Nano Banana 2 Quick-Start Guide
Nano Banana 2 (official name: Gemini 3.1 Flash) is Google's next-gen AI image generation and intelligent editing model. It combines Pro-level image quality with Flash-level response speed. Use it for text-to-image, image-to-image, local edits, background swaps, old photo restoration, and nearly every image task — often under one minute per result.
| Core dimension | Specs |
|---|---|
| Resolution | 512px to 4K, full range |
| Aspect ratios | 16:9, 9:16, 2:1, 4:1, 8:1 and more (including ultra-wide) |
| Consistency | Up to 5 characters + 14 objects kept across iterations |
| Text rendering | Accurate Chinese and English, with translation and localization |
| Search boost | Real-time web search for facts and reference images |
| Think Level | Minimal (fastest) / High & Dynamic (more precise), adjustable |
| Generation speed | Under 1 minute in fast mode (Pro tasks may take 2–5 minutes) |
Pro remains an option for specialized jobs; for daily retouching, fast iteration, and batch output, choose Nano Banana 2.
🚀 Quick start
Step 1: Find the entry point
Nano Banana 2 is already built into multiple platforms — no extra install:
- Casual users: gemini.google.com or the Gemini App → Tools → Generate image (banana icon)
- Power users: Google AI Studio → select Gemini 3.1 Flash Image
- Developers: Gemini API or Vertex AI, model ID
gemini-3.1-flash-image-preview - Direct link: app.nanobanana2.xin
Step 2: Try a few prompts
| Scenario | Prompt example |
|---|---|
| Portrait retouch | Make the character's face slimmer, optimize lighting for translucent skin, keep natural pores and texture. |
| Background swap | Replace the background with a neon-lit cyberpunk city street at night, keep the subject and lighting consistent. |
| Product shot | Professional product photo on pure white background, studio three-point lighting, soft shadow below. |
| Old photo restore | Restore this damaged vintage photo, remove scratches and tears, colorize naturally while preserving the era feel. |
Step 3: Continuous editing and tips
Nano Banana 2 supports conversational editing — upload an image once, then describe each next change in plain language.
Tips:
- Up to 14 reference images per request for fusion and style refs
- After 4–5 rounds, start a new session to avoid context drift
- Use Minimal Think Level for speed; High & Dynamic for complex multi-element prompts
- For portraits and products, be specific about lighting, materials, and composition
🧭 Ten scenarios · step by step
1. Portrait retouch and smart beauty
Upload a photo → send prompts like "optimize lighting for translucent skin, keep natural pores, soften shadows." Nano Banana 2 understands fine requests such as "smooth skin without plastic look" — ideal for ID photos, avatars, and e-commerce model shots.
2. Old photo restore and colorization
Upload yellowed, torn, or scratched vintage photos with prompts like "repair damage, remove noise, colorize naturally, keep the era feel." The model fills missing areas and converts B&W to natural color while preserving expression and composition.
3. Smart background swap and multi-image fusion
Upload your subject → specify the new background in the prompt, e.g. "replace with a neon cyberpunk street, keep subject lighting consistent." Supports multiple reference images with up to 5 characters + 14 objects kept consistent.
4. Commercial product and e-commerce hero shots
For a sneaker product photo, try "pure white background, studio three-point light, 45° angle, show leather texture and sole detail." Generate multi-angle product sets ready for Amazon, Shopify, or marketplace listings.
5. PPT visuals and infographics
Great for slides, training decks, and research presentations. Describe flows and concepts, e.g. "four-step AI workflow infographic: data collection → training → evaluation → deployment, blue isometric style." Chinese and English titles render accurately.
6. Sketch upgrade and commercial-grade rendering
Upload game/anime line art or hand-drawn sketches → "convert to commercial 3D render with realistic materials and cinematic lighting." Structure is preserved while environment, materials, and atmosphere are added — ideal for indie games and IP concept art.
7. Image marking — precise local edits
When you only need to change one area, mark or describe the region, e.g. "change only the top-left sky to sunset orange, leave everything else." Perfect for logo tweaks, local recolor, or adding/removing a single object.
8. Virtual try-on and 3D figurine generation
Upload garment image + person photo → "naturally dress the model in this denim jacket, keep fit and fabric detail." Also supports chibi figurine or 3D collectibles previews for e-commerce and fan merchandise.
9. Multi-round iteration and image-to-image
Nano Banana 2 natively supports chat-style editing — after the first result, send "change the apple to green and add a shadow" or "swap background to pure white" without re-uploading. Ideal for A/B tests, color variants, and fine tweaks.
10. Real-world knowledge generation
The model connects to live search for current product looks, landmarks, and real-world scenes before generating. Ask for "a 2026 car model" or "Shanghai Bund this season" — outputs stay grounded instead of pure training-data guesswork.
⚙️ Resolution, aspect ratio, and Think Level
| Setting | Options | Best for |
|---|---|---|
| Resolution | 512px, 1080p, 2K, 4K, etc. | 512px for fast iteration and high-concurrency tests |
| Aspect ratio | 16:9, 9:16, 2:1, 1:4, 8:1, etc. | Social, slides, ads, ultra-wide banners, and more |
| Thinking Level | Minimal (default) / High & Dynamic | Minimal for speed; High for complex multi-element prompts |
🧠 Three levels of prompt writing
Nano Banana 2 follows prompts very faithfully — the more specific your instruction, the closer the result. Even when you combine composition, text, style, lighting, and subject in one shot, it keeps the scene coherent — a signature leap over earlier models.
✨ FAQ
| Q | A |
|---|---|
| Watermarks or non-commercial use? | The official free tier has no forced watermark; outputs support commercial use — check your platform's terms. |
| Can I export PSD directly? | Direct PSD export is not supported yet; in Pro mode you can choose "keep layer info" and separate layers in Photoshop or similar tools. |
| Extreme wide or ultra-wide ratios? | Supports 4:1, 8:1 and other ultra-wide formats for banners and panoramas; 512px is for tests and prototypes, not daily use. |
| Quality drops after many iterations? | After 4–5 continuous edits, context may exceed the sweet spot — start a new session and paste your core prompt fresh. |
| How is it different from GPT Image 2.0? | Nano Banana 2 excels at realistic editing, product photography, portrait retouch, background swaps, and natural blending. GPT Image 2.0 is stronger on design-led tasks (posters, UI mockups, typography layouts). They complement each other. |
Follow this guide to start with Nano Banana 2 — Pro quality at Flash speed from your first image.