Nano Banana“终极玩法指南”上线！从合影到3D建模，AI作图“全能王”诞生

资讯配图

谷歌AI生图被玩出花了！一口气解锁20种玩法。

作者 | 江宇

编辑 | 漠影

智东西9月5日报道，Google最新一代图像生成模型Nano Banana（Gemini 2.5 Flash Image）已于近日上线，现已支持文生图、图文编辑、多图合成、迭代精修以及文字渲染等能力。自此，Nano Banana在海内外社交媒体上迅速刷屏：从X到小红书，用户纷纷晒出改图、拼图、造场景的案例。

前日，Google AI Studio还总结了六条提示词技巧，其核心要义就在于：用完整的场景叙事，而不是零散的关键词。

▲谷歌官方案例图

比如，如果想要生成一张写实感很强的照片，可以把自己当成摄影师来写提示词，重点放在氛围、镜头和细节上。要做贴纸或插画风格，就别忘了强调线条的勾勒、颜色的搭配，以及纯白背景的干净效果。如果希望文字能精准嵌入画面，那就要讲清楚字体、排版层次和整体设计风格。

要做产品棚拍或商业广告图，最关键的就是打光方式、拍摄角度和材质质感；而极简留白的风格，则需要控制画面的留白空间、主体的位置和光线的柔和度。至于分镜或漫画格，更看重的是角色动作和环境的呼应，以及光影营造出来的氛围。

不过，看完这些“官方技巧”，你可能仍觉得意犹未尽。毕竟在社交平台上，已经有人玩出了更多意想不到的花样。今天，智东西就带来一份Nano Banana的20种玩法合集，从人物合影到卡牌设计、从商品植入到3D场景，一并整理。

一、人物篇（共7类）

人物篇共有7类玩法，包含生成双人合影、猫猫自拍、极速变装、Cosplay、改姿势、老照片修复和表情包贴纸。

1、双人合影

Prompt: A photorealistic portrait of the two people from the reference image, standing side by side as friends, one taller and one shorter, smiling warmly. Their facial features and details should remain consistent with the reference photo. They are outdoors in a park during spring blossom season, with blooming flowers in the background. The lighting is soft and natural, creating a cheerful atmosphere. Captured with a 50mm lens, highlighting their joyful expressions.

指令：写实风格的人像画面，来自参考图片中的两位人物，并肩站立成朋友姿态，一高一矮，面带笑容。面部细节需与参考照片保持一致。他们身处春日公园的花海中，背景有盛开的花树。光线柔和自然，营造愉快氛围。使用50mm镜头拍摄，突出他们的喜悦神情。

▲参考图（reference image）

Nano Banana能够直接把扎克伯格和萨姆·阿尔特曼拉到同一张画面里，两人并肩，被“塑造”成了春日踏青的好友。

不仅，Nano Banana把“公园”和“花海”等指令细节复刻得相当到位，还自动给两人加了个搭肩搂腰的姿势，看起来氛围格外“融洽”。细节上，扎克伯格的发型和参考图几乎一模一样，阿尔特曼的发型稍有跑偏，但脸部特征还是非常精准。

2、猫猫自拍

Prompt: A photorealistic portrait originally showing the cat with Batman, but now replace Batman with Elon Musk. Keep all other elements exactly the same as the original setting (including the cat, background, lighting, and atmosphere). Elon Musk must have the same facial features and expression as in his reference image.

指令：写实风格的人像合影，原本设定为小猫与蝙蝠侠的合影，现在改为埃隆·马斯克与小猫合影。其他元素保持完全一致（包括小猫、背景、光线和氛围）。埃隆·马斯克需保持参考图中的面部特征和表情。

▲参考图（reference image）

这张“小猫自拍”的原图在海外已经被玩出了无数版本，有蝙蝠侠，也有超人，角色的一致性都保持得很不错。

智东西这次让小猫换个“搭档”，和马斯克来了一张合影。

在几次抽卡的结果里，有一张保留了马斯克参考图中的“搞怪”表情，效果意外有点喜感。

3、极速变装

Prompt: A photorealistic full-body portrait of Sam Altman, with his outfit swapped to the trench coat from the reference image. He is standing confidently in a runway/neutral studio backdrop, with the same styling and pose as the reference model. The scene is illuminated by soft, diffused spotlight, creating a modern, stylish atmosphere. Captured with a 50mm full-body portrait lens, emphasizing the fabric texture and flowing silhouette of the trench coat. The composition should be vertical, fashion photography format. Ensure the clothing and styling are consistent with the reference runway outfit. Ultra-realistic, with high clarity. Maintain a natural head-to-shoulder ratio, consistent with realistic human proportions. Ensure the head, face, and body remain well-proportioned and harmonious.

指令：写实风格的全身人像，萨姆·阿尔特曼换上参考图中的风衣。他以与参考模特相同的姿态与造型站立在秀场/中性摄影棚背景中。场景采用柔和的聚光灯照明，营造现代、时尚的氛围。画面以50mm全身人像镜头拍摄，突出风衣的布料质感与流畅廓形。构图为竖版时尚摄影格式。需确保服装与造型与参考图的秀场服装保持一致。成像效果超写实，清晰度高。保持自然的头肩比例，符合真实人体的比例感。保证头部、面部与身体协调统一。

这一条提示词是按照Google官方的指令tips专门优化过的。起初在没有强调“头、脸、身体比例统一”时，生成的人像多少会有些别扭。加上这一层约束后，身体比例就自然了许多。

4、Cosplay

Prompt: A photorealistic full-body portrait of Sam Altman cosplaying as Captain America, wearing a detailed blue suit with a white star on the chest, red-and-white stripes, and holding the iconic circular shield. He is standing confidently in front of a dramatic battlefield backdrop, with stormy clouds and rays of sunlight breaking through. The lighting is cinematic and natural, highlighting the realistic textures of the fabric, metal shield, and his facial features. The overall mood is heroic and inspiring.

指令：写实风格的全身人像，萨姆·阿尔特曼Cos成美国队长，身穿细节丰富的蓝色制服，胸前有白色星星与红白条纹，手持标志性的圆形盾牌。他自信地站立在一片战场背景前，天空布满阴云，阳光从云层间穿透。画面光效具有电影感且自然，突出服装面料、金属盾牌以及面部特征的真实质感。整体氛围英勇而鼓舞人心。

Prompt:A photorealistic full-body portrait of Elon Musk cosplaying as Batman, wearing a detailed black armored Batsuit with a flowing cape and the iconic bat symbol on his chest. He stands in front of a dark, rainy Gotham City rooftop backdrop, with neon lights reflecting off wet surfaces. The lighting is dramatic and moody, highlighting the textures of the armor, the cape’s folds, and his facial features. The overall atmosphere is intense, heroic, and cinematic.

指令：写实风格的全身人像，埃隆·马斯克Cos成蝙蝠侠，身穿细节丰富的黑色装甲战衣，披着飘逸的披风，胸口有标志性的蝙蝠徽章。他站立在暗雨夜中的哥谭市屋顶，霓虹灯倒映在湿漉漉的地面上。突出了战衣的质感、披风的褶皱以及面部特征。整体氛围紧张、英勇，极具电影感。

对于这种Cosplay场景，Nano Banana几乎是“信手拈来”。说是AI生成的图，但效果能和几年前的好莱坞大片媲美了。

5、改姿势

Prompt:A stylized full-body illustration of the cartoon pig character from the reference image, performing the pose from the action reference image. The pig’s appearance, colors, and smooth cartoon style must remain exactly the same as in the original reference. Generate three camera perspectives of the same pose:

Front view–full-body, centered composition.

45-degree angle view–showing depth and dimension.

Low-angle shot–emphasizing power and dynamism.

The background, lighting, and atmosphere should remain consistent with the original pig reference image. Only the pose is changed, matching the action reference exactly.

指令：风格化全身插画，让参考图中的卡通小猪角色摆出动作参考图中的姿势。小猪的外观、颜色与平滑的卡通风格需与原参考图保持完全一致。生成该姿势的三个不同视角：正面视角——全身、居中构图；45度斜角视角——展现空间感和立体感；仰角视角——强调动态效果。背景、光线和氛围需与小猪参考图保持一致。唯一改变的是姿势，应与动作参考图完全匹配。

▲参考图（reference image）

智东西给Nano Banana上了点难度。动作参考图的人体结构相对准确，智东西直接让动画电影中的角色浪浪山小猪妖去模仿。结果生成的画面还挺让人惊喜：小猪妖清澈的眼神和整体形象都复刻得很稳。三个角度的生成没有任何问题，甚至连它标志性的猪鬃毛都保留了下来，整幅画也依旧保持了上美厂的原始画风。

6、老照片修复

Prompt: Colorize old black-and-white photos with realistic, historically accurate tones. After natural colorization, retain mild film grain and a soft vignette to preserve the vintage feel. Emulate Kodak/Fuji color negative film aesthetics — warm highlights, stable mid-tones, and restrained shadows. Use moderate-to-low saturation, avoiding neon or trendy filters, so the final image feels authentic, as if it could have been shot in color at that time.

指令：还原并增强一张老旧的黑白照片，修复划痕、折痕和褪色细节。提升画面的清晰度与锐度，同时保留复古质感。对人物肤色、衣物和背景元素进行细致而真实的上色处理，让照片焕发新生，同时维持原始影像的怀旧氛围。

这一回，智东西把Nano Banana攻克的目标放在了英国作家萧伯纳的黑白肖像上。在修复时，为整张照片做了细致的上色。肤色自然、衣物的质感都被处理得比较贴近真实。

7、表情包贴纸

Prompt:A set of cartoon-style stickers of Donald Trump, each showing different emotions: happy, angry, surprised, confused. Bold outlines, simple cel-shading, and a vibrant color palette. The background must be white.

指令：一组卡通风格的唐纳德·特朗普表情贴纸，每张展示不同的情绪：开心、生气、惊讶、困惑。设计需有粗线条轮廓、简单的赛璐璐上色、鲜艳的配色。背景必须为白色。

生成结果出来后，这套贴纸几乎可以直接拿去特朗普大楼“上架”。四种情绪都紧贴特朗普的人设：开心时眉飞色舞，生气时火力全开，四种表情都夸张到位。

如果想要复刻类似风格，可以尝试在提示词里加上“cel-shading（赛璐璐上色）”，加上“cel-shading”就是告诉模型：别给我太写实的阴影，要画成卡通风格的色块效果，就像动画人物那样。

二、产品广告篇（共4类）

产品广告篇共有4类玩法，包含明盒盲盒、商品植入、产品内部图和广告分镜脚本。

8、明盒、盲盒

Prompt 1:Create a photorealistic product mockup featuring a stylized figurine of the character from the reference image, posed according to the action reference image. The figurine should stand on a display base, placed next to its packaging box decorated with original artwork. Set the scene on a white studio surface with bright, even lighting to accentuate textures and branding. Use a 50 mm product portrait lens to capture fine detail and depth. The packaging design must align with the figurine’s theme—vibrant colors, clear typography, and a glossy finish. Ensure visual and identity consistency between the figurine and reference images, including pose, style, lighting, and proportions, resulting in a cohesive commercial-quality product shot.

指令1：写实风格的产品Mockup，展示根据参考图中的角色制作的手办，并摆出与动作参考图中姿势一致的动作。手办置于展示底座上，旁边放置印有原始艺术形象的包装盒。场景设定于白色摄影棚台面，采用明亮均匀的光线以突出材质质感和品牌细节。使用50毫米产品肖像镜头捕捉精细细节与景深。包装设计需与手办主题保持一致，色彩鲜艳、排版清晰、表面有光泽。手办造型需严格对应参考图的角色设定，包括姿势、风格、光线和比例，整体画面统一，表现出专业的商业产品摄影效果。

Prompt 2:Create a 1/7 scale commercial figurine of the character from the reference image, in a photorealistic style and realistic environment. The figurine is placed on a computer desk, standing on a round transparent acrylic base with no text. On the computer screen, display the ZBrush modeling process of the figurine. Next to the computer screen, place a BANDAI-style toy packaging box featuring the original artwork of the character.

指令2：绘制图中角色的1/7比例商业化手办，写实风格，真实环境。手办放置在电脑桌上，使用圆形透明亚克力底座，底座上无任何文字。电脑屏幕里显示的是该手办的ZBrush建模过程。在电脑屏幕旁，放置一只印有原画的、具有BANDAI风格的玩具包装盒。

▲参考图（reference image）

在网络上已经爆火的盲盒款之外，智东西还试了试“明盒款”。生成结果里，玩具的透明包装盒细节相当逼真，手办的质感和比例也十分到位，看上去就像是真实上架的商品照。如果不是亲手跑出来，很容易以为是哪家厂商提前放出的宣传图。

9、商品植入

Prompt: A sleek, photorealistic commercial image featuring the female model from the reference image holding and presenting the figurine from the other reference image. The scene should have the atmosphere of a high-end toy advertisement, with soft, natural lighting and a minimalist, lifestyle-oriented background. Use a 50 mm product portrait lens to capture sharp detail and depth. Ensure the model’s pose, lighting, and styling are consistent with her reference image, and the figurine’s proportions, textures, and colors match its reference as well. The figurine must appear naturally integrated, with accurate shadows, reflections, and scale relative to the model. The overall composition should feel cohesive, aspirational, and suitable for promotional use.

指令：一幅精致写实风格的商业广告图，基于参考图中的女模特姿态与光线，手持并展示另一个参考图中的玩偶。场景营造高端玩具广告风格，背景简洁极简且具生活氛围，光线柔和自然，突出温馨与时尚感。使用50mm产品肖像镜头拍摄，以捕捉细节与景深。女模特的姿势、光线与造型需与其参考图保持一致；玩偶的比例、材质质感与色彩需与其参考图匹配。确保玩偶与模特融合自然，阴影、反光与尺寸比例恰当。画面整体统一协调，具备宣传级质感。

在这组效果里，模特原本手里拿着的是一块蛋糕，被Nano Banana“无缝”替换成了LABUBU玩偶。手部的衔接尤其值得一提，手指弯曲的角度和指尖都处理得很细致，没有出现常见的“手指畸形”。与此同时，玩偶毛茸茸的质感也被完整保留下来，和模特动作自然融合，整张图看上去就像一张真的玩具广告照。

10、产品内部图

Prompt: A hyper-realistic exploded-view photograph of a refrigerator, showing only the user-visible parts such as shelves, drawers, doors, and compartments. Each part is separated and floating apart in a controlled explosion layout, as if suspended in mid-air. The lighting is soft and even, highlighting realistic materials such as glass, plastic, and metal surfaces. All components must remain in accurate scale and perspective. The refrigerator must be fully complete with no missing parts. The background should be clean white or neutral, creating the look of a professional product display.

指令：一幅高度写实的冰箱爆炸视图照片，仅展示用户可见的部分，如搁板、抽屉、门体和储物格。各个部件以“controlled explosion（可控爆炸视图）”的方式分离悬浮，仿佛漂浮在空中。光线柔和均匀，突出玻璃、塑料和金属表面的真实质感。所有部件需保持比例与透视准确。冰箱必须完整呈现，不能缺少任何部件。背景为干净的白色或中性色，呈现专业产品展示的效果。

Nano Banana生成的画面里，冰箱的搁板、抽屉、门体都被整齐地分离开，材质质感清晰，还原得很逼真。

11、广告分镜脚本

Prompt:A four-frame storyboard for a new running shoe commercial in a sports event montage style. The sequence should feel like dynamic highlights from an intense competition, with photorealistic detail. Strong lighting contrasts, motion blur, and heavy trailing effects. Each frame should capture energetic moments of movement and intensity, leading to a final product-focused shot with the shoe and brand logo.

指令：一组四格分镜脚本，用于新跑鞋广告，以体育赛事蒙太奇风格呈现。整体氛围如同激烈比赛的动态集锦，细节写实。画面中光影对比强烈，运动模糊与拖影效果加重。每一格都展现充满能量与张力的动态瞬间，最后结尾于以跑鞋和品牌Logo为核心的产品镜头。

▲参考图（reference image）

每一格画面都准确捕捉了运动瞬间，最后一格也自然过渡到产品展示，跑鞋与品牌Logo完整呈现，画面干净利落。值得一提的是，英文Logo与文字部分也很稳定，没有出现常见的乱码问题。

三、美术视觉篇（共7类）

美术视觉篇共有7类玩法，包含换风格换角度、线稿上色、打光、生成连环画、宝可梦卡片、游戏UI和多元素拼接。

12、换种风格、换个角度

Prompt 1: Recreate the reference image in a bright and cheerful style, with vivid colors, lively atmosphere, and enhanced vibrancy. The overall mood should feel joyful and uplifting, while keeping the original subject recognizable.

指令1：将参考图重新创作成明艳欢快的风格，画面色彩鲜明，氛围活泼，整体更具亮度与生机。在保持原始主体可识别的同时，营造出愉悦、积极的观感。

智东西把参考图中的“黑化版”懒洋洋交给Nano Banana处理，结果立刻“洗白”成了本色的蠢萌路线。

Prompt 2: Recreate the reference image from multiple different camera angles—such as front view, top-down view, low-angle view, and 45-degree side view—while keeping the subject recognizable. Arrange all variations together within a single image on a clean white background, ensuring consistency in style and lighting.

指令2：将参考图以多个不同的拍摄视角重新创作，如正面视角、俯视角、仰视角和45度侧视角，在保持主体可识别的同时展现差异。将所有不同视角的画面排布在同一张白色背景的图中，保证风格与光线一致。

智东西选用了詹姆斯中国行时的一张上海游客照，交给Nano Banana去生成多个角度的版本。

▲参考图（reference image）

结果不仅人物在不同视角下保持了高度一致性，还自然衍生出几种合影姿势，看起来十分写实。细节上，他手里拿着的篮球也会随着视角变化而调整位置和角度，整体效果就像是一组真实拍下的多角度照片。

13、线稿上色

Prompt: Colorize an action line drawing featuring two characters. The action poses and composition must remain exactly the same as in the line art, while the characters’ appearances, proportions, and details must strictly match the reference images. Apply colors in the same style as the references, with balanced tones, clear highlights, and shadows. Keep the line work sharp and visible. Add a dynamic background in the style of Japanese anime battle scenes—fiery, high-energy, and dramatic—with glowing effects, speed lines, or explosive lighting to enhance the intensity of the action. Ensure that the anatomy and body structure of the characters are realistic and proportional.

指令：为一张包含两个人物的动作线稿进行上色。动作姿势与线稿图保持完全一致，人物的外观、比例和细节需严格遵循参考图。上色风格需与参考图统一，色彩均衡，明暗和高光清晰。保证线条清晰可见。背景添加日本动漫热血战斗风格，充满燃感和张力，可以加入光效、速度线或爆炸般的光影效果，以强化动作的激烈氛围。同时要保证人物的人体构造合理，比例自然。

▲参考图（reference image）

网上的部分案例中，Nano Banana的线稿上色效果相当成熟，生成结果都能做到细致填色。智东西还希望Nano Banana不仅能上色，还要能加上角色形象和日漫式战斗氛围。

起初，当输入一张元素较为复杂的线稿时，生成结果在部分元素的处理上出现了偏差。

▲复杂版线稿生成效果

但在换成结构更清晰的线稿后，Nano Banana就顺利完成了任务，角色细节到位，背景里也加上了速度线、光效等典型的热血元素，看起来颇有一股动漫原画的气势。

14、打光

Prompt: A photorealistic portrait illuminated by a strong frontal spotlight. The direct light clearly reveals the subject’s features and textures, while casting a sharp, dramatic shadow onto the wall behind them. Ensure the subject is positioned at a certain distance from the wall so that the shadow falls clearly onto it. The atmosphere should feel intense and theatrical, with the contrast between the brightly lit subject and the deep shadow on the wall enhancing the visual impact.

指令：一幅写实风格的人像作品，由正面的一束强烈聚光灯照亮。正面光清晰展现人物的特征与质感，同时在背后的墙面上投射出戏剧化的影子。要保证人物与墙面之间留有一定距离，使影子能够清晰地落在墙面上。整体氛围应充满张力与舞台氛围，通过明亮的人物与深邃的墙面阴影之间的对比，强化视觉冲击力。

智东西选用了一张迪丽热巴近期“火出圈”的一张红毯出发图，生成结果里，正面光不仅准确地给出了人物的动作，墙面上的投影也极为逼真。

15、连环画

Prompt:A sequence of three comic panels in a colorful superhero style featuring Donald Trump and Captain America in conversation:

(1) Captain America lands dramatically on a rooftop where Trump is waiting. Trump says in a bold dialogue box: “You think you can protect everyone?”

(2) The two face each other, engaging in tense dialogue. Captain America replies: “That’s my duty—and I won’t back down.”

(3) A wide shot shows them both in dynamic poses, their expressions intense. A bold caption box at the bottom reads: “The clash of ideals begins!”

Use vibrant colors, bold outlines, and classic comic book style with clear speech bubbles.

指令：一组三格漫画，采用色彩鲜艳的超级英雄风格，主角为唐纳德·特朗普和美国队长的对话：

（1）美国队长戏剧性地降落在特朗普等待的屋顶上。特朗普在对白框中说：“你以为你能保护所有人吗？”

（2）两人面对面展开紧张的对话。美国队长回应：“那是我的职责，我绝不会退缩。”

（3）远景展现两人充满张力的动态姿势，神情激烈。画面下方的旁白框写着：“理念的碰撞，正式开始！”

画面采用鲜明的色彩、粗犷的线条和经典漫画风格，配以清晰的对白气泡。

在这一组漫画分镜中，Nano Banana虽然准确生成了超级英雄风格的画风、对白框和旁白框，但还是出现了典型的“小翻车”：同一画格里跑出了两个美国队长、两个特朗普。好在整体氛围和细节都在线，看上去已经很接近一张标准的漫画分镜稿。

16、宝可梦卡片

Prompt: A custom Pokémon card generated based on the reference image. The card design should strictly follow the style, layout, and details of the reference, while keeping the overall look consistent with the classic Pokémon TCG style. Ensure the character, stats, and abilities are adapted from the reference image, and the final card feels authentic and collectible.

指令：根据参考图生成一张自定义宝可梦卡牌。卡面设计需严格遵循参考图的风格、排版与细节，同时整体效果保持与经典Pokémon TCG风格一致。角色形象、数值和技能需基于参考图进行改编，最终呈现出真实且具收藏感的卡牌。

Nano Banana直接把扎克伯格生成进了宝可梦卡牌里。最妙的是，它还很“懂人设”，自动给小札分配的技能是“好友请求”和“数据”。画面风格和经典Pokémon TCG完全对齐。

17、游戏UI

Prompt:A game UI mockup generated based on the reference image. The layout, color scheme, and styling should strictly follow the reference, while maintaining the look and feel of a functional game interface. Include core elements such as health bars, inventory slots, and a quest log, all adapted to match the reference’s design language. Ensure the final UI feels polished, cohesive, and visually consistent with the reference style.

指令：根据参考图生成一份游戏UI界面模型。界面布局、配色和风格需严格遵循参考图，同时保证具备完整的游戏交互感。包含生命值条、物品栏和任务日志等核心元素，并根据参考图的设计语言进行改编。最终的UI需精致统一，视觉效果与参考风格保持一致。

生成的效果第一眼看上去，就像是老式那种经典卡牌游戏界面：生命值条、物品栏、任务日志一应俱全，配色也带着一点复古味道。虽然是AI生成的UI mockup，但整体精致统一，有“能直接玩起来”的感觉。智东西后续有尝试生成塞尔达和原神这类大世界探索游戏的UI设计，但Nano Banana无法理解。

18、多元素拼接

Prompt: Compose a full-body photorealistic image of the model from one reference image, wearing all the clothing, shoes, and accessories from the other reference image. Ensure every item is transferred accurately, including proportions, textures, colors, and fine details, and that the entire outfit fits the model’s body and pose naturally. The final result should look seamless and realistic, as if the model is truly wearing the complete set of garments, footwear, and accessories from the clothing reference.

指令：生成一张模特的全身写实合成图，以一张参考图中的模特为主体，并让其穿上另一张参考图中的所有衣服、鞋子和配饰。需完整呈现每一件单品，保持比例、材质、颜色和细节的准确性，并确保整体穿搭自然贴合模特的体态与姿势。最终效果应无缝融合，真实自然，就像模特本就穿着参考图里的完整服装、鞋履和配饰一样。

▲参考图（reference image）

在这次测试中，服饰的每个细节都被完整保留下来：布料的材质和配饰的颜色都精准复现。生成的画面几乎没有违和感，看起来就像模特真的穿上了另一张图里的全套穿搭。效果相当惊艳，让人感觉Nano Banana完全可以胜任“虚拟试衣间”的角色。

四、空间合成篇（共2类）

空间合成篇共有2类玩法，包含3D建筑建模和地标AR图。

19、3D建筑建模

Prompt:A detailed 3D model of the building from the reference image, presented with dashed lines to highlight structural outlines and hidden sections. The model should maintain realistic proportions and architectural accuracy, while the dashed lines clearly illustrate the building’s form and internal layout.

指令：根据参考图生成该建筑的精细3D模型，以虚线方式呈现结构轮廓和隐藏部分。模型需保持真实比例与建筑细节的准确性，同时通过虚线清晰展示建筑的形态与内部布局。

智东西选取了巴塞罗那大教堂作为测试对象。生成结果不仅给出了45度俯视角的效果，还生成了一张完整的立体模型图。

20、地标AR图

Prompt:A photorealistic photo of the landmark from the reference image, with AR overlays highlighting historical facts, structural details, measurements, and visitor information. The overlays should look like an augmented reality interface, blending naturally with the real-world photo while remaining clear and easy to read.

指令：根据参考图生成该地标的写实照片，并添加AR覆盖层，显示历史信息、结构细节、尺寸测量和游客指南。覆盖层需具备AR的效果，与真实照片自然融合，同时保持清晰易读。

▲参考图（reference image）

智东西选择了上海的陆家嘴作为生成对象。结果画面在保持真实摄影风格的同时，叠加了信息层：包括高度、建成年份、功能分区等数据，还有游客可见的观景台提示。整体效果像是手机举起来时的AR导航界面，信息点和建筑自然对应，读起来直观清晰。

就这轮20种玩法来看，Nano Banana在写实合成、产品级渲染和多视角一致性上惊喜不少，偶尔也会在复杂元素与分镜里“翻车”，但整体可玩性与可用度都已足够高。

不过，随着模型继续迭代，未来的惊喜也值得期待：比如更好的人物一致性、更自然的手部和表情生成、更懂语境的分镜连贯性，甚至可能支持更长链路的创作流程，让一次指令就能完成“成品”。

到那时，AI或将真正变成创作者的常用工具。