Nano Banana 2.1 · Gemini 3.1 Flash Image
Google Gemini 3.1 Flash Image,现已入驻 OVOV。生成带清晰多语言文字的 4K 画面,支持从超宽横幅到竖长塔的极端画幅,最多 14 张参考图锁定角色一致性 —— 每帧 7-14 秒返回。
各模型名称与标识均为其各自所有者的商标。OVOV 为独立产品,与任何模型提供方均无隶属、背书或赞助关系。
See Nano Banana 2.1 in action
Real Nano Banana outputs — no cherry-picking, no post-editing.
Generated with Nano Banana 2.1
Generated with Nano Banana 2.1
Generated with Nano Banana 2.1
Generated with Nano Banana 2.1
Generated with Nano Banana 2.1
Generated with Nano Banana 2.1
Generated with Nano Banana 2.1
Generated with Nano Banana 2.1
Generated with Nano Banana 2.1
Why Nano Banana 2.1 is different
Built on Google Gemini 3.1 Flash Image to fix what breaks AI images in production — inconsistent characters, locked aspect ratios, wrong-language text and slow turnaround.
Extreme Aspect Ratios
From 8:1 ultra-wide banners to 1:8 vertical towers. Website heroes, story panels, tall infographics — canvases no other model will render.
Up to 14 Reference Images
Feed as many as 14 reference photos to lock a character, product or style across every generation. Best-in-class consistency for series work.
Gemini World Knowledge
Grounded in Gemini 3.1 — real landmarks, accurate logos, plausible physics. Fewer hallucinated details, more usable first drafts.
Native Multilingual Text
Readable headlines and labels in English, Chinese, Japanese, Korean, Arabic and more — rendered inside the image, not pasted on after.
4K Photo-Realism
Skin pores, fabric weave, lens falloff — 4K outputs that pass as real photography, straight out of the model.
Fast 7-14s Turnaround
Flash-tier speed: most images return in 7 to 14 seconds. Iterate on a concept dozens of times in the span of one coffee.
From prompt to 4K in seconds
Nano Banana 2.1 handles the whole loop — you describe, optionally attach references, it renders.
Describe & set the canvas
Write a natural-language prompt and name your aspect ratio — anything from 8:1 wide to 1:8 tall.
Attach up to 14 references
Optionally drop in reference images to lock a character, product or art style before rendering.
Download 4K, no watermark
Export commercial-grade 4K images with full usage rights — no watermark, no cleanup.
What you can build
Nano Banana 2.1 shines wherever fixed canvases, consistent characters or dense text defeat other models.
Ultra-Wide Banners & Heroes
Website hero images, GitHub social cards, ad leaderboards — 8:1 and 21:9 canvases composed for the format, not cropped into it.
- True 8:1 / 21:9 ultra-wide composition
- Text and focal point placed for the ratio
- 4K export ready for retina displays
Consistent Character Series
Comic panels, brand mascots, product catalogs — lock the same face or object across a whole set with up to 14 references.
- Up to 14 reference images per generation
- Same character across panels and poses
- Product kept on-model for catalog shots
Vertical Stories & Infographics
1:8 and 9:16 tall canvases for Instagram Stories, Pinterest pins and scrolling infographics that stay sharp end to end.
- Extreme 1:8 vertical towers
- Readable multi-section text top to bottom
- Story- and pin-ready dimensions
Grounded Real-World Visuals
Landmarks, product mockups and editorial scenes that respect real-world facts thanks to Gemini world knowledge.
- Accurate landmarks and geography
- Plausible lighting and physics
- Fewer hallucinated details to fix
Prompt writing guide for Nano Banana 2.1
Gemini 3.1 Flash Image reads natural language well — these techniques unlock its edge on aspect ratio, consistency and text.
Best with specific, canvas-aware prompts
Name the extreme aspect ratio up front
Nano Banana 2.1 supports ratios from 8:1 to 1:8. State the target canvas early so the model composes for it instead of cropping.
- •Say "8:1 ultra-wide" or "1:8 vertical tower" explicitly
- •Put the ratio near the start of the prompt
- •Describe where the focal point sits on that canvas
Example
"An 8:1 ultra-wide banner, mountain range on the left, empty sky for a headline on the right"
Attach references to lock consistency
Provide up to 14 reference images to keep a character, product or style identical across generations.
- •Use several angles of the same subject
- •Keep references clean and well-lit
- •Describe what should stay fixed vs. change
Example
"Same character as the references, now sitting in a cafe, warm morning light"
Wrap exact text in quotation marks
Quoted strings are rendered literally; everything else is scene description. This gives crisp, correct text in any language.
- •Text inside quotes is rendered as-is
- •Use straight quotes (") for best results
- •Combine multiple quoted strings for multi-line copy
Example
"A poster reading "未来已来" in bold brush calligraphy over a misty mountain"
Lean on Gemini world knowledge
Name real places, brands or eras and the model draws on Gemini knowledge for accurate detail.
- •Reference specific landmarks or products by name
- •Add the era or region for period accuracy
- •Great for editorial and mockup realism
Example
"A rainy evening at Shibuya Crossing, Tokyo, accurate signage and building layout"
Ask for 4K when detail matters
Request 4K for print, hero images and close-ups. Use 1k or 2k for fast drafts while you iterate.
- •State "4K" for final, print-ready output
- •Drop to 1k/2k while exploring concepts
- •Higher resolution rewards richer descriptions
Example
"4K product hero of a ceramic coffee cup, soft window light, shallow depth of field"
State constraints at the end
Tell the model what to leave out — extra text, watermarks, clutter. Direct negatives work best.
- •Append "Constraints:" at the end of the prompt
- •Forbid extra text, watermarks and logos explicitly
- •Avoid vague phrasing like "try not to"
Example
"Constraints: no extra words, no watermark, clean uncluttered background"