Nano-banana 是什麼？
Nano-banana 是 Google 最新推出的圖片生成模型 Gemini 2.5 Flash Image Preview 的代號。這是一款全新的多模態影像模型，能結合文字理解與影像生成，大幅提升創作效率與靈活度！
nano-banana 亮點功能
-
文字驅動影像生成與編輯
只要輸入一句文字指令，就能快速完成背景替換、色調調整、角色姿勢變化，甚至加入新元素，操作直覺又高效。 -
多影像融合
能同時理解並處理多張圖片，將它們合併為全新作品。
例如：把角色置入不同場景，或將產品套入全新設計版面。 -
角色與對象一致性
生成的角色或物件能保持外觀一致，特別適合品牌素材設計、漫畫角色繪製、遊戲人物創作等需要長期延續性的應用。
以下是網路上看到的實用範例
轉換風格
轉換風格對於 nano-banana 真的是小菜一碟。
提示詞參考：
請把這個角色轉換成洛伊德機械風格。
訂製專屬模型
喜歡的東西沒有模型化嗎，就讓 nana-banana 來幫你。
提示詞參考：
Create a photorealistic, highly detailed PVC character figure, meticulously crafted based on a photograph. The figure stands proudly on a sleek, round plastic base. Behind it, its collectible box features a high-quality print of the original character art. Adjacent to the box, a computer monitor displays the intricate 3D modeling process in Blender, showcasing the digital sculpting and texturing. The entire figure is rendered in a clear, translucent PVC material, giving it an ethereal, ghost-like quality. The scene is set within a professional design studio with soft, diffused indoor lighting that highlights the figure's form and transparency.
替換造型、髮型
請 nano-banana 幫你換造型也是很簡單的事，而且主體都不會跑掉。
提示詞參考：
Take the person in [Photo of a person]. Please make two subtle changes. First, change her hair color to a light pink. Second, add a pair of modern, black-rimmed glasses to her face. Ensure the changes look natural, with correct lighting, shadows, and reflections. Do not alter her facial features or expression.
證件照生成
上傳照片並下指令，就可以快速生成證件照了，還可以指定髮型、角度等等。
提示詞參考：
根據使用者提供的照片，設計九種完全不同的髮型（日式、韓式的甜美風格），要求輸出白色背景的證件照，穿白色襯衫，並在同一張圖片中展示不同髮型。人物微微側身約 15 度，面帶微笑。
來源：X@hahazwei
模特兒換裝
有了 nano-banana，就可以透過 AI 幫你快速讓指定角色穿著不同衣服了。
提示詞參考：
Take the model from [Model image] and dress her in the complete outfit shown in [Clothing item image #1]. Place her in a bright, outdoor park setting during a sunny day. Ensure the clothes fit her body realistically, with natural folds and shadows. The model's face, hair, and identity must remain consistent.
指定模特兒動作
以往要 AI 指定模特兒動作都是差強人意，現在只要給 nano-banana 指定的動作，就可以快速生成對應動作的圖片了。
提示詞參考：
Using the model from as the subject, and the pose from as the reference, generate a new, photorealistic image. The subject must perfectly replicate the reference pose. Crucially, maintain the subject's identity, facial features, and clothing from her original photo. The background should be a clean, minimalist studio setting with soft lighting.
指定角色與動作生成對應圖片
只要給指定的角色以及指定的動作，就能讓這些角色生成一張你想要的互動圖。
提示詞參考：
Create a dynamic, anime-style action scene. Use the character with black hair from [Character Image 1] and the character with pink hair from [Character Image 2]. Position them according to the stick-figure sketch in [Pose Sketch]. The black-haired character is performing a flying kick, and the pink-haired character is blocking with a punch, creating a bright yellow energy impact. The setting is ancient stone ruins at dusk, with fire and embers in the background. High detail, cinematic lighting.
AR 眼鏡與圖片標示功能
你可以請 nano-banana 幫你識別圖片及加上標注，可以得到像是 AR 眼鏡的效果，或是產品的介紹圖。
提示詞參考：
Analyze the provided [Scene Image]. Identify the 1-3 most prominent landmarks, buildings, or points of interest. For each identified point, draw a neat, glowing blue outline around its silhouette. Next to each outline, overlay a clean, semi-transparent dark UI box. Inside each box, display plausible information such as a generated Name, Type/Location, and a brief, one-sentence description. The overall style should be modern and informative, like an augmented reality (AR) view.
組合元素
你可以試著把各種元素組合成一張圖，感受 nano-banana 的強大。
出處：X@MrDavids1
透過平面設計規劃圖生成 3D 圖
模擬居家擺設
空蕩蕩的房間想知道怎麼擺設比較好嗎？交給 nano-banana 就對了。
提示詞參考：
This is a photo of an empty living room [Empty Room Photo]. Redesign it in a cozy Scandinavian style. Add a light grey fabric sofa, a natural oak wood coffee table, several green potted plants, and a large, soft wool rug. The lighting should be bright and airy, coming from the window.
從 Google 地圖生成真實的照片
給 nano-banana 一張 Google 地圖並標注地點，就可以請他生成當地的照片。
不過當然跟真實的還是有些差距，但如果以生成的圖片來說已經可以說是可以以假亂真了。
提示詞參考：
請以真實照片的形式，忠實地呈現我目前位置向東所看到的景色。希望能像真的從當地拍攝的照片一樣，連建築物等細節也精確呈現。請生成這樣的圖片。
圖片轉換成塑膠模型
把還沒模型化的怨念機體轉換為塑膠模型。
提示詞參考：
請把這張圖變成真實世界的塑膠模型，白色背景，如果有需要適時地增加腳架
更換髮型
來源：X@balconychy
試著替人物換上不同髮型。
提示詞參考：
以九宮格的方式生成這個人不同髮型的頭像
製作人像公仔
提示詞參考：
產生一張擺放於桌面上的動漫風格公仔照片，以日常隨手用手機拍攝的輕鬆休閒視角呈現。公仔模型以附件中人物照片為基礎，精確還原照片中人物的全身姿勢、臉部表情以及服裝造型，確保公仔全身完整呈現。整體設計精緻細膩，髮質與服飾採用自然柔和的漸層色彩與細膩質感，風格偏向日系動漫風，細節豐富，質感真實，觀感精美。
證件照
用 Nano-banana 就不用在出去拍證件照啦，只要上傳一張人物照片，就可以生成高品質的證件照了。
提示詞參考：
將照片人物修改成為白色背景證件照
Google Map座標轉換等距圖像
上傳一張 Google 地圖並標示地點的截圖後
使用以下提示詞：
使用這個Google位置將地標製作為3D等距圖像（僅限建築物），以遊戲的風格創建它 ，並直接覆蓋在原本地圖上
拆解物件
把指定的物件拆解
提示詞參考：
An exploded view of [specific iconic object], disassembled parts floating in the air, detailed, high-resolution, photorealistic, studio lighting, neutral background, product photography, technical drawing elements, knolling layout, rendered 3D objects, floating components, detailed mechanical design.
紙雕風格生成
提示詞：
“A hyper-detailed origami version of [CHARACTER], folded from high-quality colored paper with crisp edges and realistic paper texture. The folds capture the character’s iconic features, outfit, and silhouette in a simplified geometric form. Placed on a clean studio surface with soft shadows. Minimalist collectible art style, photorealistic detail, sharp focus. 1080x1080 square format”
Source: X@TechieBySA
Google Map座標轉換等距圖像
上傳一張 google 地圖的截圖後，使用以下提示詞：
使用這個Google位置將地標製作為3D等距圖像（僅限建築物），以遊戲主題公園的風格創建它 ，並直接覆蓋在原本地圖上
轉黑白線稿
上傳照片並使用以下提示詞：
請把這張照片轉成黑白線稿
把 logo 轉成動物
上傳 logo 圖片後輸入以下提示詞：
Create a mascot for [BRAND NAME] logo following the JSON aesthetic below:
{
"style": "brand-inspired creative mascot generation",
"subject_handling": {
"adapt_to_uploaded_image": true,
"preserve_uploaded_logo_shape_and_text": true,
"extract_dominant_brand_colors": true,
"analyze_logo_style_for_mood": true,
"use_uploaded_image_as_primary_reference": true
},
"mascot_generation": {
"character_type": "open — could be an animal, humanoid, object, robot, fantasy creature, or abstract form",
"design_strategy": "generate a mascot concept that visually complements the uploaded brand based on extracted brand colors, mood, and style",
"style_alignment": "mascot should feel like part of the brand family — matching tone (playful, elegant, techy, minimal, etc.)",
"color_usage": "use brand colors as dominant tones in mascot’s body, clothing, skin, or materials",
"personality_embedding": "include expressive traits aligned with brand tone (e.g. curious, bold, serene, confident)"
},
"composition": {
"include_uploaded_logo_in_final_image": true,
"arrangement": "mascot and logo presented together in a clear and balanced composition",
"environment": "minimal scene or brand-aligned abstract background — not too busy",
"interaction": "mascot may pose near, hold, or gesture to logo — but must not obscure or alter it"
},
"lighting": {
"type": "neutral or soft directional lighting depending on brand tone",
"shadow_behavior": "soft contact shadows to ground elements"
},
"camera": {
"view_angle": "centered frontal or three-quarter depending on layout",
"focus": "both mascot and logo in clear focus"
},
"post_processing": {
"enhance_color harmony between mascot and logo": true,
"disable_style_overrides or artistic distortion": true
},
"image_constraints": {
"transparent_background": false,
"aspect_ratio": [INSERT ASPECT RATIO],
"include_text_if_present_in_logo": true,
"preserve_uploaded_logo_geometry": true,
"prevent_logo_modification": true
},
"notes": "Use the uploaded logo or product image as the central brand reference. Automatically generate a unique mascot that feels like an original creation belonging to the brand — inspired by extracted colors, shapes, and mood. The mascot must not copy other mascots or use predefined templates. The final image should creatively showcase both the logo and its new mascot side-by-side, visually unified but distinct."
}
