提示词库
Seedream 5.5 提示词，每条附生成的图。
每条提示词都原样摘自模型厂商的指南或创作者的帖子，旁边是它生成的成品。点开卡片看全文。
字节跳动 Seed 在 2026 年 11 月 9 日发布 Seedream 5.5，作为新一代图像创作模型。三项升级：多轮编辑时未改动的区域保持像素级一致，最多可连续改 100 轮；原生输出带全透明和半透明通道的 RGBA 图；光线和肤质更写实，小字更清楚。官方说 Pro 版随后推出。下面先放创作者在 X 上分享的提示词，每条链到原帖，后面是字节在 Seedream 5.5 官方页给出的示例。每条都逐字引用，并附各自生成的图。
来源：ByteDance Seed: Seedream 5.5 · Introducing Seedream 5.5
试用 Seedream 5.5：Higgsfield
这是推广链接，通过它订阅，AgentCraft 会获得佣金，你付的价格照旧。
Seedream 5.5 上手：照着字节官方示例写提示词 →
更新于 2026-11-11 · 共 20 条，最新的在前
其他模型页：GPT Image 2.6 · Nano Banana 2.1 · Seedance 2.5 · 可灵 4.0 · MiniMax H3 · 万相 3.0
女子手持干花站在苏联公交车旁
卷纸艺术舞者配彩虹伞
屋顶上身穿绿色仿毛大衣的下棋男子
地铁扶梯上回眸的女子，35毫米胶片质感
BERRY CLUB 蓝莓角色设定板，中英双语
蓝闪蝶中英双语信息图
云海滑翔船工坊里的熊、小熊猫和水獭
超广角镜头下从篱笆缝里探头的驴
夏夜欧洲小馆里用餐的两位老人
窗光下抱着熟睡婴儿躺在沙发上的父亲
FROM SEED TO SALAD 六步种菜信息图
柠檬黄墙、紫色木门和亚麻布，午后硬光
PRINT PLAY 涂鸦拼贴海报
雨后水洼里倒映的金色建筑
莎士比亚文学报，七篇短文环绕插图
地铁站台上川流的通勤人群
从迷你泳池提起的水珠拼出 seedream
夜晚酒吧瓶罐间的白猫，闪光灯胶片感
大风中低头看手机的女子，整幅画面倾斜
没有匹配的条目，换个短一点的词试试。
值得照着改的 Seedream 5.5 提示词
本页每个分类里点赞最多的一条，全文贴出，并说明它为什么好用。复制一条，换掉主体就能用。
霍格沃茨图书馆烛光中微笑的斯莱特林少女
用参考图锁定角色外形，提示词再逐项列出校服和配色，让服装保持一致。光线来自两个方向，左侧暖色烛光与窗外冷蓝暮光，形成明暗层次。
Use @ image as the exact character design: long pale-blonde hair, grey eyes, Slytherin uniform with grey sweater vest, green-and-silver striped tie and black robe with green lining. High-quality anime illustration, upper-body shot of her sitting at a long wooden table in the Hogwarts library at dusk, turning her head toward the viewer with a sly, confident smile. Her hands are out of frame. Warm candlelight from the left lights her face and hair, cool blue twilight comes through the tall arched window behind her, glowing dust in the air, stacks of old books softly blurred in the background. Clean line art, rich soft lighting, detailed hair strands, no text. 女子手持干花站在苏联公交车旁
提示词按固定顺序组织：先写具体车型和场景，再列出服装清单，然后用取景规则（膝盖以上、偏离中心、平视）和布光描述控制构图。50mm 镜头、胶片颗粒和负面提示词让画面接近抓拍胶片，而不是现代精修大片。
Photorealistic fashion editorial with the atmosphere of a winter morning in the early 1990s, in a Soviet-era residential neighborhood. A beautiful adult brunette woman with delicate facial features stands beside the open folding doors of an aged cream-and-red Soviet LiAZ-677 city bus. She holds a small bouquet of dried wildflowers wrapped in vintage brown kraft paper, gently gripping it with both hands in front of her thighs.
She wears an oversized charcoal wool coat, a soft ivory knitted scarf, a dark burgundy ribbed sweater, straight dark trousers and black leather winter boots. Loose dark brown hair falls naturally around her face, a few strands lifted by the freezing wind. Flushed cheeks, minimal makeup, thoughtful expression, looking slightly past the camera, shoulders gently raised against the cold.
Snow-covered pavement, compacted gray snow beside the bus wheels, frosted windows with condensation, faded red paint, weathered metal. Blurred Soviet apartment blocks and a modest bus shelter in the distance. Fine snowflakes in the air, visible breath. The delicate dried flowers add a subtle warm color accent against the muted winter surroundings.
Three-quarter portrait framed from the knees up, eye-level camera, woman placed off-center, the bus filling one side of the composition. Cold overcast daylight with direct on-camera flash, crisp face, deep background, subtle analog film grain, restrained colors, candid 1990s street photography combined with luxury fashion editorial. Real skin texture, tactile wool, believable hands and bouquet construction, shot on a 50mm lens --ar 3:4 --style raw --s 150 --no modern cars smartphones logos text watermark 卷纸艺术舞者配彩虹伞
用卷纸条和卷纸蝴蝶点明媒介，并以哑光深灰卡纸背景配顶部棚拍光，让画面像真实的纸雕作品。按顺序列出渐变色，使裙摆、伞面和水纹的彩虹过渡保持一致。
3D paper quilling art of a graceful woman dancing in side-profile. She wears an elegant sleeveless v-neck dress made of intricately coiled paper strips in a vivid rainbow gradient—transitioning seamlessly through yellow, orange, red, magenta, deep purple, royal blue, and emerald green.
Her long brown hair flows in the wind behind her as she gracefully holds an open quilled umbrella featuring matching rainbow paper panels overhead. She wears bright red high heels, standing atop quilled paper water ripples reflecting vibrant hues. Surrounding her in mid-air are multiple detailed quilled paper butterflies in shades of bright orange, yellow, pink, and sky blue.
The background is a solid matte dark charcoal grey cardstock, featuring subtle overhead studio directional lighting that casts soft 3D drop shadows underneath the layered paper elements, emphasizing realistic texture and depth. High detail, paper craft sculpture aesthetic, 4k resolution. --ar 2:3 屋顶上身穿绿色仿毛大衣的下棋男子
提示词把相机固定在棋盘高度并使用20毫米广角镜头，让前景的国王棋子显得巨大、棋盘向远处延伸；同时写明机身和曝光参数来锁定摄影质感。结尾关于不磨皮、不做CGI的约束让皮肤和面料纹理保持真实。
I tested all four models directly in
@magnific
Prompt: An ultra-realistic high-fashion editorial photograph of a striking 28-year-old Japanese man with sharp facial features, a defined jawline, slightly messy medium-length black hair and narrow rectangular black sunglasses, sitting confidently on a rooftop terrace while playing chess.
He wears an oversized luxurious emerald-green faux-fur coat over a black silk shirt, tailored black trousers and a subtle silver chain necklace. His expression is calm, confident and slightly intimidating, looking directly toward the camera.
Camera composition: An extreme low-angle shot taken from the very edge of a chessboard, with the camera positioned only a few centimeters above the board's surface. The black-and-white chessboard dominates the entire bottom third of the frame, its squares appearing exceptionally large in the foreground and rapidly shrinking toward the seated man.
The man is centered in the composition, visible from approximately the knees upward, leaning slightly forward. His right arm extends directly toward the camera, with his hand dramatically enlarged by perspective. He is gripping the top of a polished ivory wooden king chess piece between his thumb and index finger, moving it toward an empty square near the camera. The king is dramatically oversized in the immediate foreground due to the wide-angle perspective, while the remaining chess pieces appear progressively smaller toward the background.
His other hand rests naturally near his knee. The chess pieces are physically accurate and positioned naturally on the board, with realistic contact shadows and detailed polished wooden surfaces.
Environment: An open urban rooftop with subtle concrete architectural details, low surrounding rooftops and an enormous vivid cyan-blue sky with a few soft, scattered white clouds. The bright sky occupies the upper half of the composition, creating strong visual contrast against the emerald-green coat.
Photography: Shot on a Sony A1 II with a 20mm wide-angle lens, camera placed at chessboard level, f/8, ISO 100, 1/1000s. Deep environmental perspective, exaggerated foreground scale, realistic optical distortion, crisp textures across the chessboard and subject. Bright natural midday sunlight, accurate skin tones, natural facial asymmetry, individual hair strands, detailed fabric and fur fibers, subtle imperfections and physically realistic shadows.
Authentic luxury fashion campaign photography, bold editorial framing, natural photographic color grading, exceptionally realistic skin texture, no artificial skin smoothing, no CGI appearance, no illustration, no plastic textures, no excessive cinematic effects. 同类提示词，在其他模型上测过
每条在括号里注明成图来自哪个模型。拿到 Seedream 5.5 上跑一遍，对照着看。
透明背景中的洗发水瓶抠图 （在 GPT Image 2.6 上测）
明确的输出约束让抠图更干净：产品居中、无光晕、标签文字原样保留，并且不加背景或阴影。禁止任何风格改动，使产品与原图保持一致。
Extract the product from the input image and isolate it on a fully transparent background.
Output: centered product, crisp silhouette, no halos/fringing.
Preserve product geometry and label legibility exactly.
Add only light polishing. Do not add a solid backdrop, checkerboard, scenery, or shadow.
Do not restyle the product; remove the background and preserve clean alpha transparency. 透明背景的烘焙店标志 （在 GPT Image 2.5 上测）
要求设计一个原创扁平标志，在大小尺寸下都清晰可辨，并明确写出透明背景的要求：干净的透明边缘，不要底色、棋盘格或水印。调用 API 时配合 background=transparent 使用。
Create an original, non-infringing logo for a company called Field & Flour, a local bakery.
The logo should feel warm, simple, and timeless. Use clean, vector-like shapes, a strong silhouette, and balanced negative space.
Favor simplicity over detail so it reads clearly at small and large sizes. Flat design, minimal strokes, no gradients unless essential.
Fully transparent background. Deliver a single centered logo with generous padding, clean alpha edges, and no solid backdrop, scenery, checkerboard, or watermark. 可爱仙人掌贴纸 （在 GPT Image 2 上测）
对手臂、眼睛、腮红和花瓣数量的精确限定让角色结构稳定。粗描边、白色模切边和透明背景等风格要求让成品可直接用于印刷。
Create a cute printable sticker illustration on a transparent background: a kawaii green cactus in a terracotta flower pot, centered and filling most of a square canvas. The cactus has one large oval ribbed central body and exactly 2 curved side arms, one on the left and one on the right, both raised upward. Give it a happy face with exactly 2 glossy black oval eyes, 2 pink circular blush cheeks, and 1 open smiling mouth with a dark interior and pink tongue. Add a red-pink flower on top-left of the cactus with exactly 6 rounded petals and a cluster of small yellow dots in the center. Cover the cactus with many small star-shaped cream spines with orange bases, distributed across the main body and both arms. Place the cactus in an orange terracotta pot with a thick rim, darker orange shading, small speckles, and glossy highlights; show brown soil and several rounded pebbles at the top. Use a polished digital cartoon style with thick dark green outlines, soft gradients, shiny highlights, saturated colors, and a clean die-cut sticker look with a thick white border around the entire cactus and pot. Background should be transparent, optionally previewed by a light gray checkerboard only to indicate transparency. No text, no watermark, no extra objects. 詹姆斯韦伯望远镜百科信息图卡片 （在 GPT Image 2 上测）
用 TOPIC 占位符加一份模块化面板清单（基础档案、分类、优缺点、评分总结等）。收藏手册和自然史图鉴这类风格锚点，配合浅色背景和圆角框，让整个系列版式统一。
Create educational encyclopedia style images with GPT Image 2 on
@itsPolloAI
Prompt: Based on { TOPIC }, create a high-quality vertical “encyclopedia-style educational infographic image.”
This image should NOT look like a regular poster or a simple illustration. Instead, it should feel like a structured knowledge guide that combines:
the feeling of a collectible reference handbook
a modern encyclopedia page
a lifestyle knowledge card
and a highly shareable social-media infographic
The overall style should resemble a premium natural-history guidebook mixed with modern editorial infographic design.
The image should include:
One beautiful and highly detailed main subject image
Several zoomed-in detail sections highlighting important features
Multiple modular information panels with rounded corners
Clear title hierarchy and highlighted key labels
Concise yet information-rich encyclopedia content
Visual scoring systems, quick summaries, or “Top 5” modules
The information sections should automatically adapt to the topic. Select and combine relevant categories such as:
Basic profile
Classification / taxonomy
Physical characteristics
Behavior / ecology / habits
Structure or formation mechanisms
Growth conditions or usage methods
Care, maintenance, or optimization tips
Risks, warnings, and important notes
Suitable users or application scenarios
Pros and cons comparison
Quick rating tags or summary cards
Visual requirements:
Clean light-colored background
Soft and elegant color palette
Gentle shadows
Small refined icons
Rounded information boxes
Organized editorial layout
High information density without feeling crowded
Comfortable reading experience
The final result should feel like a real publishable encyclopedia knowledge card designed for reading, collecting, and creating as part of a consistent series — NOT like a commercial advertisement poster.
The image must strongly emphasize:
“knowledge integration + modular information design + handbook/reference-style presentation.” 濒危孟加拉虎注解信息图海报 （在 GPT Image 2 上测）
提示词要求以写实主体动物为中心,配合注解标注、图标和色块,而不是千篇一律的分区。先让模型查找真实物种资料,版面的信息才密集而准确。
Animal Infographic by GPT Image 2 on ChatGPT app
Prompt: Create a visually rich infographic about an endangered animal. Start by finding one online, research its habitat, diet, and unique traits. Present information through annotated visuals and structured callouts, not generic sections. Style it like a bold graphic illustration: a detailed, photorealistic central animal as the focal point, supported by diagrams, callouts, and concise text elements. Use clean backgrounds and a mix of photorealism with strong graphic elements (shapes, icons, color blocking) in a layered composition. Make it dense, tactile, and professionally authored. 东京夜晚人行道上的模糊闪光抓拍 （在 GPT Image 2 上测）
叠加运动中闪光、手持抖动、方向性模糊、过曝高光和大量噪点等手机随手拍特征，营造未经修饰的抓拍感。负面提示排除影棚光、锐利对焦和真人复刻。
GPT Image 2
{
"prompt": "Ultra-realistic candid smartphone photo, 3:4 vertical. Original non-identifiable young woman inspired only by general proportions/vibe. Night Tokyo sidewalk beside white ceramic tile wall. Mid-walk, half-profile, turning back in surprise, hand raised toward lens, shy restrained smile, continuing forward, imperfect framing, nearly exiting frame. Flash fired during motion. Strong handheld shake, heavy directional blur, smeared facial details, blurred hand and hair, ghosting, streaked tile wall, deep darkness around. Harsh flash highlights, blown exposure, hard shadows. Raw accidental snapshot, intrusive, playful, fleeting memory. Extremely noisy, heavily blurred, authentic phone-camera look.",
"negative_prompt": "identity match, face copy, real person replication, anime, illustration, painting, stylized, studio lighting, beauty lighting, sharp focus, clean face, smooth skin, posed portrait, fashion editorial, cinematic, film still, tripod, professional photography"
} 身穿阿迪达斯风衣的男子在卷帘门前遮嘴 （在 ChatGPT / Gemini 上测）
提示词明确写出服装细节（双袖条纹、胸前三叶草标、带抽绳连帽）、姿势（手遮住口鼻，下方手部带动态模糊）和背景，因此输出中每一项都能对应。强烈的直射闪光打在深色条纹卷帘门上，形成高反差的街拍质感。
A raw, street-style, candid photograph of a young Asian man standing in a medium waist-up shot. He is wearing a white oversized Adidas original windbreaker jacket with iconic black three-stripes down both sleeves and a small trefoil logo on the chest, a high neck with a hoodie and drawstrings, and black pants. His dark, naturally messy and voluminous hair frames his face. He is looking down and away from the camera, bringing his right hand up to cover his lower face (nose and mouth) in a cool, pensive gesture. His left hand, held lower, shows noticeable motion blur, adding realism. The background is a weathered, dark blue-black metal security shutter door with repetitive horizontal ribs, creating an industrial, urban environment. The lighting is harsh direct flash, brightly illuminating the man and casting deep shadows on his face and sharp shadows onto the shutter behind him, making him stand out. The color palette is high-contrast black, white, and deep urban blues/grays. The texture is gritty with film grain. Composition is eye-level, slightly off-center to the left. 35mm photograph, high contrast, moody atmosphere, cinematic urban street photography, 4k, photorealistic, cinematic lighting 橄榄绿西装男子，皮肤质感自然 （在 Nano Banana 2.1 上测）
在角色设定图之外再传入两张纹理参考图，提示词锁定人物身份、姿势、服装和构图，只改变皮肤微纹理和发丝结构。提示词还包含明确的对比度规则，可在不修图的前提下柔化斑驳的脸部，文中提示词针对 Seedream 5.0 Flash 编写。
How it works:
→ Image 1 — character sheet
→ Image 2 — skin texture
→ Image 3 — hair texture (light or dark)
Universal prompt for Seedream 5.0 Flash (works across characters): Keep <<<image_1>>> exactly as it is: same person, same face shape and features, same expression, same hairstyle and hair shape, same hair color, same skin tone, same makeup, same wrinkles, freckles, moles, scars and facial hair, same clothing and accessories, same pose, composition, layout, background and framing.
Change only the surface texture:
- Skin: apply the micro-texture from <<<image_2>>> — fine natural skin lines, pores and soft realistic surface. Do not copy its color or its glossy shine.
- Hair: apply the strand structure from <<<image_3>>> — fine individual strands, soft natural flyaways. Do not copy its color.
Balance the contrast to a natural photographic level: if image 1 looks too harsh or blotchy, soften it; if it looks flat or washed out, add depth. Aim for a well-balanced realistic photo — rich but not crushed shadows, clean highlights, smooth even tonal transitions, natural shading with real volume. No retouching, no beautifying, no plastic look. 这个模型的提示词怎么写
像写一份简短的拍摄说明
字节的 15 条示例都很短：中位数 417 个字符，约 63 个英文词、四句话。其中 9 条把画面称为照片或抓拍，13 条写了质感、颗粒或细节，8 条写了光线。对比一下，我们 Nano Banana 2.1 页上创作者提示词的中位长度是 1,749 个字符。写 Seedream 5.5，几个具体的决定就够用。 来源：ByteDance Seed: Seedream 5.5
写出原字，再写字体和版式
15 条示例里有 6 条用引号给出要画的原字，比如报头、大标题或中文标语，再补上字体风格和位置。字节在自己的中英文测试里报告严重文字错误率 0.4%，轻微错误率 1.7%，所以密集的小字可以放心要求。 来源：Introducing Seedream 5.5
定好横竖，再把数量写清
9 条示例开头就写了竖幅、横幅或 3:4 这样的比例。版面类的提示词用数字列出组成部分：角色设定板要四个变体、五个表情、六个配件，种菜信息图写了六个阶段，用五个箭头连起来。 来源：ByteDance Seed: Seedream 5.5
在提示词里写明透明背景
Seedream 5.5 直接输出 RGBA 图。字节的透明背景示例都用文字写明：每个角色单独出一张透明背景的图，或者每个圆形之外的区域保持透明。写清哪些部分实，哪些部分透。 来源：ByteDance Seed: Seedream 5.5
分轮修改，每轮改一处
字节的演示里，连续 100 轮编辑都拿上一轮的结果做输入，依次换衣服、配饰和桌上的物件，其余部分保持不变。每一轮只点名要改的那一样东西，模型会让未改动的区域保持像素级一致。 来源：Introducing Seedream 5.5
去哪试 Seedream 5.5
国内用即梦。海外可以在 CapCut 旗下的 Dreamina 网页版用，开发者可以走 BytePlus ModelArk 调用，或在 BytePlus 体验中心试。Higgsfield 也上了 Seedream 5.5，本页顶部指向它的链接是推广链接。 来源：Jimeng · Dreamina · BytePlus playground · BytePlus Seedream
关于这个模型的常见问题
Seedream 5.5 是什么？
字节跳动 Seed 的图像创作模型，2026 年 11 月 9 日发布，能生成也能编辑图片。相比上一版有三项升级：多轮编辑更稳，原生输出透明 RGBA 图，画面更写实、文字更清楚。 来源：Introducing Seedream 5.5
和 Seedream 5.0 比，新在哪？
字节在六类生产场景里测试，Seedream 5.5 每一类都比 Seedream 5.0 Pro 得分高，影视和办公两类的可用率翻了一倍多。在两两对比里，图生图任务的写实度得分达到 +55.62%。 来源：Introducing Seedream 5.5
Seedream 5.5 能出透明背景的图吗？
能。它直接生成带 alpha 通道的 RGBA 图，支持全透明和半透明，贴纸、角色和字效出来就能直接放进版面。 来源：Introducing Seedream 5.5
同一张图能连续改几次？
字节演示了对同一张人像连续改 100 轮，每轮都用上一轮的结果，到最后戒指、眼镜框这样的小物件依然改得准。 来源：Introducing Seedream 5.5
Seedream 5.5 和 GPT Image 2.5、Nano Banana 2.1 比怎么样？
按字节自己的专家评测，Seedream 5.5 在文生图和图生图的文字结构上都领先 GPT Image 2.5 Sunburst 和 Nano Banana 2.1，文生图写实度优势明显。这是厂商自己的结果；我们的 GPT Image 2.6 和 Nano Banana 2.1 页可以拿提示词对照着看。 来源：Introducing Seedream 5.5
有 Seedream 5.5 Pro 吗？
字节说 Seedream 5.5 系列的 Pro 版随后推出。 来源：Introducing Seedream 5.5
Seedream 5.5 的提示词该写多长？
字节的官方示例从 183 到 536 个字符不等，中位数 417。最短的那条，水珠拼出一个单词，有效是因为一个清楚的点子撑起了整张图。 来源：ByteDance Seed: Seedream 5.5