Nano Banana Prompts: Google Gemini Image Trend Guide

Nano banana prompts are text instructions for Google's Gemini image models, the family users call Nano Banana in the Gemini app and on social feeds. As of August 2026 the trend covers figurine-style portraits, retro mall looks, and fast photo edits inside Gemini. You upload a selfie or write a scene brief, pick Fast, Thinking, or Pro in the image tool, and iterate in plain language. This guide explains what the nickname means, how the three image tiers differ, reusable templates you can slot for new subjects, a step-by-step workflow, mistakes that waste free daily caps, and a soft path to PromptMake /image when you need a reference still turned into Midjourney or FLUX text after you lock a look.

What nano banana prompts are

Nano Banana is not a separate app. It is the public nickname Google adopted for Gemini image generation and editing inside gemini.google.com and the Gemini mobile app. Developers call the same capability through the Gemini API with image-capable model IDs such as gemini-3.1-flash-image. Search traffic for nano banana prompts usually means people want paste-ready lines for viral portrait styles or quick photo transforms, not API documentation.

The name stuck after Gemini 2.5 Flash Image went viral for fast edits in mid-2025. Google kept the banana branding in product copy. By August 2026 most casual users mean Nano Banana 2 (Gemini 3.1 Flash Image), the default fast generator in the app, when they share a trend prompt. Power users still reach for Nano Banana Pro (Gemini 3 Pro Image) when they need stricter instruction follow, sharper in-image text, or higher resolution exports.

Who this fits: creators chasing the current Gemini portrait wave without rewriting from scratch every week, marketers who need a repeatable figurine or retro look for a campaign, and anyone who already uses ChatGPT Image or Midjourney but wants Google's edit-in-place workflow on a phone photo. Who should skip pure trend chasing: teams that need locked brand assets in Midjourney v7 with --sref, regulated work that cannot upload faces to consumer apps, and jobs that require 4K print masters without a Pro tier.

This page is a craft guide for nano banana prompts as reusable templates. It is not a meme dump. Pair it with trending ChatGPT image prompts on this blog when you export the same look to GPT Image or FLUX after you test inside Gemini.

How Gemini image tiers map to Nano Banana names

Google ships several image models under one banana umbrella. Names change in release notes. Treat the table below as a mid-2026 snapshot and verify in the Gemini app model picker before you paste old prompts from a forum thread.

Nano Banana (original): Gemini 2.5 Flash Image. Still useful for cheap, fast edits. Good when you want a quick background swap or a simple stylize pass and you do not need Pro-level text in the frame.

Nano Banana 2: Gemini 3.1 Flash Image. Default for most app users in 2026. Balanced speed and crowd-pleasing quality. Most viral nano banana prompts you see on social in August 2026 target this tier.

Nano Banana Pro: Gemini 3 Pro Image. Slower, higher control. Pick Pro when the prompt lists five constraints, when lettering must stay readable, or when you export at 2K or 4K for a client deck.

Free Gemini accounts carry daily image caps and may apply a visible watermark on outputs as of mid-2026. Paid Gemini tiers raise limits. Plan generations like a budget: one clear prompt plus one edit pass beats five vague retries on the same selfie.

Fast vs Thinking vs Pro in the app

Inside Gemini, open the tools menu and choose Create images. The model menu offers Fast, Thinking, and Pro variants tied to the Nano Banana family. Fast fits first drafts and social-sized portraits. Thinking spends more compute on composition and multi-step scene logic when your prompt is dense. Pro maps to the flagship image model when your account exposes it.

Match the menu to the job. A figurine-style portrait with one subject and a plain backdrop rarely needs Pro on the first pass. A poster with a headline, a product hero, and a strict color palette usually needs Pro or a Thinking pass after a failed Fast run.

Write the tier choice in your notes next to the winning prompt. Teammates who paste your shell into Fast when you tested on Pro will get a different look.

Photo upload vs text-only generation

Nano banana prompts split into two lanes. Text-only: you describe a scene from scratch and Gemini invents pixels. Photo upload: you attach a selfie, product shot, or location still and ask for an edit, stylize, or background change while preserving face or product geometry.

Trend portraits almost always use upload plus edit language: "Turn this photo into a collectible figurine on a display base" or "Restyle this portrait as a 1980s mall studio photo with soft flash." Text-only works for original scenes, posters, and concept frames without a reference face.

Gemini's edit model keeps identity cues from the upload when you ask for a stylize pass. It can drift if you stack three unrelated style names in one line. One medium per prompt holds better.

Anatomy of a strong nano banana prompt

Viral nano banana prompts share a fixed half and a variable half, same as any image trend. The fixed half locks medium, lighting, background, and constraints. The variable half swaps subject, wardrobe, props, and crop. Paste the entire viral string for a new person and leftover nouns from the original fight your brief.

Gemini image models read natural prose. Complete sentences beat comma tag soup for Nano Banana 2. Pro still rewards explicit lists when you need four constraints obeyed at once. Lead with the job type: create, edit, restyle, replace background. Name the subject next. Close with light, medium, background, aspect intent, and hard nos (no extra people, no readable logos, no watermark text).

The subsections below split edit prompts from create prompts and show how to bracket variables so one shell survives the next trend cycle.

Edit prompts: identity, change, preserve

Edit prompts need three clauses. Identity: what must stay (face likeness, hairstyle, glasses, product shape). Change: the new look (figurine material, mall backdrop, watercolor wash). Preserve: what must not drift (single subject, eye line, crop).

Example edit skeleton: "Edit this uploaded photo. Keep the person's face likeness, hairstyle, and expression. Restyle the image as [medium] with [light] and [background]. Single subject only. No extra limbs. No text overlays."

Swap [medium] for polymer clay figurine on a round display base, 1980s mall studio portrait with soft on-camera flash, or soft watercolor illustration. Swap [light] for soft studio key, warm tungsten mall glow, or overcast outdoor fill. Swap [background] for plain gradient, retro photo backdrop paper, or minimal shelf scene.

Create prompts: scene, composition, quality

Create prompts invent the frame. Stack subject, action, place, light, medium, and aspect in that order. "Create an image of a ceramic coffee mug on a wooden table, morning window light from the left, soft shadows, product photography, square composition, high detail, no text."

Trending create prompts on Gemini often mimic editorial or toy photography rather than pure illustration. Name materials (matte plastic, glossy resin, brushed metal) when the trend is figurine or product hero. Name era cues (1980s, 1990s grunge, Y2K mall) when the trend is retro portrait.

Add aspect intent in plain words: square for profile avatars, vertical 4:5 for feed posts, wide 16:9 for YouTube thumbs. Gemini accepts ratio language even when the UI picker also offers presets.

Step-by-step workflow for nano banana prompts

You can go from a phone selfie to a reusable template in one sitting. Budget twenty minutes the first time. Later runs take five once your shell library exists. The workflow below assumes the Gemini app with Create images enabled. API users follow the same prompt text inside generateContent calls with an image-capable model.

Keep a notes row per trend: trend name, fixed shell, variable fill, tier used (Fast, Thinking, Pro), date, and kill date if the look went stale. That row beats a camera roll full of unexplained PNGs.

Step 1: Pick the job and tier

Decide upload edit vs text create. Decide Fast vs Pro before you write. Open Gemini, choose Create images, select the tier, and attach a photo if the trend needs likeness lock.

Write one sentence offline: "Figurine stylize of my selfie, matte resin, display base, soft studio light, plain backdrop." That sentence becomes the seed for a fuller prompt.

Step 2: Draft, generate, and score

Expand the seed into a full nano banana prompt using the skeletons in the previous section. Generate once. Score against your sentence: Did likeness hold? Did the medium read clearly? Did the background stay simple? If one dimension failed, change only that clause in the next pass.

Use conversational edits when the first output is close: "Keep everything but replace the base with a clear acrylic disk" or "Soften the flash; less grain." Gemini's edit thread is the product advantage over one-shot generators.

Step 3: Strip to a reusable shell

After a win, delete subject-specific nouns from the prompt text. Replace them with [subject], [wardrobe], [background], [era]. Save the shell under a look name: "figurine display," "mall studio flash," "soft paint portrait."

Test the shell with two different subjects or two different products. If the look holds, file it. If the second subject collapses, tighten the fixed medium and light clauses, then retest. One good shell beats ten saved viral strings you never reuse.

Reusable nano banana prompt templates

These templates mirror look families that stayed active in Gemini feeds through August 2026. Fill the brackets. Do not treat them as eternal trends. Capture new shells with the workflow above when the feed moves.

Template 1, collectible figurine: "Edit this uploaded photo. Keep face likeness, hairstyle, and expression. Restyle the person as a detailed collectible figurine on a round display base, matte painted resin, subtle mold seams, soft studio light, plain gradient backdrop, single subject, no text, square composition."

Template 2, retro mall portrait: "Edit this uploaded photo. Preserve face likeness. Restyle as a 1980s indoor mall studio portrait, soft on-camera flash, pastel backdrop paper, slight film grain, warm skin tones, chest-up crop, photographic, no logos."

Template 3, 3D avatar character: "Edit this uploaded photo. Keep likeness. Render as a cute 3D animated character with smooth skin shading, clean stylized hair, simple [wardrobe color] outfit, bright soft key light, minimal [background color] backdrop, friendly expression, single subject, no text."

Template 4, editorial poster frame: "Create a poster-style portrait of [subject description], dramatic soft key light with light haze, [color] gradient background, subtle film grain, bold composition, vertical 4:5, photographic, no readable logos, no watermark."

Template 5, product hero stylize: "Edit this uploaded product photo. Keep product shape and label layout. Place on [surface], [light direction] softbox light, clean shadows, commercial product photography, minimal props, no extra objects, square composition."

Template 6, seasonal overlay (generic): "Edit this uploaded photo. Preserve face likeness. Add [season or event] themed background elements with restrained color, soft natural light, festive but not cluttered, single subject, no offensive symbols, no text overlays unless specified: [optional text]."

For each template, run a second fill with different bracket values before you share the shell with a client. Pro tier may be worth the cost when Template 4 needs readable headline text inside the frame.

Common mistakes with nano banana prompts

Mistake 1: Pasting the full viral string without stripping the original subject. Leftover nouns from someone else's selfie leak into your render.

Mistake 2: Stacking three trends in one prompt. Figurine plus mall flash plus watercolor fights itself. One medium per generation.

Mistake 3: Skipping the preserve clause on edits. Without "keep face likeness," stylize passes drift toward a generic face.

Mistake 4: Using Pro for every first draft. Burn Pro caps on exploration. Fast or Thinking covers most trend tests.

Mistake 5: Ignoring daily limits. Five vague retries on the same upload empty the free tier. Edit one clause per round.

Mistake 6: Uploading photos you do not have rights to use. Trend prompts do not override likeness, publicity, or client NDAs. Use your own face, licensed stock, or client-approved assets.

Mistake 7: Expecting Midjourney parameters inside Gemini. --ar and --sref belong in Midjourney. In Gemini, write "vertical 4:5 composition" and describe style in prose. Export dialect separately when you move the look to another model.

Model and tool notes for August 2026

Nano banana prompts target Google's image stack first. Other models can approximate the look with different dialect. Midjourney v7 wants short phrases plus --ar and --style raw for photographic trends. FLUX family builds prefer natural-language materials and light. GPT Image inside ChatGPT accepts conversational edit chains similar to Gemini when you upload a reference still.

Gemini 3.5 Flash and Gemini 3.1 Pro remain the text-side workhorses when you ask Gemini chat to rewrite a viral prompt into a shell without generating pixels yet. Use Flash for quick strip passes. Use Pro when the viral paste is long and contradictory.

When you outgrow consumer caps, the Gemini API exposes the same Nano Banana models with per-token billing. Treat API keys like production secrets. Redact client names before you paste prompts into public tools.

When to stay inside Gemini

Stay in the Gemini app when you iterate on one selfie, when you want conversational edits, or when you need a fast social portrait without learning Midjourney flags. The Create images tool plus upload is the core nano banana workflow.

Gemini also fits patriotic or event-themed edits when your audience expects quick shareable frames and you accept watermark or tier limits on free accounts.

When to export the look elsewhere

Export when you need --sref style lock in Midjourney v7, batch FLUX renders through an API, or SDXL with LoRA for a brand palette. Capture the winning look as a fixed shell, then translate dialect per target.

PromptMake /image helps on the export path: upload a reference still that already matches the trend, pick a goal mode such as Recreate Exactly, Change Style, Adjust Lighting, or Create Variation, and copy Midjourney, FLUX, DALL·E, Stable Diffusion, or Leonardo formatted text. Soft start: https://promptmake.net/image. Guest accounts receive about three image runs per day; free registration raises that to about five. Image quota is separate from /text.

When PromptMake /image fits nano banana work

Gemini wins while you test nano banana prompts on your own photos and refine in chat. You already have upload and edit in one thread. Stay there until the look is stable.

PromptMake /image wins when the look lives in a PNG from Gemini and you need the same composition in Midjourney or FLUX without hand-rewriting parameters. Recreate Exactly drafts language from the still. Change Style applies a new medium clause while holding pose. Adjust Lighting fixes a trend where the medium worked but the key did not. Create Variation builds sibling frames for a moodboard from one approved hero.

Run image-to-prompt after you strip a viral string to a shell, not before. Text templates teach the trend grammar; the reference still calibrates texture and crop when words alone under-specify the look.

FAQ

These answers cover the searches people type after their first nano banana prompt misfires. Topics include what the name means, which Gemini tier to pick, free limits, template reuse, likeness rules, export to other models, and how PromptMake /image fits once you lock a look.

What are nano banana prompts?

Nano banana prompts are plain-language instructions for Google's Gemini image models, the family nicknamed Nano Banana in the Gemini app. They cover text-to-image creation and photo edits such as figurine stylize, retro portrait looks, and product hero frames. Most August 2026 trend posts target Nano Banana 2 (Gemini 3.1 Flash Image). You write or paste the prompt in Create images, optionally upload a photo, pick Fast, Thinking, or Pro, then iterate.

Is Nano Banana the same as Gemini?

Nano Banana is the consumer nickname for Gemini's image generation and editing models, not a separate product. Google uses the banana branding in marketing and the Create images tool. Developers access the same capability through the Gemini API with image-capable model IDs. Chat-side Gemini 3.5 Flash and Gemini 3.1 Pro help rewrite prompts; image tiers carry the Nano Banana names.

Which Nano Banana tier should I use in 2026?

Use Nano Banana 2 (Gemini 3.1 Flash Image) for most trend portraits and fast edits. Use Nano Banana Pro (Gemini 3 Pro Image) when you need stricter instruction follow, readable in-image text, or higher resolution output. Use the original Nano Banana tier (Gemini 2.5 Flash Image) for cheap simple edits if your account still exposes it. Start on Fast in the app; move to Thinking or Pro when a dense prompt fails twice.

How do I write a nano banana prompt that keeps my face?

Upload your photo, then lead with preserve language: keep face likeness, hairstyle, and expression. Follow with one clear stylize clause: figurine material, era, or illustration medium. Add single subject and no extra limbs constraints. Generate once, then edit in chat with small deltas instead of rewriting the entire prompt. Avoid stacking multiple unrelated styles in one pass.

Are nano banana prompts free to use?

Gemini offers free tiers with daily image caps as of mid-2026. Paid Gemini plans raise limits and may remove or reduce watermarking depending on current policy. Exact caps change; check the Gemini app for your account. Treat free tiers as a learning budget: one structured prompt plus one edit beats five blank retries.

Can I reuse a viral nano banana prompt for a different subject?

Reuse the structure, not the full viral paste. Strip subject-specific nouns, bracket variables like [subject] and [wardrobe], and test two fills before you save the shell. Leftover nouns from the original trend post cause the wrong objects and faces to appear. Your workflow notes should store the shell, not the one-off string.

How do I move a Gemini trend look to Midjourney or FLUX?

Lock the look in Gemini first with a reusable shell. Export a reference PNG. Translate dialect: Midjourney v7 wants concise phrases, aspect flags, and --style raw for photographic trends; FLUX wants material-heavy sentences. PromptMake /image can draft those formats from the still with goal modes like Recreate Exactly or Change Style. Soft path: https://promptmake.net/image.