Nano Banana 生成結果
在浏覽器中直接體驗 Google Gemini 2.5 Flash Image:用自然語言描述修改需求、跨場景保持人物面部一致,支援最多融合 10 張圖像。每張生成約 15 秒,僅消耗 6 點數。
Nano Banana 是 Gemini 2.5 Flash Image 的代號,是由 Google DeepMind 開發的多模態圖像生成與編輯模型。Google 於 2025 年 8 月 26 日 發布並在同一天面向開發者開放(Gemini API、Google AI Studio、Vertex AI),在 LMArena 盲测中被评為圖像編輯领域的第一名。2025 年 10 月 2 日 推出了穩定版,支援 10 種畫面比例,在 Gemini 應用中已生成超過 50 億張圖像。
它最受欢迎的能力是局部編輯。你只需上傳照片並描述想要修改的內容(例如“移除左邊的人物”、“給黑白照片上色”),模型就能精准修改該區域,同時讓面容、宠物或產品特徵保持高度一致。Google 随後推出了基於 Gemini 3 Pro 的 Nano Banana Pro 和基於 Gemini 3.1 的 Nano Banana 2,而原版 Nano Banana 是當前性價比最高、速度最快的選项。
在 Flux1 AI 上,你無需 Google 帳號或 API 密钥即可直接使用。無论選擇何種比例或添加多少參考圖,生成單張 1K 圖像均為統一的 6 點數。同賬戶还可無縫切換到其他 40 多種模型。
The four jobs Google built Gemini 2.5 Flash Image for, and how to get each one right here.
There is no brush or mask. You write the change and the model works out where it goes. Google's examples include blurring a background, removing a stain or a whole person, changing a pose and colorizing a black-and-white photo.
Google focused this model on keeping a likeness intact: a friend or a dog should still look like themselves in a new outfit, city or decade. That suits profile photos, a recurring character and product shots from several angles. If features start to drift after many rounds of edits, start again from the original photo instead of the latest result.
Upload several images and describe how they combine: put a product into a lifestyle scene, place yourself and your pet on the same beach, or restyle a room with the wallpaper from another photo. The uploader here takes up to 10 references, but Google notes that this model works best with up to three, so pick the few that matter and refer to them by number.
Apply the color and texture of one image to an object in another: Google demonstrates flower-petal patterns on rain boots and a dress cut from butterfly-wing prints. Gemini's world knowledge means "turn this sketch into a product render" works without a list of style keywords.
A kawaii-style sticker of the image. The design features bold, clean outlines, simple cel-shading, and a vibrant color palette, with a white background.
Replace the background with a beautiful sunny beach scene. The person wears an elegant white flowing dress and holds a refreshing iced lemonade with garnish. Bright summer day atmosphere with ocean waves and warm sunlight.
Four steps from photo to download. Credits are only spent when a run succeeds.
Use Text to Image to create something new from a prompt, or Image to Image to edit or combine photos you already have. Both run on the same model at the same price.
Drop in up to 10 reference images. For a single edit, one clear photo is enough; for a blend, keep to two or three images that each have an obvious role.
Write what should change and what must stay. Choose one of ten ratios, or match the input image, then press Generate. Most results arrive in 10–15 秒.
Each result appears in your history. To keep going, send it back to the uploader as a reference and ask for the next small change, then download the version you like.
Google's own guidance for this model comes down to one rule: describe the scene in sentences instead of listing keywords. These rewrites apply it to common edits.
All five run on Flux1 AI with the same credits. The original is the cheapest Nano Banana for quick 1K edits; the newer ones cost more and go further.
| 模型 | 适用場景 | Max output | References | 典型耗時 | Credits / image |
|---|---|---|---|---|---|
| Nano Banana | Quick photo edits, consistent faces, multi-image blends | 1K | Up to 10 | 10–15 s | 6 |
| Nano Banana 2 | Pro-level detail at Flash speed, 2K and 4K output | 4K | Up to 14 | 10–20 s | 10–21 |
| Nano Banana 2 Lite | Fastest drafts on the newer Gemini 3.1 generation | 1K | Up to 10 | 5–10 s | 6 |
| Nano Banana Pro | Text-heavy designs, infographics, 4K finals | 4K | Up to 14 | 10–15 s | 14–28 |
| GPT Image 2 | Photoreal scenes and short English text | — | Up to 10 | 15–30 s | 12 |
Nano Banana 2 is Gemini 3.1 Flash Image, which Google describes as Pro-level capability at Flash speed. Here it costs 10 credits at 1K and up to 21 at 4K, takes up to 14 references and returns in 10–20 seconds. Stay on the original for everyday 1K edits; move up for print size, cleaner text or more references. If you want the newer generation at 1K, Nano Banana 2 Lite costs 6 credits and is the model Google lists as this one's replacement.
Nano Banana Pro is built on Gemini 3 Pro and aimed at text-heavy designs such as posters and infographics. It costs 14–28 credits and outputs up to 4K. On this site both return in about 10–15 秒, so the choice is cost: the original is the cheaper pick for photo edits, outfit swaps and blends. Explore on Nano Banana, then send the chosen version to Pro or GPT Image 2.5 (12–24 credits) when the final needs fine type or 4K.
One flat price per image. You see the total before you press Generate, and failed runs are refunded automatically.
| Run | What you get | Credits |
|---|---|---|
| Text-to-image | One 1K image from a prompt, in any of the ten ratios | 6 |
| Photo edit | One 1K edit of a single uploaded image | 6 |
| Multi-image blend | Up to 10 references in one run, no extra charge per image | 6 |
| Failed or blocked run | Credits go back to your balance automatically | Refunded |
Basic · 300 credits/mo
≈ 50 Nano Banana images at 1K
Pro · 800 credits/mo
≈ 133 Nano Banana images at 1K
Max · 2000 credits/mo
≈ 333 Nano Banana images at 1K
Credits are shared across every image and video model, so Nano Banana drafts, a few 4K finals on Nano Banana Pro and a short video all come out of one balance.
Honest notes on what Nano Banana does and does not do on Flux1 AI, so you pick the right model the first time.
Every image comes back at 1K, about one megapixel (1024×1024 for a square). For 2K or 4K, use Nano Banana 2 or Nano Banana Pro.
The Gemini app lets you keep editing in one conversation. Here each run is separate and returns one image; carry an edit forward by reusing the result as a reference.
The uploader accepts up to 10 images, but Google says this model performs best with three or fewer. If you need to combine many items, Nano Banana 2 handles more references.
The model's knowledge stops in June 2025 and it cannot search the web. For current prices, logos, events or product details, write the facts into the prompt or upload a reference.
Google lists long-form text, fully reliable character consistency and small factual details as areas it is still improving on this model. Short labels usually work; for posters with paragraphs of copy, use Nano Banana Pro and proofread every word.
Explicit or "spicy" sexual content is not supported, and Google's safety filters can block prompts or uploads that break its policy. A blocked run is refunded.
Not on Flux1 AI: each image costs 6 credits. Credits come with any Flux1 AI plan and work on every model on the site, and failed runs are refunded automatically. Google's Gemini app is a separate service with its own limits, and it now runs newer Nano Banana versions.
Yes. "Nano Banana" is Google's nickname for Gemini 2.5 Flash Image (model code gemini-2.5-flash-image). Later models reuse the name: Nano Banana Pro is Gemini 3 Pro Image and Nano Banana 2 is Gemini 3.1 Flash Image.
Google announced it on August 26, 2025 in the Google 開發者s Blog and the Gemini app, and made it generally available on October 2, 2025. Google Cloud's model lifecycle table lists its retirement on Vertex AI for March 15, 2027, with Gemini 3.1 Flash-Lite Image (Nano Banana 2 Lite) as the replacement.
Nano Banana is the original: 1K, 6 credits, best for quick edits and blends. Nano Banana 2 is the newer Flash model with Pro-level detail, up to 4K, for 10–21 credits. Nano Banana Pro is the top tier for text-heavy layouts and 4K finals at 14–28 credits. See the comparison table above.
Yes. Google offers it to developers through the Gemini API and Vertex AI. The Flux1 AI API also serves Nano Banana and the other Nano Banana models on the same credit balance as this page; parameters and examples are in the API docs.
No. Gemini 2.5 Flash Image outputs about one megapixel at every ratio (1024×1024 at 1:1, 1344×768 at 16:9). If a site offers 4K under the Nano Banana name, check which version it runs. On Flux1 AI the 4K options are separate models: Nano Banana 2 and Nano Banana Pro.
每次最多 10 張 here, at no extra cost. Google's documentation says the model works best with three or fewer inputs, so give each one a clear job and refer to them by number in the prompt.
Every image created or edited with Gemini 2.5 Flash Image carries Google's SynthID, an invisible digital watermark that identifies it as AI-generated or AI-edited. It does not change how the picture looks.
Upload a photo, describe the change, and get a 1K edit back in about 15 seconds for 6 credits.