in frame. Preserve each face, apparent age, hairstyle, skin tone, clothing, and position. Restyle the setting to SETTING_VALUE with soft even daylight across all faces, a matched color grade, and natural group spacing. Change only the background, lighting, and grade. Do not add, remove, or reposition anyone. Output 4:5.

Customize: the setting. Check every child’s face against the source at full zoom, because a small face in frame is the hardest part of the edit to eyeball.

61. Festival Traditional Wear Portrait

Job: restyle. Input: one portrait. Crop: 4:5.

Use the uploaded portrait as the identity reference. Preserve the facial structure, skin tone, expression, and apparent age. Change the wardrobe to GARMENT_VALUE in COLOR_PALETTE_VALUE with realistic fabric weight, drape, and embroidery detail, and place the subject in FESTIVAL_SETTING_VALUE with warm practical lighting. Change only the wardrobe, setting, lighting, and grade. Output 4:5, photorealistic, with no invented religious symbols or brand marks.

Customize: the garment, palette, and setting. Name the garment specifically rather than by occasion, or the model picks its own interpretation of the occasion.

62. Beach Golden Hour Portrait

Job: restyle. Input: one portrait. Crop: 4:5.

Use the uploaded portrait as the identity reference. Preserve the face, hairstyle, skin tone, and body proportions. Place the subject on an open beach at golden hour with warm backlight, soft rim light on the hair, gentle wind in the clothing, and distant water in creamy bokeh. Change only the background, lighting, wardrobe, and color grade. Output 4:5 with realistic skin texture and no beauty smoothing.

Customize: the time of day. Backlight is what sells the location, so keep it even when you move the hour.

63. Birthday Studio Portrait

Job: restyle. Input: one portrait. Crop: 4:5.

Use the uploaded portrait as the identity reference. Preserve the face, expression, skin tone, and hairstyle. Restyle the frame as a birthday studio portrait in COLOR_THEME_VALUE with a single spotlight, a deep seamless backdrop, controlled confetti or bokeh in the background, and restrained props. Change only the wardrobe, backdrop, lighting, and grade. Output 4:5 with no rendered text or numbers.

Customize: the color theme. Keep numbers out of the image and add the age as a text overlay later, because rendered digits break more often than letters.

64. Father and Child Portrait

Job: restyle. Input: one photo containing both people. Crop: 4:5.

Use the uploaded photo of the adult and child as the identity reference. Preserve both faces, apparent ages, expressions, relative height, and clothing. Restyle the setting to SETTING_VALUE with warm directional light, a soft background, and a natural candid composition. Change only the background, lighting, and color grade. Do not age either person up or down, and do not add or remove people. Output 4:5.

Customize: the setting. The do-not-age clause matters here more than anywhere, because the model reads a child’s face as adjustable.

Quick Copy: The 12-Point Nano Banana Prompt Checklist

Copy this block. A Gemini AI photo prompt copy paste template that answers all twelve lines survives contact with the model, and one that answers seven does not.

Fields 9 and 10 are the pair that separates a working prompt from a wasted attempt. Naming only what to preserve produces a photo that barely changes; naming only what to change produces a photo of someone else.

The 12 fields that turn a basic Nano Banana prompt into a structured, repeatable image-generation instruction.

How to Use These Gemini AI Photo Prompt Copy Paste Templates

Seven steps. The first three take under two minutes and remove most of the retries.

1. Pick the job, not the style. Decide whether you are generating from nothing, editing an upload, or compositing several references. The three need different prompt shapes, and mixing them is the most common cause of a non-result.
2. Prepare the input. One clear, well-lit, front-facing source image beats a better prompt on a blurry one. Heavily filtered uploads carry their filter into the output.
3. Copy the template that matches the job. Every template is complete after its uppercase variables are replaced. Replace the uppercase variables before sending.
4. Check nothing in uppercase is left. Anything still written in UPPERCASE_VALUE form is an unfilled variable. Leave the preserve and change clauses exactly as written.
5. Score it against the 12-point checklist. Twelve fields answered, ship it. Fewer than ten, fix the gaps before spending an attempt.
6. Run one image, not ten. Judge the first output against the job, not against the style you imagined.
7. Refine in the same thread. Reply to the returned image rather than starting a new chat. The model keeps the context and the correction lands more often.

The seven decisions behind those steps are unpacked in full below the templates, along with the model facts, the scoring matrix, and the cost of retries.

Anyone who has already worked through prompt engineering as a discipline will recognise the shape. What differs for image work is that the Preserve line carries more weight than the descriptive language most text-prompt habits push you toward.

Write Your Own Nano Banana Prompt in Five Lines

Every Gemini AI photo prompt copy paste template on this page collapses into one shape. Learn it and you stop needing a library.

Edit the uploaded photo. Preserve SUBJECT_ELEMENTS_VALUE exactly and keep the original identity. Change only CHANGE_ELEMENTS_VALUE. Add STYLE_VALUE, LIGHTING_VALUE and MOOD_VALUE. Keep skin texture and fine detail natural. Output ASPECT_RATIO_VALUE.

Line one is the job verb, and without it the model writes a description instead of an edit. Line two is the identity lock, and it is the only thing standing between a better photo of you and a good photo of a stranger.

Line three is the permission slip. A prompt that locks the subject but never names what is allowed to move returns your upload almost unchanged.

Line four is the part everyone writes first and it matters least. Line five stops the over-processed plastic finish and prevents the silent fallback to a square crop.

Five-second check before you spend an attempt: photo uploaded, preserve list written, change list written, one style named, aspect ratio in its own sentence at the end.

Step 1: Define the Job Before You Write the Prompt

These seven steps expand the quick version near the top of the page. Fill this one in first, because it takes ninety seconds and it decides which tier you belong in.

The last two matter more than people expect. A job with a same-day deadline and a hard commercial-use requirement is not a free-tier job, and finding that out after eleven attempts is an expensive way to learn it.

Step 2: Separate Must-Preserve From Free-to-Change

Every image edit splits into two lists. Write both, every time.

The reject column is the part most guides leave out. A failed preserve is not something you fix with a follow-up prompt, because the model has no memory of the original value once it has replaced it.

Locking too much is its own failure. A prompt that preserves the subject, the background, and the lighting while asking for a restyle has asked for nothing, and the model returns the upload almost unchanged.

Step 3: Score the Prompt Before You Run It

Score the draft prompt out of 5 on each criterion, multiply by the weight, and total it. This costs fifteen seconds and it is cheaper than an attempt.

Decision thresholds:

The last row is the rule that matters. A prompt scoring 4.4 with an empty Preserve line will still return a different person, because the total cannot compensate for a missing lock.

Step 4: Count the Real Cost of Retries

The cost of image work in the Gemini app is not a price per image. It is a compute budget, and retries are what spend it.

Gemini Apps run on compute-based usage limits that account for prompt complexity, the models used, and chat length, and the Gemini usage limits help page states plainly that “your limit refreshes every 5 hours until you reach your weekly limit”. That page also puts AI Plus at roughly twice the standard allowance, AI Pro at four times, and AI Ultra between five and twenty times, checked July 28, 2026.

Source: Gemini Apps usage limits help page and the Nano Banana 2 rollout note (official documentation, checked 2026-07-28, US consumer subscription tiers).

Two of those lines carry a source worth reading before you commit. Output resolution and the free-versus-paid split come from the Nano Banana 2 rollout note, which also confirms that image editing stays restricted to users 18 or older even though image generation opened to under-18 accounts.

One budgeting rule covers most jobs. Multiply your expected attempts by the number of images in the job, and if that number is above roughly twenty in a single working session, treat the paid tier as a project cost rather than a subscription.

The false economy is the one nobody prices. Three people each burning a free allowance on the same deliverable costs more coordination time than one paid seat costs money.

Step 5: Check Rights, Provenance, and Model Access

Run these before the first attempt on anything client-facing, not after.

Every image the models produce carries an imperceptible SynthID watermark, and Google’s Nano Banana Pro announcement also sets the visible policy: the Gemini sparkle appears on images from free and Google AI Pro tier accounts, and is removed for Google AI Ultra subscribers and inside Google AI Studio.

The provenance tool has a limit that matters for anyone doing verification work. Gemini checks images with SynthID and Content Credentials, but the AI verification help page states that it “can currently only recognize content created by Google AI tools”, inside a daily quota of roughly 10 image checks.

Source: Nano Banana Pro announcement, Gemini AI verification help page, and the Gemini API image generation documentation (official documentation, checked 2026-07-28).

If Gemini reports that no SynthID watermark was detected, the file was not identified as created or edited by Google AI. That does not prove it is a photograph, and it does not rule out another AI system.

Some results are simply inconclusive. The help page names cases where there is not enough detail to watermark, where the edit was likely too small, and where a file has been altered so many times that the watermark is no longer detectable.

Consent is the check with no technical fallback. Google’s Generative AI Prohibited Use Policy sets what the model will refuse, and the model also declines edits it believes involve recognisable public figures, which misfires on ordinary faces often enough to plan around.

Step 6: Run One Image Before You Scale the Prompt

Eight tasks. Run them on a single image and a single prompt before you apply either to a batch.

Task 6 is the one people skip, and it decides whether you have a template or a lucky result. A prompt that produces one good image and one unusable one is not ready for a batch of forty.

Step 7: Decide to Ship, Refine, or Abandon

One rule covers a whole team: never spend a fourth attempt on the same prompt. Three failures on one wording means the wording is wrong, and the fourth attempt is how a five-minute job becomes an afternoon.

What Nano Banana Means in 2026, and How It Maps to Gemini AI Photo Prompts

Everything from here down is the craft layer: which model you are writing for, how to build your own template, and the seven decisions that stop an attempt being wasted.

Nano Banana is the informal name for Google’s Gemini image models. That is why a nano banana prompt and a Gemini AI photo prompt copy paste template are the same instruction under two names, which is why the copy-paste libraries and the model documentation rarely use the same vocabulary.

Four models now carry the name in the API: Nano Banana 2 Lite (gemini-3.1-flash-lite-image), Nano Banana 2 (gemini-3.1-flash-image), Nano Banana Pro (gemini-3-pro-image), and the original Nano Banana (gemini-2.5-flash-image), according to Google’s image generation documentation.

Nano Banana 2, formally Gemini 3.1 Flash Image, arrived on February 26, 2026 and replaces Nano Banana Pro across the Fast, Thinking and Pro models in the app. Output runs at 1K for free accounts and 2K for paid ones, per the Gemini app release note, checked July 28, 2026.

That same announcement documents maintaining character resemblance for up to five characters and the fidelity of up to ten objects in a single workflow. The Gemini API documentation separately lists input-reference limits for Gemini 3.1 Flash Image of up to ten object images and up to four character images.

Treat those as two different ceilings rather than one interchangeable number. The first describes what the model holds consistent inside a workflow, the second describes how many reference files the API accepts.

Nano Banana Pro, formally Gemini 3 Pro Image, was announced November 20, 2025 with 2K and 4K output and consistency across up to 14 images and up to 5 people, according to Google’s Nano Banana Pro announcement.

The practical read for a buyer: the free tier now runs the newer model, so the upgrade question is no longer about model access. It is about resolution, retry headroom, and whether the visible watermark is acceptable on client work.

One naming trap costs people real time. A prompt written against Pro-only capabilities, such as 4K output or a 14-image consistency chain, will silently produce something weaker on a free account rather than returning an error.

Example Filled-In Prompt Brief

A worked case, so the twelve fields are not abstract. The job: a four-person consultancy needs matching headshots for a proposal deck, due the same day, from phone photos the team already has.

Scored against the Step 3 matrix, that brief lands at 4.6, with the only weak line being input quality on one of the four photos.

The last row is the honest lesson. No prompt recovers detail that was never in the source file, and a reshoot on a phone costs less than six more attempts.

Red Flags That Should Stop a Prompt

Ten conditions where the correct move is to stop rather than spend another attempt. No Gemini AI photo prompt copy paste template fixes any of them.

Several are the reason a job belongs in a dedicated AI photo editor instead of a chat prompt.

Rows 4, 5 and 10 sit outside the model entirely. Google’s Prohibited Use Policy governs what the tool refuses, but nothing in the product decides whether you had the right to make the edit in the first place.

Common Mistakes That Waste the Most Retries

The first two are opposites, and people tend to overcorrect from one into the other. A prompt that fails by changing too little is fixed by trimming the preserve list, not by adding adjectives.

Most of these are the same failures that show up across every image model, which is why the structure transfers even when the wording does not. Anyone maintaining a library for more than one tool will recognise the pattern from generative AI work generally.

What to Do After You Have a Working Prompt

Match your outcome to the next step rather than starting over. Working nano banana prompts are assets, and most teams lose theirs by never writing down what made one work.

A second model is the fastest sanity check on a prompt that keeps failing. The same brief run through a different engine tells you whether the wording is weak or the job is simply outside what this model does well, and the ChatGPT photo editing prompts library here uses the same field structure for that comparison.

If the question is the tool rather than the prompt, start with the model’s own capabilities and limits in the Gemini review before you commit a workflow to it.

For teams standardising prompts across writing, design and social at the same time, the same twelve-field logic carries over to ChatGPT prompt templates and to the wider set of AI tools for content creation.

Frequently Asked Questions About Gemini AI Photo Prompts

What is the Nano Banana prompt checklist, and do I need all twelve fields?

The checklist is the twelve fields a prompt has to answer before it stops wasting attempts: job, subject, action, setting, framing, lighting, medium, exact text, preserve, change, aspect ratio and exclusions. You do not need all twelve on a simple generation.

You do need fields 1, 9, 10 and 11 on every edit of an existing photo, because those four are what stop identity drift and a wrong crop.

How do I customize a copy-paste Gemini AI photo prompt?

Replace every uppercase placeholder in the Gemini AI photo prompt copy paste template with your own subject, setting, text, colour or style details, then paste the completed prompt into Gemini. Keep the preserve and change clauses unless you deliberately want to move the edit boundary.

If the result comes back too close to your upload, reply in the same thread asking for the restyle to be visible, rather than starting a new chat.

How do I score a prompt before I run it?

Score each of the eight criteria in the Step 3 matrix from 1 to 5, multiply by the weight, and total it. Above 4.2, run it.

Between 3.0 and 4.1, fix the lowest-scoring field first. Below 3.0, rewrite.

A zero on either the job verb or the preserve list means rewrite regardless of what the total says.

Which parts of an image must I preserve, and which can I let the model change?

Preserve anything a viewer would recognise or verify: a real person’s face, a product’s shape and label, exact brand colour, and any on-image text. Let the model change what you are paying it to change, which is usually background, lighting, wardrobe and colour grade.

Locking the background and the lighting while asking for a restyle is the most common way to get nothing back.

Why does Gemini hand back my photo almost unchanged?

The prompt read as a description rather than an instruction. Open with an action verb, state which elements must visibly differ from the original, and check that your preserve list does not include the things you wanted restyled.

Those three fixes account for most passthrough cases. If it persists after two rewrites, the lock list is fighting the change list and the prompt needs restructuring, not more adjectives.

Are Nano Banana prompts and Gemini AI photo prompts the same thing?

Nano Banana is the informal product name for Google’s Gemini image models, so the same core prompt structure usually transfers. The Gemini app and the Gemini API expose different controls, though.

API users may need to set aspect ratio, image size and output format as parameters rather than relying on prompt wording alone. The image generation documentation lists Nano Banana 2 Lite, Nano Banana 2, Nano Banana Pro and the original Nano Banana as of July 28, 2026, and they do not share the same resolution and reference-image ceilings.

How many photos can I edit per day in the Gemini app?

There is no published fixed daily image count. Gemini Apps use compute-based limits that account for prompt complexity, model and chat length, and the help centre states that the limit refreshes every five hours until you reach a weekly cap, checked July 28, 2026.

Paid tiers raise the ceiling rather than changing the mechanic. Any figure quoting an exact images-per-day number is describing an older system.

Do Nano Banana images carry a watermark?

Every generated image carries the invisible SynthID watermark, regardless of tier. Some Gemini app tiers add a second, visible mark on top of it.

Google documents the visible Gemini sparkle for free and Google AI Pro users, and removes it for Google AI Ultra subscribers and for images generated inside Google AI Studio, checked July 28, 2026. Removing the visible mark does not remove SynthID.

Why is Gemini ignoring my aspect ratio?

Put the ratio in its own sentence at the very end of the prompt and write it numerically, as 9:16 rather than vertical. Supported ratios depend on the active model and on whether you are in the Gemini app or the API, so verify the current list before building a workflow around an unusual crop.

The API documentation lists Gemini 3.1 Flash Image with ratios including 1:1, 4:5, 9:16, 16:9 and 21:9, checked July 28, 2026. An unsupported number is rarely the cause of a wrong crop.

If it still returns a square, ask for the crop as a follow-up turn on the returned image instead.

What do I do after a prompt works?

Save it with three things attached: the input type it expects, the crop it produces, and the score it earned. That turns a lucky result into a template someone else on the team can run.

Then run the identical prompt a second time before you batch it, because a prompt that produces one good image and one unusable one is not ready for a set of forty.

About the author

Macedona

Macedona is the founder and lead reviewer at SaaS CRM Review, where he has published 175+ in-depth reviews, pricing guides, and comparisons of CRM and SaaS tools. Each review is based on hands-on testing or verified documentation, and every article states clearly which method was used. Pricing and features are checked against official vendor sources, with the verification date noted in the article. Macedona follows a published review methodology and editorial policy. SaaS CRM Review earns affiliate commissions from some links, which never influence ratings or rankings. Read the full affiliate disclosure.

Follow the author: