Qwen-Image-2.1多图合成与局部编辑教程：多人合照、换装与局部重绘（附整合包下载）
Qwen-Image-2.1多图合成与局部编辑教程：多人合照、换装与局部重绘（附整合包下载）
SEO关键词：Qwen-Image-2.1、Qwen-Image-2.1多图合成、多人合照生成、AI换装、局部重绘、全景图生成、图像编辑教程、掩码编辑
文章摘要：本文整理 Qwen-Image-2.1 多图合成与局部编辑的官方完整提示词，包含 360 度全景图、六人合照、五图换装、十图室内布置四个多图输入案例，以及圈选编辑、掩码补绘、涂抹添加三个局部编辑案例，最多支持 10 张图按编号引用，提示词保持原文不变，文首附整合包下载地址，文末附写法要点与常见问题。
大家好 这里是「代码简单说」
欢迎大家关注同名公众号,不定时更新更多实用有趣的教程 也欢迎大家在评论区一起讨论交流!~
这篇文章主要分享一下 Qwen-Image-2.1 多图合成和局部编辑的提示词写法。
这是 Qwen-Image-2.1 图像编辑里最实用的两块能力：一是多图输入，最多可以同时喂 10 张参考图，把不同的人、不同的商品合成到一张新图里；二是局部编辑，用圈选、掩码、涂抹的方式告诉模型只改标记区域。我把官方七个案例的提示词原样整理出来，做合照、虚拟试穿、家居布置、局部修补都能直接套。
一、整合包下载
如果不想自己配置环境，可以直接使用 Qwen-Image-2.1 整合包，支持文生图 + 图片编辑，6GB 及以上显存可用，适合本地直接体验。
| 项目 | 地址 |
|---|---|
| 整合包下载 | Qwen-Image-2.1整合包：支持文生图+图片编辑 |
二、多图合成前的准备
- 图片按顺序上传：多图案例的提示词靠编号引用图片（
<image1>、<image2>……或【图1】【图2】……），上传顺序必须和提示词里的编号一致，这是多图合成能不能对上号的关键。 - 单主体清晰：每张参考图里只放一个主体（一个人、一件家具），主体完整、背景干净，身份和外观保持得最稳。
- 智能改写：官方多图和局部编辑案例全部关闭智能改写，保证编号引用语句原文发送。
- 局部编辑标注：圈选案例直接在原图上画彩色框，掩码案例上传「原图 + 标注图」两张，涂抹案例在图上涂白标记，标注颜色只是给模型看的定位信息。
三、多图合成示例（图像编辑）
示例：全景图（360-degree panorama）
上传一张普通视角照片，扩展生成完整的 360 度等距柱状全景图，主体只出现一次、左右边缘无缝。
输入参考图（按编号顺序，提示词中可引用「第 1–1 张图」）
#1
参考输出
完整提示词（智能改写：关）
Generate a complete 360-degree equirectangular panorama from the input perspective image.
The scene is an open paved urban plaza beside a very tall, slender observation tower with a twisting white lattice structure and a narrow antenna at the top, rising into a partly cloudy blue sky. A woman with long straight dark brown hair is close to the camera, wearing a brown baseball cap with a light-colored embroidered emblem, a black short-sleeved T-shirt, and a dark green shoulder bag. She has a gentle closed-mouth smile and one arm extended toward the camera. Preserve her facial appearance, expression, cap, clothing, bag, pose, and relationship to the tower. The plaza has broad gray stone paving, trimmed hedges, mature green trees, a red-framed low structure, slim light poles, a few distant pedestrians, and partial urban buildings beyond the landscaping. Extend the scene into a coherent full 360-degree plaza with continuous paving, green planted borders, nearby paths and city surroundings behind the camera. Keep the tower present only once, retain its recognizable lattice geometry and its position relative to the woman, and include its full upward extent through the correct spherical projection. Preserve the reference image's natural photographic style, soft bright daylight, cloud patterns, believable skin tones, and consistent shadows.
Use a true equirectangular projection covering the entire 360-degree horizontal and 180-degree vertical field of view, including the sky overhead and the ground below. Continue the environment behind the camera consistently, with seamless left and right edges and a single continuous scene. Keep the woman present only once. Output resolution: 2880 x 1440 pixels, with an exact 2:1 aspect ratio.
示例：多人合照（Group portrait，6 图）
六个不同的人合成一张复古情景剧宣传照：每人以各自参考图为身份基准，站位、服装、发型在提示词里逐人指定。
输入参考图（按编号顺序，提示词中可引用「第 1–6 张图」）
#1
#2
#3
#4
#5
#6
参考输出
完整提示词（智能改写：关）
Using the six people in <image1>, <image2>, <image3>, <image4>, <image5> and <image6> as identity references, generate a brand-new vertical group portrait in the style of a 1980s sitcom promotional photo, with all six posed together in a vintage bar interior. Scene and background: on a warm beige plaster wall, a gilded carved eagle with spread wings is mounted high; below it hangs a horizontal framed panel of four black-and-white vintage photographs showing a ship hull under construction in a shipyard; at the left edge of the frame a red triangular pennant protrudes, bearing white lettering "Boston RED SOX WORLD" [partially occluded]; on the right runs a dark-brown wooden staircase with turned baluster spindles rising vertically; a Tiffany-style stained-glass pyramid pendant lamp hangs from a metal chain at upper center-right; the whole scene is lit with soft, warm indoor studio light and a muted film-like color tone, in an eye-level medium-full group composition. Arrangement and new states: the heavyset man from <image5> stands at the far left of the back row, now wearing a brown herringbone tweed suit jacket over a white shirt with a dark maroon tie, body angled toward the center with one hand resting on the seat back beside the woman in red; the woman from <image1> is at center-left, keeping the hairstyle from <image1> exactly unchanged — the same golden-blonde upswept style with soft volume on top and fringe bangs across the forehead, identical in length, parting, volume and shape, not restyled into loose or shoulder-length waves — wearing a bright red V-neck knit sweater and small earrings, her right hand (screen left) resting on the shoulder of the young man seated in front of her; the man from <image3> stands tallest at center back, now in a light-gray suit jacket, white shirt and red striped tie, smiling faintly; the man from <image2> stands at the far right of the back row, now in a dark navy zip jacket over a light-blue shirt, one hand resting on the shoulder of the man in the light-gray suit jacket; the young man from <image6> sits on the floor at front left, now in a pink-mauve finely striped polo shirt with a pale-yellow collar, one arm resting on his knee with a thin cord bracelet on that wrist and his other hand placed on his own shoulder; the woman from <image4> sits at front right, turned sideways toward the camera with her hands resting naturally in her lap. The six are arranged in two rows, some standing and some seated, with relaxed natural expressions and a warm vintage ensemble mood.
示例：多品穿戴（Outfit styling，5 图）
虚拟试穿的典型玩法：一个模特 + 多件单品图，提示词用【图N】指定每件单品穿在哪儿，姿势和背景保持不变。
输入参考图（按编号顺序，提示词中可引用「第 1–5 张图」）
#1
#2
#3
#4
#5
参考输出
完整提示词（智能改写：关）
让【图1】中的模特换上【图3】中的玛丽珍鞋，拿着【图4】中的手提包，并戴上【图5】中的绒毛帽。将【图2】中的羽绒服敞开穿在外面，露出原有的内搭上衣。保持模特姿势和背景不变。
示例：多物布置（Room furnishing，10 图）
单张空房间图 + 九件家具软装图，合成一张完整布置好的室内效果图，每件家具的位置、朝向、上面摆什么都写得清清楚楚。
输入参考图（按编号顺序，提示词中可引用「第 1–10 张图」）
#1
#2
#3
#4
#5
#6
#7
#8
#9
#10
参考输出
完整提示词（智能改写：关）
Using the empty living room in <image1> as the scene and the furniture and decor in <image2>, <image3>, <image4>, <image5>, <image6>, <image7>, <image8>, <image9>, <image10> as references, generate a brand-new fully furnished interior image. The scene is the living room of <image1>: light wood plank floor, gray cement-textured walls, a dark gray door on the left wall, two narrow black-framed windows on the right wall, a dark gray skirting board along the wall bases, and multiple warm recessed downlights set into the ceiling in straight rows running the depth of the room; eye-level indoor viewpoint, natural perspective, square 1:1 framing, overall warm lighting. Layout: mount the black-framed wine shelving unit from the background of <image5> on the back wall, with several wine bottles and a small dried-flower vase on its shelves; in front of it place the dining table from <image5>, with a woven-leather dining chair behind it and three round stools lined up on the near side, two wine bottles and a wine glass on the tabletop; to the front-right of the dining table place the high-back armchair from <image4>, facing the room center; on the left side of the room place the rocking chair from <image3> with its ottoman in front; lay the rug from <image6> flat in the middle of the room and on it place the multi-tier storage coffee table from <image2>, its raised wooden top holding a dessert plate, a drink can, a vase of white flowers and a phone, its trays and lower shelves holding a camera, game controllers, a handheld console, a watch and snack drinks, with packaging texts “CRISPY BISCUITS”, “Classic Classique”, “纯牛奶”, “CHA GEILI”, “合味道”, “Coca-Cola” and “RIO” visible; along the right wall, occupying the right side of the frame, place the sofa from <image9>: a large curved kidney-shaped sofa in cream bouclé fabric with one continuous rounded backrest wrapping into low rolled ends, a single seamless seat cushion, no armrests and no visible legs, resting on a recessed dark-wood plinth base; on it place the cushion from <image8> plus a light gray pillow; behind the sofa set a low wooden console top along the wall, on which stand the arch-shaped stone table clock and the dried-flower vase from <image7>. In the near-left foreground corner of the frame, standing directly on the wood plank floor, place the potted rubber plant from <image10>: large glossy dark-green oval leaves on upright stems in a matte greige cylindrical ceramic pot with a matching round saucer, its height reaching roughly the seat level of the nearby seating. Only the objects themselves are taken from <image9> and <image10> — the brick-view window, sheer curtain, floor lamp, framed picture, round marble table and beige rug behind the sofa, and the window, fireplace and sofa behind the plant, do not appear. All furniture rests naturally on the floor with soft contact shadows, lit consistently by the room's downlights, with realistic proportions and occlusions.
四、局部编辑示例（图像编辑）
这一组是「指定区域、精准修改」的三种标注方式：圈选、掩码、涂抹。
示例：局部编辑：圈选（Local editing: annotations）
直接在原图上画蓝、红、绿三种颜色的框，提示词里按颜色逐个说明要改什么，最后明确标注线不能出现在结果里。
输入参考图（按编号顺序，提示词中可引用「第 1–1 张图」）
#1
参考输出
完整提示词（智能改写：关）
将蓝色框选区域内佩戴在抬起手腕上的棕色皮表带金属手表整体移除，并以与手臂肤色及明暗过渡一致的皮肤自然延续填补腕部区域；将红色框选区域内男生的蓬松金色头发改为黑色：黑色头发沿原有发丝走向与蓬松质感呈现完整的明暗层次，高光处为带微光泽的深灰色、阴影处为深黑色；将两处绿色框选区域内的衣服——画面左下左臂上的白色短袖袖口与右肩处的白色衣袖部分——换成灰色短袖亚麻睡衣：睡衣面料呈灰色，带有清晰的亚麻织纹与自然褶皱，袖口为宽松的短袖剪裁，边缘贴合无缝痕；蓝色、红色与绿色标注线不得渲染在图像中。
示例：局部编辑：掩码（Local editing: mask）
上传两张图：原图 + 一张画了圆圈标注的图，提示词描述要在标注处补什么内容。
输入参考图（按编号顺序，提示词中可引用「第 1–2 张图」）
#1
#2
参考输出
完整提示词（智能改写：关）
图中画圈标注的地方需要补上一名骑坐姿态的西部牛仔男子。他头戴棕色宽檐牛仔帽，蓄着浓密的络腮胡，脸侧向画面左方；上身穿棕色帆布夹克，里面搭配深蓝色牛仔衬衫，颈间系着浅棕色围巾；下身穿着带流苏的棕色皮质护腿，脚穿皮靴踩进马镫，一只手搭在鞍部附近，整体呈现自然的骑乘状态。
示例：局部编辑：涂抹（Local editing: painted region）
在原图上涂出白色掩码区域，提示词描述要在涂白区域添加的内容。
输入参考图（按编号顺序，提示词中可引用「第 1–1 张图」）
#1
参考输出
完整提示词（智能改写：关）
在图中右侧白色掩码标注的区域添加一名水肺潜水员：潜水员身穿黑色湿衣，佩戴潜水面镜和背负式气瓶，身体略微倾斜地悬停在水中，面向镜头，调节器呼出的一串气泡向水面上升。
五、多图与局部编辑提示词写法要点
- 编号引用必须严格对应上传顺序：英文提示词用
<image1>到<image10>，中文提示词用【图1】到【图5】，两种写法都有效，但编号和上传顺序错一个，内容就会张冠李戴。 - 每张图的角色单独交代「变与不变」：多人合照案例里每个人都是「来自 imageN + 换什么衣服 + 保持什么不变」的三段式描述，发型这类容易变的特征还要额外强调「exactly unchanged」。
- 场景主体只出现一次：全景图案例反复强调「Keep the tower present only once」「Keep the woman present only once」，扩展画面时主体重复是最常见的翻车点。
- 标注只是定位信息：圈选案例结尾写「蓝色、红色与绿色标注线不得渲染在图像中」，多图布置案例写「只取 image9 和 image10 中的物体本身，它们背后的背景不要出现」，都要显式排除参考图里的标注和干扰元素。
- 局部编辑先定位后描述：三种标注方式的提示词结构一致——先说「哪个标注区域」，再说「要添加/移除/替换什么」，最后补上目标的材质、颜色、姿态细节。
六、常见问题
1、多图合成时人物或物品对错了号怎么办？
检查上传顺序和提示词编号是否一一对应，然后确认每张参考图里只有一个主体；关闭智能改写保证引用语句原文发送。
2、参考图里的背景被一起带进结果了怎么办？
官方多物布置案例的做法是显式声明「只取某图中的物体本身，其背后的窗、壁炉、沙发不要出现」，把自己不要的参考图元素列出来排除掉。
3、圈选编辑后标注线还留在图里怎么办？
提示词结尾必须写「标注线不得渲染在图像中」这类排除语句；同时标注线尽量画得细一些，只框住要改的区域。
4、掩码和涂抹有什么区别？
效果上是同一类局部补绘：掩码案例上传「原图 + 画圈标注图」两张，靠第二张图定位；涂抹案例直接在原图上涂白标记，单图输入。按自己客户端支持的方式选即可。
5、一次最多能传几张图？
官方示例最多演示了 10 张图输入（空房间 + 九件家具），日常使用建议控制在 10 张以内，图越多每张的细节保持难度越大。
七、总结
以上就是 Qwen-Image-2.1 多图合成与局部编辑的完整提示词教程，覆盖全景扩展、多人合照、虚拟试穿、室内布置和圈选、掩码、涂抹三种局部编辑。核心思路是编号对应上传顺序、逐个主体交代变与不变、显式排除标注和干扰元素，照这个结构写就能稳定合成自己的多图项目。
本系列其他文章（Qwen-Image-2.1 官方提示词分场景整理）：
- UI 界面篇：AI 生成 App 界面
- 角色立绘篇：透明背景立绘与三视图
- 试卷生成篇：AI 生成数学试卷
- 信息图篇：学术信息图与科普长图
- 商品图篇：电商产品图与广告摄影
- 人像篇：发型表情编辑与老照片修复
- 分镜篇：漫画分镜与广告故事板
浙公网安备 33010602011771号