Nano Banana Pro 生成結果
Google DeepMind 的 Gemini 3 Pro Image 模型:清晰多語言文字、根據事实構建資訊圖,以及由最多 14 張參考圖引導的 1K、2K 或 4K 高畫質編輯。
Nano Banana Pro 是 Google DeepMind 旗舰級 Gemini 3 Pro Image 的代號。相比注重輕量速度的 Flash 系列,Pro 系列專為高要求商業視覺、概念設計與超清印刷而打造。
它具備行業顶尖的復杂提示词遵循能力,能准確理解多主體互動、專業攝影布光與細腻景深,更能在畫面中清晰渲染復杂的標語與文字,告別傳統 AI 圖像的拼寫错誤與亂碼。
在 Flux1 AI 上直接在浏覽器中運行 Nano Banana Pro,無需 Google 帳號、API 密钥或 Gemini 訂閱。任選 1K、2K 或 4K 解析度,按需添加參考圖,每張只需 14 或 28 點數,與涵盖 GPT Image 2.5 和 Seedream 5.0 Pro 等 40+ 款模型共用統一點數餘額。
What Google built into Gemini 3 Pro Image, and how each capability shows up in the images you make here.
Google calls Pro its best model for correctly rendered, legible text, from a one-line tagline to a long paragraph, in a wide range of fonts, textures and calligraphy. It can also translate the text inside an image: Google's own demo turns English can labels into Korean and leaves the rest of the design untouched. Its docs list 15 best-supported languages, including Arabic, Hindi, Japanese and Chinese.
Because the model plans before it renders, it handles structured layouts: a step-by-step recipe card, a labeled plant care sheet, a four-shot storyboard, handwritten notes turned into a clean diagram. Give it the content (numbers, labels, steps) and say how to arrange it; it lays out facts, it does not check them.
Blend products, outfits, rooms and faces into one scene. Google's documentation breaks the 14-image limit down for Pro: up to 6 object images kept at high fidelity, up to 5 images of people for character consistency, and up to 3 images used purely as style references. Name each image's job in the prompt and Pro keeps faces, logos and products recognizable across the composite.
Edit with plain instructions: turn day into night, move the focus, add harsh chiaroscuro light, change the camera angle or regrade the colors while the subject stays put. You can reframe to a new aspect ratio while the subject stays locked in place, and output at 4K when the image is going to print, packaging or a large display.
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 idea to download. Credits are only spent on runs that succeed.
In the panel above, brief it like a designer: subject, setting, layout, lighting and any exact text in quotation marks. Full sentences beat keyword lists.
Upload up to 14 images to edit or combine: a product shot, a face, a logo, a palette. References do not change the price, so add every one the job actually needs.
1K and 2K both cost 14 credits, so 2K is the sensible default. Pick 4K (28 credits) for print or large screens, then one of ten aspect ratios, or keep the ratio of your input image.
The image arrives in 10–15 秒 and is saved to your history. Proofread every word, then reuse the result as a reference and ask only for the change.
Pro follows detailed direction well, so most weak results come from prompts that leave the important decisions to the model.
All five run on Flux1 AI from the same credit balance. Choose by the job, not by the version number.
| 模型 | 适用場景 | Max output | References | 典型耗時 | Credits / image |
|---|---|---|---|---|---|
| Nano Banana Pro | Dense text, infographics, multi-person composites, realism | 4K | Up to 14 | 10–15 s | 14–28 |
| Nano Banana 2 | Everyday generation and edits at lower cost | 4K | Up to 14 | 10–20 s | 10–21 |
| Nano Banana | Quick, low-cost edits at 1K | 1K | Up to 10 | 10–15 s | 6 |
| GPT Image 2.5 | Photoreal scenes and sharp text up to 4K | 4K | Up to 14 | 30–60 s | 12–24 |
| Seedream 5.0 Pro | Data-heavy infographics, multilingual posters | 2K | Up to 10 | 1.5–3 min | 6–10 |
Google describes Nano Banana 2 (Gemini 3.1 Flash Image) as Pro capabilities at Flash speed; here it costs 10 credits at 1K, 14 at 2K and 21 at 4K, against 14 and 28 for Pro. Pro keeps more people consistent (5 vs 4 in Google's docs), alone accepts dedicated style references, and independent side-by-side tests (fofr.ai's, for example) found its photos more natural and less processed. Draft on Nano Banana 2 or the 6-credit Nano Banana 2 Lite, then finish hero images and text-heavy layouts on Pro.
GPT Image 2 costs 12 credits flat and takes 15–30 seconds; GPT Image 2.5 costs 12–24 credits and goes to 4K, but usually takes 30–60 seconds. Nano Banana Pro returns in 10–15 秒 and is the stronger pick when an image must hold a lot of structured text or keep several real people consistent. For photoreal scenes, run the same prompt on both and keep the better one.
You see the credit total before you press Generate, and failed runs are refunded automatically.
| Run | What you get | Credits |
|---|---|---|
| 1K | Text-to-image or an edit, 1024 × 1024 px at 1:1 | 14 |
| 2K | Same price as 1K at twice the resolution, the best value for most work | 14 |
| 4K | Highest resolution, for print, packaging and large displays | 28 |
| References | Up to 14 images per run at any size, no surcharge | Included |
Basic · 300 credits/mo
≈ 21 images at 1K or 2K, or 10 at 4K
Pro · 800 credits/mo
≈ 57 images at 1K or 2K, or 28 at 4K
Max · 2000 credits/mo
≈ 142 images at 1K or 2K, or 71 at 4K
Credits are shared across every image and video model, so Nano Banana Pro finals, cheaper drafts on other models and a few videos all come out of one balance.
Where this Nano Banana Pro workflow differs from Google's own apps, and what the model still gets wrong.
In the Gemini API, Pro can look up live data such as weather or sports scores before drawing. This generator has no search switch, so the model works only from your prompt and its training. Write current prices, dates and statistics into the prompt yourself.
You receive the finished image, not the model's interim thought images. For a series such as a storyboard, run once per frame and reuse earlier frames as references.
The Gemini app keeps a chat, so you can say "now make it night". Here each run is a fresh request: upload the last result as a reference and describe only the change. There are no mask or brush tools.
Small print, long paragraphs and less common languages can still slip. Google's own tip: settle the exact wording first, then ask for the image. Check names, numbers and spelling before you publish.
Google embeds an invisible SynthID watermark in every image Gemini 3 Pro Image creates or edits, including API output. It does not change the picture, but the Gemini app can identify it as AI-generated.
Explicit, "NSFW" or sexual content is not supported, and Google's safety filters can block prompts or reference images that break its usage policy. A blocked run is refunded automatically.
Not unlimited. At launch Google gave free Gemini app users a limited Nano Banana Pro quota, after which they fall back to a lighter model. On Flux1 AI each image costs 14 credits at 1K or 2K and 28 at 4K, credits come with any plan and work on every model, and failed runs cost nothing.
Nano Banana (Gemini 2.5 Flash Image) is the original fast editor, capped at 1K here. Nano Banana 2 (Gemini 3.1 Flash Image) is Google's faster, cheaper workhorse with 4K output. Nano Banana Pro (Gemini 3 Pro Image) is the premium model for dense text, complex layouts and multi-person scenes. See the comparison table above.
Google DeepMind announced Nano Banana Pro on November 20, 2025, launching it in the Gemini app, Google AI Studio, the Gemini API and Vertex AI, and rolling it out to Google Ads, Workspace (Slides and Vids) and Flow.
Yes: 1K, 2K or 4K (5504 × 3072 pixels at 16:9). 4K costs 28 credits and takes a little longer; for screens and social posts, 2K at 14 credits is usually enough.
Google offers it as gemini-3-pro-image through the Gemini API and Vertex AI. The Flux1 AI API also serves Nano Banana Pro at 2K and 4K, billed in the same credits as the web app; see the API docs for request examples.
每次最多 14 張, at no extra cost. Google's documentation says Pro keeps up to 6 objects at high fidelity, up to 5 people consistent and up to 3 images as style references within that limit, so give each image a clear role in the prompt.
No. Google has not released the weights, so there is no offline version, and "Nano Banana Pro download" apps are not official. You run it online and download your finished images.
No. Explicit or sexual content is not supported on Flux1 AI, and Google's own filters block it in every product that offers the model. If a prompt or reference is blocked, the run fails and its credits are returned automatically.
Write a brief, add your references, and get a 2K image in about 15 seconds for 14 credits.