提示词库

GPT Image 2.5 提示词库：每张图都附作者写的原始提示词

这里的每张图，都是作者连同提示词一起公开发出来的。先看提示词怎么写，再看它出了什么图，然后复制走，或者直接送进 Seadanse 的生成器，用 Sunburst 或 Flare 跑一遍。

打开生成器怎么写提示词

17 条提示词17 张图署名作者 13 位X 17 条，Reddit 0 条每周更新

生成器

不用离开这一页就能跑

点下面任意一条提示词的「用这条」，文字就落进这个框里。可以改词、选 Sunburst 或 Flare、选比例和分辨率，然后直接在这里出图。

0/10000
按提示词需要什么筛选
按检查点筛选
跳转到Showcase 4Photoreal & Portrait 4Infographic & Diagram 1Product & Ads 1Design & Poster 3Illustration & Anime 2Editing & Restyle 217 / 17

Showcase4

没有落进其他分类的那些——大家日常在做的东西。

A four-view character sheet on one seamless backdropGPT Image 2.5

作者 Abdul Shakoor8119 浏览原帖

A four-view character sheet on one seamless backdrop

One woman photographed front, three-quarter, profile and rear across four panels, with wardrobe and lighting pinned so the four read as one shoot.

  • 只要提示词
  • Character sheet
  • 4 panels
  • Consistency lock
提示词757 个字符
Multi-panel studio photographs of the same real woman across four panels: front-facing, three-quarter turn, full profile, and a rear view, all on a seamless neutral studio backdrop with soft even lighting. She has long straight golden-brown hair with warm honey-blonde balayage, loosely swept back with soft face-framing pieces, an oval face, warm brown eyes, softly tanned olive-warm skin, full lips, small stud earrings. She wears a rust-terracotta linen strapless wrap top tied at the back, matching wide-leg rust-terracotta linen trousers, and a woven raffia belt cinched at the waist. Same lighting, same skin texture, same wardrobe in every panel. Natural film-still photograph quality, visible skin pores and texture, no retouching, no beauty filter.

为什么有效

  • The four angles are named up front, so the model lays out panels rather than guessing at a collage.
  • Identity is written as a list a stylist could work from — hair colour and parting, face shape, eye colour, skin tone, earrings — then wardrobe down to the belt material.
  • "Same lighting, same skin texture, same wardrobe in every panel" is the line doing the consistency work, and the closing "no retouching, no beauty filter" is what keeps the skin from going plastic.
Your own photo, restyled as a 1985 snapshotGPT Image 2.5

作者 Mira7679 浏览原帖

Your own photo, restyled as a 1985 snapshot

An edit that keeps the face exactly and changes everything around it, with the era carried by film defects rather than by the word "vintage".

  • 需要参考图
  • Photo edit
  • Identity lock
  • Analog defects

需要 A photo of yourself

提示词568 个字符
Using my uploaded photo, show me what I would have looked like around 1985. Preserve my identity, facial features, skin tone, age, and recognizable appearance. Reimagine my hair, clothing, accessories, and surroundings with bold, unmistakably mid-1980s styling—expressive silhouettes, statement accessories, layered details, distinctive colors, and textures. Make it feel like a genuine 1985 photograph with analog grain, faded color, direct flash, and subtle softness. Add a period-accurate 1980s red-orange date stamp in the lower corner. No modern objects or text.

为什么有效

  • It splits the job in two explicitly: preserve identity, features, skin tone and age; reimagine hair, clothing, accessories and surroundings. The model is told what not to touch before it is told what to change.
  • The period is specified as photographic artefacts — analog grain, faded colour, direct flash, subtle softness — which is what dates a picture more reliably than the clothes do.
  • The date stamp is given a colour and a corner, and "no modern objects or text" closes the usual leak.
A whole website homepage rendered as one long screenshotGPT Image 2.5

作者 leolee725 浏览原帖

A whole website homepage rendered as one long screenshot

A full Chinese design-studio homepage, nav to footer, specified section by section with each one given a share of the height.

  • 只要提示词
  • Full page mock
  • Percentage layout
  • Chinese typography
提示词1,316 个字符
请使用网页版当前最新的 GPT Image 2.5 生图能力直接生成图片，不要仅返回提示词。
生成一张完整的创意设计公司中文官网视觉设计图，正面平视、无透视的桌面网页长截图，9:16 竖幅，尽可能高分辨率，完整显示从导航到页脚。白底，钴蓝、明黄、珊瑚红和浅天蓝配色，细窄留白、整齐网格、直角卡片、大胆斜切几何拼贴。整体年轻、明快、有设计实验感。
顶部为纤细白色导航栏，左侧原创中文粗体字标“见造”，小标语“以创意，让世界更有趣”。右侧导航依次为“关于我们、创意服务、精选作品、加入我们、最新动态”，最右是蓝色“联系我们 →”按钮。
首屏占全图约 27%，左侧以非常醒目的钴蓝粗黑体排两行“让有趣，”“改变世界。”，左下小字“以自由的想象，连接人与品牌，让每一个好点子，成为真实的改变。”右侧是一名明确成年的 20 岁中国女性，黑色齐肩发被微风吹动，自然淡妆，真实皮肤，微仰头望向右上方。穿钴蓝棉麻立领短袖上衣，简洁斜襟与少量同色盘扣，合身但不紧绷。人物与蓝色立方体、黄色球体、珊瑚红圆环、白色几何台座组合，背景浅蓝天空，明亮自然日光，材质和投影可信。
首屏下方白色理念横栏占约 8%，左侧小字“我们的理念”，中间黑色粗标题“用创意，为社会留出想象。”，右侧正文“我们相信，好的设计不止解决问题，也让人与世界产生新的连接。”及“了解见造 →”。
服务区占约 16%，左侧窄栏标题“创意服务”、简短介绍和“查看全部服务 →”；右侧三张等宽卡片：蓝底卡片以手持灯泡摄影配“品牌创意”，黄底卡片以彩色几何积木配“传播设计”，珊瑚红底卡片以打开的笔记本电脑配“数字体验”。每张卡片下方有两行简短中文介绍和右箭头。
作品区占约 17%，左侧标题“精选作品”和“查看全部作品 →”，右侧四列平面作品缩略图，分别为有彩色颜料点缀的 20 岁中国女性肖像海报、城市几何装置海报、蓝白红编辑设计、山野艺术装置海报。四个项目名称依次为“自在生长”“城市游乐场”“灵感发生地”“山野之间”，缩略图中的可见文字也全部使用中文。
招聘区占约 18%，整幅钴蓝横栏，左侧白色大标题“和有趣的人，一起创造新风景。”，正文“让热爱成为日常，让灵感自由生长。”及白底按钮“查看开放职位 →”。右侧三名明确成年的 20 岁中国女性坐在黄色与珊瑚红几何台阶上自然交流，分别穿蓝色立领上衣、米白棉麻衬衫、浅黄色简约针织上衣，搭配宽松长裤与干净平底鞋，黑色长发或齐肩短发，自然淡妆，青春、自信、亲切，非古装、非职业套装、无夸张饰品。
下方约 6% 为白底“最新动态”栏，三行消息为“2026.09.09 见造全新品牌网站正式上线”“2026.09.08 城市游乐场创意计划发布”“2026.09.01 秋季创意伙伴招募开启”。最后约 8% 为简洁页脚，中文“见造”、导航和“© 2026 见造创意。保留所有权利。”，留出舒适下边距。所有界面文案为可读、准确的简体中文，标题与正文层级鲜明。无日文、无乱码、无第三方标志、无水印、无浏览器边框、无设备外壳。输出一张完整网页设计图。
More Prompt ➡️ http://genvizu.com.

为什么有效

  • Every band is allocated a percentage of the image — hero 27%, philosophy 8%, services 16% — so the layout is designed rather than discovered.
  • All the copy is written out in the language it should render in, headline and body and button labels alike, which is what makes a page mock legible rather than lorem.
  • It opens by telling the assistant to generate the picture rather than return a prompt, a reminder that this one was written for a chat window.
A country whose border is drawn from its own landmarksGPT Image 2.5

作者 simeon-sanai605 浏览原帖

A country whose border is drawn from its own landmarks

One idea, stated once and precisely: arrange famous buildings so their collective silhouette traces the national outline.

  • 只要提示词
  • Single strong idea
  • 3D diorama
  • Short prompt
提示词779 个字符
Create a unique 3D miniature [COUNTRY] where the entire national border is formed from its most iconic landmarks and cultural symbols. Arrange famous buildings, monuments, bridges, mountains, temples, towers, traditional architecture, trees, and other recognizable elements so their silhouettes collectively create the accurate outline of the country. Keep the interior mostly clean and minimal, with [CITY] represented by its signature landmark at the correct location. Use layered sculpted forms, tiny architectural details, soft shadows, handcrafted textures, warm ivory background, and elegant studio lighting. Premium collectible travel-diorama aesthetic, playful but sophisticated, highly recognizable. Add small refined typography: “[COUNTRY]” and “[CITY] • [FAMOUS FOR]”.

为什么有效

  • The constraint is geometric and unusual, and it is stated as the first sentence — the model gets the whole concept before any style words arrive.
  • "Keep the interior mostly clean and minimal" is what stops the shape being swallowed by detail, and one city is pinned "at the correct location".
  • At around 700 characters it is the shortest full recipe here, which shows how little length a strong single idea needs.

Photoreal & Portrait4

想让人以为是照片的人像和场景。这类提示词把字数花在镜头、光线和皮肤上，而不是形容词上。

A handwritten essay in Einstein's hand, from four linesGPT Image 2.5

作者 Chetaslua9.2万 浏览原帖

A handwritten essay in Einstein's hand, from four lines

A photo of a pencil-written essay on lined paper, asked for as a phone snapshot rather than as a document scan.

  • 只要提示词
  • Handwriting
  • Phone-photo realism
  • Four sentences
提示词294 个字符
a photorealistic, taken by phone photo of a handwritten essay in pencil, albert einstein handwriting, on an 8.5x11 piece of lined paper, about the history of money in world. make sure there is variance in the writing in a very einstein way. give it a slight coffee stain on the top right corner

为什么有效

  • "Taken by phone photo" does the work a list of camera settings would: it sets the angle, the lighting and the slight imperfection all at once.
  • The handwriting is specified twice — whose, and that it should vary "in a very einstein way" — which is what stops the model rendering a uniform font.
  • The coffee stain is placed ("top right corner"), not just mentioned. Every physical detail in this prompt has a position.
Two women at a Mediterranean archway, shot on a Y2K digicamGPT Image 2.5

作者 Abdul Shakoor7106 浏览原帖

Two women at a Mediterranean archway, shot on a Y2K digicam

A single long paragraph that places two figures at different depths and then spends its second half on one specific camera and its flaws.

  • 只要提示词
  • Two subjects
  • Direct flash
  • Y2K digicam
提示词1,621 个字符
Two women framed by a sun-bleached stone archway — one in the near foreground at frame right, a woman with long straight golden-brown hair loosely pulled back with soft face-framing pieces, warm honey-blonde balayage catching the light, an oval face with warm brown eyes, softly tanned olive-warm skin and full lips, small stud earrings, wearing a rust-terracotta linen wrap top and matching wide-leg linen trousers cinched with a woven raffia belt, hands clasped in front of her, turned back over her shoulder toward camera; a second woman, features unspecified, stands further back in the courtyard in a deep forest-green silk blouse and wide-leg cream trousers, leaning against a whitewashed stone pillar, gazing out toward the sea. Setting: the threshold of a Mediterranean cliffside courtyard, hand-laid terracotta mosaic floor and a weathered stone archway thick with trailing bougainvillea in the foreground, a sun-drenched courtyard beyond with wrought-iron lanterns, potted olive trees and a low stone fountain, whitewashed rooftops and a pale hazy sea filling the horizon, soft late-afternoon light. Shot on an early-2000s point-and-shoot digicam with harsh direct on-camera fill flash that aggressively illuminates the near woman and the archway, flattening her features and putting sharp specular highlights on the linen, the stone and her shoulders; the courtyard and sea beyond stay visible but hazy, milky and underexposed relative to the flash. Warm, slightly saturated CCD tones, strong bloom and halation at the archway edges, minor digital noise. Candid amateur, Y2K editorial, unpolished but stylized.

为什么有效

  • Each woman is given a position in frame and a relationship to the camera — near foreground at frame right, turned back over her shoulder; further back, gazing out to sea — so depth is decided by the writer.
  • The camera is a character here: an early-2000s point-and-shoot with harsh on-camera fill flash, and the prompt says exactly what that flash does to the near subject and to everything behind her.
  • The look is named through defects rather than adjectives: CCD tones, bloom and halation at the archway edges, minor digital noise.
A 1980s portrait from a face you supplyGPT Image 2.5

作者 Saul Goodman4725 浏览原帖

A 1980s portrait from a face you supply

An identity-preserving restyle written in four blocks: keep the face, change the era, shoot it on film, light it believably.

  • 需要参考图
  • Face reference
  • 35mm film
  • Editorial portrait

需要 A face reference photo

提示词1,235 个字符
Create an authentic 1980s retro-vintage portrait in a 4:5 vertical aspect ratio, using the provided person as the exact facial reference. Preserve their identity, facial structure, recognizable features, skin tone, and natural expression with high accuracy—do not alter or beautify the face.
Give the subject a classic 1980s hairstyle and stylish period-accurate fashion with bold silhouettes, authentic textures, and effortless vintage attitude. Compose the portrait naturally with a strong editorial feel, keeping the subject as the clear focal point.
Capture the image as if shot on a 35mm analog film camera, with realistic film grain, subtle dust and texture, gentle softness, natural skin detail, slight color fading, and authentic analog imperfections. Use warm nostalgic color grading, soft neon highlights, subtle ambient glow, and direct on-camera flash to create the distinctive look of an iconic 1980s photograph.
Keep the lighting cinematic yet believable, with soft shadows, realistic highlights, natural contrast, and a slightly imperfect film exposure. The final image should feel genuinely photographed in the 1980s—not digitally recreated, with a timeless, nostalgic, fashionable, and effortlessly cool atmosphere.

为什么有效

  • "Do not alter or beautify the face" is stated in the same breath as the reference, which is the instruction most edits of this kind are missing.
  • The film look is broken into components a lab would recognise — grain, dust, slight fade, imperfect exposure — instead of being asked for as "vintage".
  • The last block pulls back toward plausibility: soft shadows, natural contrast, "cinematic yet believable". It reads as a correction the author added after the first attempts came back overcooked.
A candid night portrait beside a lit city wallGPT Image 2.5

作者 Aqsa1400 浏览原帖

A candid night portrait beside a lit city wall

A long descriptive prompt that keeps returning to the light: warm lanterns behind, gray-blue sky above, and skin that has to survive both.

  • 只要提示词
  • Night portrait
  • Practical lighting
  • Candid pose
提示词1,934 个字符
A realistic candid travel photo of a young East Asian woman standing on an old stone walkway beside a historic city wall at night. Behind her is a beautiful traditional Chinese-style building glowing with warm golden lights and deep red accents, with layered roofs, curved upturned eaves, detailed wooden architecture, and softly illuminated windows. Tall red lanterns hang from an ornate black lamp post nearby, adding a warm glow to the scene.
She has long, straight dark brown hair falling naturally over her shoulders and back, with a few loose strands around her face. She has soft youthful features and a gentle, genuine smile as she looks naturally toward the camera. She is wearing an oversized dark charcoal-gray leather jacket with a relaxed, slightly loose fit, paired with wide-leg light-gray jeans. A small black shoulder bag hangs at her side with a cute little white plush charm attached to it.
She is casually leaning against the black metal railing, resting one arm comfortably on it while her body is turned slightly toward the camera. The pose should feel spontaneous and relaxed, like a friend captured the moment while she was enjoying an evening walk. The old dark brick wall stretches behind her, with warm lights highlighting parts of the stonework, while the paved walkway continues into the distance with a few people and subtle city lights.
The sky is a deep gray-blue with soft clouds, contrasting naturally with the warm golden architecture and glowing red lanterns. Keep everything believable and lived-in, with realistic skin texture, natural facial proportions, individual hair strands, authentic leather and denim textures, soft evening shadows, subtle reflections from the lights, and realistic depth of field. Shot with a modern smartphone camera, slightly imperfect and natural rather than overly polished, with soft cinematic tones and a subtle dreamy atmosphere. Wide horizontal 16:9 composition.

为什么有效

  • The background is built before the subject, so the warm architectural light is established as the scene's key rather than added afterwards.
  • The pose is written as an accident — leaning on the railing, one arm resting, "like a friend captured the moment" — which is more useful direction than "candid" on its own.
  • It closes on the textures that break first at night: skin, individual hair strands, leather, denim, soft shadows and reflections.

Infographic & Diagram1

图表、剖面图、带标注的分区和讲解图。版面是写死的结构，不交给模型自由发挥。

A papercraft travel campaign with a working phone notificationGPT Image 2.5

作者 Saul Goodman1539 浏览原帖

A papercraft travel campaign with a working phone notification

A template with one placeholder for the destination, built as mixed physical materials and a UI card whose text is quoted exactly.

  • 只要提示词
  • Reusable template
  • Mixed materials
  • Quoted UI text
提示词2,501 个字符
[LOCATION] =
Create a premium, scroll-stopping Instagram travel campaign for [LOCATION] in a 4:5 vertical portrait format.
A large modern smartphone stands vertically near the upper center, acting as the main visual anchor. A single elegant map-navigation notification card overlaps the upper-middle of the phone, integrated naturally with realistic depth, soft shadows, and subtle reflections.
The notification should feature a warm-white, lightly frosted rounded rectangle with a refined mobile UI aesthetic and perfectly legible typography:
[map icon] MAPS — now
You keep searching.
Come visit [LOCATION] instead.
Keep the notification within the upper 25–40% of the composition. It should feel physically connected to the phone and must not cover the miniature destination scene.
Below the smartphone, a folded papercraft map labeled “[LOCATION]” unfolds into a handcrafted miniature world. Include only 4–6 highly recognizable elements that naturally represent the destination—such as iconic architecture, landscapes, nature, food, culture, or atmosphere.
Visual Materials
- Phone & distant background: realistic photography
- Buildings & landmarks: handcrafted painted 3D clay
- Map, terrain, roads & water: layered hand-cut paper
- Vegetation & small details: miniature clay and paper
Make the physical materials clearly visible through paper fibers, cut edges, layered surfaces, subtle clay texture, tiny imperfections, and soft dimensional shadows.
Add one large location pin and a tiny backpacked traveler standing on the map, looking toward the miniature world. Keep the traveler small and secondary, like a hidden discovery detail.
Use only 2–3 subtle travel doodles, such as a paper airplane with a dotted path, a tiny heart, and a simple hand-drawn route arrow. Keep them thin, imperfect, and journal-like.
Visual Hierarchy
1. Notification
2. Miniature [LOCATION] world
3. Location pin
4. Tiny traveler
5. Folded map
Maintain generous negative space and avoid clutter, excessive landmarks, unnecessary text, icons, props, or competing focal points.
Use a refined 3–5 color palette inspired by [LOCATION], warm cinematic or golden-hour lighting, shallow depth of field, realistic miniature photography, tactile materials, and sophisticated editorial advertising composition.
The final image should feel like a world-class handcrafted travel campaign-instantly recognizable, destination-specific, clean, emotionally inviting, and rich with subtle details worth discovering.

为什么有效

  • Materials are assigned per element — photographic phone, painted clay buildings, hand-cut paper terrain — so "handcrafted" becomes a set of instructions rather than a mood.
  • The notification copy is written out in full and given a position and a size limit ("upper 25–40% of the composition", must not cover the scene below).
  • It caps the number of landmarks at four to six and the doodles at two or three, which is why the result reads as designed rather than crowded.

Product & Ads1

产品图、样机和广告画面。大多数会先说清材质、光和角度，再说产品是什么。

A lookbook poster with its garments labelledGPT Image 2.5 Sunburst

作者 诗泳研AI538 浏览原帖

A lookbook poster with its garments labelled

A fashion catalogue frame where the annotations are part of the design: thin arrows pointing at each piece, each with its own quoted label.

  • 只要提示词
  • Annotated callouts
  • Lookbook
  • Type hierarchy
提示词1,353 个字符
Create a clean fashion e-commerce lookbook poster featuring a young female model in a cute, simple autumn outfit.
Use a soft pastel pink, blue and lavender gradient background. Keep the overall design minimal, clean and slightly playful, like a modern Korean/Japanese fashion catalog.
Show the full-body model as the main focus, with realistic skin, natural proportions and clear clothing details. The outfit should be simple and wearable: cardigan or hoodie, fitted top, shorts or pleated skirt, socks and sneakers/Mary Jane shoes.
Add handwritten English annotations with thin arrows pointing to individual clothing pieces, such as “Cardigan”, “Polo Shirt”, “Shorts”, “Sneakers” and “Bag”.
Add a bold English fashion headline at the top: “AUTUMN DAILY OUTFIT”, plus a few small decorative English phrases such as “NEW SEASON 2026”, “SIMPLE · COMFY · BETTER ME” and “OUTFIT IDEAS FOR A BRIGHTER YOU”.
Use a structured rectangular frame around the main content, subtle rounded corners, plenty of negative space, and only a few minimal heart/star doodles.
The final image should feel like a polished fashion e-commerce campaign poster — minimal, youthful, realistic, highly readable typography, clean product presentation, not cluttered, with the outfit as the visual focus.
Vertical 3:4 composition, high-resolution commercial fashion photography.

为什么有效

  • The labels are quoted individually — "Cardigan", "Polo Shirt", "Shorts" — and attached to arrows, so the callouts land on the right garment instead of floating.
  • The headline and the small decorative phrases are given separately, which keeps the type hierarchy from collapsing into one size.
  • The outfit is asked for as wearable and simple, and the frame, rounded corners and doodle count are all capped, which is what keeps a busy layout readable.

Design & Poster3

海报、封面、图标和界面稿——构图本身就是需求的一部分。

A Porsche poster written as a spec sheetGPT Image 2.5 Flare

作者 ᴍᴜʀᴘʜʏ5784 浏览原帖

A Porsche poster written as a spec sheet

A car poster laid out as labelled blocks — scene, product, typography, specs, lighting, mood — with nothing written as prose.

  • 只要提示词
  • Block structure
  • Numeric values
  • Product poster
提示词689 个字符
4:5 vertical social poster, ultra high resolution, 8K
SCENE / BACKGROUND:
Flat racing yellow with subtle gradient (light → deep amber)
Grain texture 2–3%
Soft rectangular overlays at 3% opacity
PRODUCT:
Porsche 911 GT3 RS
Gloss yellow with carbon details
Rear 3/4 angle, slightly elevated
Wing dominant
Position:
Right of center
TYPOGRAPHY:
“PORSCHE”
Ultra tall, condensed, stretched
Deep charcoal
Secondary:
“PORSCHE 911 GT3 RS”
EDITORIAL:
“Porsche — Where Precision Meets Passion”
Focus:
engineering, obsession, track DNA
FADED TEXT:
“GT3 RS” at 3–5%
SPECS:
386 kW / 525 PS
3.2 s
296 km/h
LIGHTING:
Clean studio, sharp highlights
MOOD:
Precision.
Iconic.
Timeless performance.

为什么有效

  • Every block is a heading with values under it, which is why the layout comes back predictable: position, colour and weight are stated per element rather than described.
  • The background is given real numbers — a gradient direction, 2-3% grain, overlays at 3% opacity — and so is the ghosted "GT3 RS" at 3-5%.
  • The typography block quotes each string separately and gives the primary one its treatment, so the model is not left to invent a wordmark.
An extreme close-up that argues against flawless skinGPT Image 2.5

作者 Jahan Zaib2615 浏览原帖

An extreme close-up that argues against flawless skin

A portrait prompt whose whole method is refusing retouching: freckles, pores, peach fuzz and two healed scars, written out one by one.

  • 只要提示词
  • Extreme close-up
  • Skin texture
  • No retouching
提示词2,991 个字符
Create an irresistibly captivating, ultra-realistic close-up portrait of a naturally beautiful young woman designed to create an instant “STOP SCROLLING” reaction.
The camera is extremely close to her face, capturing the mesmerizing warmth of her deep brown eyes as the undeniable focal point. Her eyes must look alive and intensely realistic, with intricate iris patterns, tiny natural veins, wetness along the waterline, crisp catchlights, and individual eyelashes sharply resolved.
Her beauty is raw, authentic and imperfect—visible freckles scattered naturally across her cheeks and nose, realistic pores, subtle peach fuzz, tiny beauty marks, natural skin texture and delicate variations in tone. Do NOT make her skin artificially flawless. Above one eyebrow is a thin, naturally healed scar, while a subtle healed scar curves along her cheek, giving the portrait a mysterious, unforgettable character.
Her long, glossy black hair falls naturally around her face, with a few delicate strands crossing her forehead and cheek. Her eyebrows are full, naturally shaped and individually detailed. Soft, naturally pink lips remain slightly relaxed, creating an effortless expression that feels candid rather than posed.
THE VIRAL HOOK: Capture the exact split-second where she looks directly into the camera as if she has noticed the viewer. Her expression is calm, mysterious and emotionally magnetic—subtle vulnerability mixed with quiet confidence. The viewer should feel like they accidentally captured a private, once-in-a-lifetime moment.
Lighting is soft cinematic daylight entering from one side, producing luminous highlights across her eyes, cheekbones and skin while maintaining realistic shadows. Add an extremely subtle atmospheric glow around the edges without making the image dreamy or artificial.
Composition: extreme facial close-up, eyes positioned near the visual center, face filling almost the entire frame, slightly unconventional tilted camera angle, intimate perspective, natural asymmetry, realistic facial proportions.
Camera realism: full-frame professional camera, 85mm portrait lens, f/1.4, razor-sharp focus on both eyes, extremely shallow depth of field, natural optical bokeh, realistic lens rendering, subtle filmic micro-contrast.
Visual quality: hyper-photorealistic, cinematic editorial photography, 8K-level detail, HDR, physically accurate skin, realistic reflections, lifelike eyes, natural color grading, premium fashion-magazine aesthetic, authentic human imperfections, breathtaking micro-details.
ABSOLUTELY AVOID: plastic skin, excessive retouching, doll-like face, CGI appearance, uncanny eyes, exaggerated makeup, fake freckles, over-saturated colors, excessive HDR, artificial skin smoothing, distorted facial anatomy, extra eyelashes, asymmetrical eyes, blurry eyes, watermark, text, logo.
Overall feeling: beautiful enough to stop the scroll, realistic enough to feel like a photograph, mysterious enough to make people stare twice.

为什么有效

  • The eye is treated as the subject and specified at a level most prompts never reach — iris pattern, tiny veins, wetness on the waterline, catchlights, individually resolved lashes.
  • "Do NOT make her skin artificially flawless" is the load-bearing line, and the imperfections that follow are named individually so the model has something to render instead of nothing to remove.
  • The framing is engagement-first — it asks in so many words for a "STOP SCROLLING" reaction and calls one paragraph THE VIRAL HOOK — so the useful part is the skin and eye detail, not the pitch around it.
Two landscapes merged into one animated 16-frame loopGPT Image 2.5

作者 KrishnaG2441 浏览原帖

Two landscapes merged into one animated 16-frame loop

A composite of two reference photographs into a single place, then sixteen frames of environmental motion that have to close back on frame one.

  • 需要参考图
  • Two references
  • 16-frame loop
  • Fixed camera

需要 Two landscape reference images

提示词2,716 个字符
Turn the two reference images into one breathtaking, ultra-realistic cinematic travel landscape with a seamless sunset animation loop.
SCENE
Combine the key elements from both images into one natural-looking destination. Include the grand Ottoman-style mosque with multiple minarets and domes, Istanbul-style waterfront and boats from the first image, together with the peaceful alpine lake, colorful European village, dramatic snow-capped mountains and beautiful hilltop stone castle from the second image.
Create a wide scenic waterfront view with historic architecture on one side and the castle and mountains on the other. Add colorful flowers, green trees, Mediterranean plants, a vintage lantern and an elegant terrace with a small café table in the foreground.
SUNSET & LIGHTING
Beautiful golden-hour sunset with pastel blue, pink, peach and orange clouds. Warm sunlight illuminates the mosque, castle, village, mountains and boats. The lake reflects the sunset, buildings and mountains with realistic gentle ripples and golden reflections.
ANIMATION
Create 16 consecutive frames of a smooth seamless loop. Animate only subtle environmental movement: slowly drifting clouds, gentle water ripples, boats moving slightly, birds flying in the distance, flowers and tree leaves softly moving in the breeze, and subtle candle/lantern flickering.
Keep the camera completely fixed. Keep the mosque, castle, mountains, houses and all major objects perfectly consistent across every frame. Frame 16 must smoothly transition back into Frame 1.
STYLE
Ultra-realistic cinematic travel photography, luxury tourism advertisement, breathtaking golden-hour atmosphere, realistic architecture, detailed mountains, natural vegetation, beautiful water reflections, atmospheric depth, high dynamic range, rich but natural colors, professional photography, photorealistic 8K quality.
CAMERA
Wide cinematic landscape view, 24mm lens, stable locked camera, natural perspective, realistic depth of field, consistent framing and lighting across all frames.
ANIMATION SETTINGS
16 frames, approximately 2.2-second complete loop, smooth continuous GIF animation, no camera movement, no flickering.
NEGATIVE PROMPT
cartoon, anime, painting, CGI, 3D render, low quality, distorted buildings, warped mosque, crooked minarets, malformed castle, distorted mountains, duplicated boats, floating objects, unrealistic reflections, fake water, excessive fog, oversaturated colors, extreme HDR, blurry image, motion blur, camera shake, zoom, flickering, morphing buildings, changing architecture, disappearing objects, duplicated birds, unnatural clouds, image seams, visible collage, text, logo, watermark, border, black bars, artifacts.

为什么有效

  • The merge is itemised: which elements come from the first image, which from the second, and where each sits in the final composition.
  • Motion is restricted to a named list — drifting clouds, water ripples, boats, distant birds, leaves, a flickering lantern — and the camera is held fixed, which is what keeps a loop from drifting.
  • "Frame 16 must smoothly transition back into Frame 1", plus a demand that the architecture stay identical across every frame, is the whole seam.

Illustration & Anime2

手绘和渲染风格而不是照片风格，从动漫到水彩再到等距 3D。

A landmark as a handmade miniature on a paper backgroundGPT Image 2.5

作者 simeon-sanai9392 浏览原帖

A landmark as a handmade miniature on a paper background

A bracketed template that turns any famous structure into a sculpted tabletop model, with the restraint written in as instructions.

  • 只要提示词
  • Fill-in template
  • Miniature diorama
  • Negative space
提示词1,019 个字符
Create a charming handcrafted miniature travel scene featuring [ICONIC STRUCTURE] as the main focal point.
Show the landmark as a beautifully sculpted tiny 3D model, with soft rounded details, handmade textures, delicate imperfections, and a whimsical storybook feeling. Surround it with a few subtle elements that represent its location—such as tiny trees, flowers, streets, boats, mountains, clouds, or local objects—without making the scene crowded.
Place everything on a clean warm-white textured paper background, with plenty of elegant negative space. Add a small tasteful wooden or paper travel plaque containing:
[STRUCTURE NAME]
[CITY, COUNTRY]
Famous for: [SHORT UNIQUE FACT]
Use soft natural lighting, gentle shadows, pastel yet realistic colors, miniature diorama depth, handcrafted clay/paper textures, and a premium cute travel-journal aesthetic. Centered composition, highly detailed landmark, adorable but sophisticated, clean and collectible travel-card design, no photorealistic people, no clutter.

为什么有效

  • It says what to leave out as clearly as what to include: a few surrounding elements "without making the scene crowded", plenty of negative space, "no clutter".
  • The handmade quality is specified as physical properties — soft rounded details, delicate imperfections, clay and paper texture — rather than as the word "handcrafted".
  • The plaque has a fixed three-line structure, so the caption is part of the design rather than an afterthought.
A split-screen character bible from two reference sheetsGPT Image 2.5

作者 TechieSA626 浏览原帖

A split-screen character bible from two reference sheets

A poster laid out region by region — left half, centre divider, right half, bottom palette — with every piece of text quoted.

  • 需要参考图
  • Split screen
  • Two references
  • Quoted typography

需要 Two character reference sheets

提示词1,721 个字符
“Create a premium cinematic character bible sheet for ALAN & T-REX. Use uploaded character sheets as strict visual reference for both characters. Do not change either's appearance.
LAYOUT: Split screen partner format. Two halves divided by a bold dramatic dividing element in the center.
LEFT SIDE — ALAN: Red watercolor splash behind him fading into center. Large bold brushstroke text ALAN top left in deep red. Below small text: PALEONTOLOGIST / ISLA NUBLAR. One massive dramatic cropped hero image of Alan from mid-thigh up — full sprint stance, light blue denim shirt, bold red bandana catching the light, wide eyes locked forward in pure survival mode, red watercolor splash radiating behind him.
CENTER: Bold dramatic JURASSIC PARK in deep olive green, slightly worn. Below it small text: ISLA NUBLAR. RUN. OR BE EATEN.
RIGHT SIDE — T-REX: Green watercolor splash behind him fading into center. Large bold brushstroke text T-REX top right in deep olive green. Below small text: PREDATOR / APEX. One massive dramatic cropped hero image of the T-Rex from mid-body up — jaw wide open, rows of teeth visible, fierce amber eye locked toward the center, dark olive green watercolor splash radiating behind him.
BOTTOM CENTER: Color palette — light blue, bold red, tan khaki, dark olive green, amber, deep black. Tagline centered: LIFE FINDS A WAY.
OVERALL: Clean white background, bold red watercolor left side, green watercolor right side, dramatic deep red center divider, bold flat color blocking, chunky simplified forms, hard edge shadows, thick black outlines, vibrant saturated colors, minimal clean typography, cinematic cel-shaded 3D anime, hand-painted textures, not cartoon not Disney not Pixar, print ready.”

为什么有效

  • Layout is dictated as named regions, and each region gets its own colour, its own text and its own crop, so nothing is left to composition luck.
  • Both characters are pinned to uploaded sheets with "do not change either's appearance", which is what keeps a two-subject poster from drifting.
  • Every string that appears in the image is written out, down to the small type under each name.

Editing & Restyle2

从你自己的图开始改。这类提示词的重点是告诉模型改什么，更重要的是告诉它别动什么。

A phone held across the face, showing the face behind itGPT Image 2.5

作者 TSUBAKI1080 浏览原帖

A phone held across the face, showing the face behind it

An identity-preserving portrait built on one trick: the subject hides behind a phone whose screen shows the same face, live.

  • 需要参考图
  • Frame within frame
  • Identity lock
  • Phone portrait

需要 A face reference photo

提示词2,133 个字符
Ultra-realistic iPhone creative portrait, identity preserved exactly from reference image. Young woman standing indoors in front of a glass storefront or reflective wall, photographed from a slightly high front angle. She holds a smartphone horizontally directly in front of her face, covering her eyes and upper face in real life. The phone screen faces the camera and clearly shows a live camera/photo view of her face. Composition: The real woman is visible behind the phone, but her face is mostly hidden by the horizontal phone. On the phone screen, her face appears clearly centered, smiling softly and winking. The phone acts like a frame within the frame. Her hand grips the top edge of the phone, fingers visible across the top. Pose: One hand holds the phone horizontally across her face. Her other hand is lower near her waist making a peace sign or casual gesture. She stands relaxed, facing camera. Outfit: White fitted zip-front sleeveless top, loose light-wash baggy jeans, brown belt, playful accessories, rings, bracelet, red-and-white shoulder bag, small plush/keychain accessories hanging from bag or belt. Phone screen details: The phone screen shows a clean live-camera image of her face with a wink and soft smile. Add cute sticker-style doodles around the screen/photo area: small colorful sparkles, stars, flower sticker near the face. The stickers should look like playful phone-camera decoration, not random floating objects. Environment: Indoor mall/storefront area with glass doors, reflections, polished floor, soft daylight mixed with indoor light, casual streetwear snapshot vibe. Camera + lighting: 0.5x iPhone wide-angle photo, close candid framing, realistic reflections on glass and phone screen, slight distortion on hands, raw social-media feel. Important: Match the phone-covering-face composition, horizontal phone screen showing her face, wink expression on screen, hand gripping phone, playful stickers, white zip top, baggy jeans, accessories, and indoor glass background closely. Natural skin texture, realistic screen reflections, no beauty filter, no smoothing. 4K vertical 9:16.

为什么有效

  • The conceit needs the model to hold two versions of one face at once, so the prompt states the relationship explicitly — hidden in life, clearly centred on the screen, "a frame within the frame".
  • The hand is placed on the phone edge with fingers visible across the top, which is the detail that stops the phone floating.
  • Wardrobe and props are listed to the level of the charms hanging off the bag, so the picture reads as a real outfit rather than a generic one.
Eight enamel pins built around one portraitGPT Image 2.5

作者 Mira545 浏览原帖

Eight enamel pins built around one portrait

A product-photography prompt with a fixed count: exactly eight pins, one of them a likeness taken from a reference image.

  • 需要参考图
  • Product render
  • Exact count
  • Portrait reference

需要 A portrait reference image

提示词1,416 个字符
Create a hyper-realistic enamel pin collection in a vertical 4:5 composition on a dark textured background. Design exactly 8 separate premium pins with polished gold-metal outlines, glossy hard-enamel surfaces, subtle raised edges, realistic reflections, and soft shadows. The main medium-sized center pin is a portrait of [CHARACTER], preserving the reference image’s recognizable face, hairstyle, skin tone, expression, clothing, and key features while translating them into a detailed premium enamel illustration. Add 1 small name pin reading “[NAME]” and 1 small personality-word pin reading “[WORD]”. The remaining 5 pins should represent [CHARACTER]’s interests and lifestyle, such as creative work, AI/technology, visual creation, photography, travel, friendship, hobbies, or everyday objects, using clever symbolic illustrations rather than excessive text. Arrange everything organically around the portrait with varied pin shapes and sizes, generous spacing, and no rigid grid. Use bold, flat, bright colors with sophisticated contrast, crisp details, realistic metallic depth, and high-end commercial product lighting. Make it feel like a limited-edition designer pin set photographed in a studio—modern, vibrant, personal, collectible, and instantly eye-catching. No city names or specific landmarks, no duplicated pins, no clutter, no random text, no misspellings, no cheap cartoon or plastic appearance.

为什么有效

  • The count is exact and the roles are assigned — one portrait, one name, one word, five symbolic — so the arrangement is decided before the model starts inventing.
  • The material is described the way a manufacturer would: gold metal outlines, hard enamel, raised edges, real reflections.
  • "Arrange organically … no rigid grid" plus varied shapes is what separates this from a contact sheet.
怎么写

GPT Image 2.5 提示词怎么写

上面这些提示词题材各不相同，结构却高度一致。下面就是那个结构，一段一段写好了，可以直接复制。

模板308 个字符
[FORMAT] 3:2 横构图，摄影质感，50mm 镜头 f/2 拍摄。不是插画，不是三维渲染。
[SUBJECT] 一位三十出头的女性，短黑发，灰色羊毛大衣，领子立起来。双手插兜，重心在后脚，视线看向画面左外侧。
[SCENE] 黄昏的窄街，地面是湿的，身后店铺虚焦。主光来自画面左侧的橱窗，天光作为冷蓝色补光，没有直射阳光。
[TEXT] 招牌上的字是「NORTHERN COFFEE」，窄体全大写无衬线，深绿底白字，位于画面上三分之一处。画面其他任何地方都不出现文字。
[KEEP] 大衣保持灰色，头发保持短发。不要水印，不要签名，画面里不要出现其他人，不要镜头光晕，除上面那块招牌外不出现任何文字。
  1. 01

    先说清这是一张什么样的图

    介质、比例、镜头，都写在主体前面。开头就写主体，模型给你的是它的默认审美；开头先定格式，出来的才是你要的那种图。

  2. 02

    按镜头能看见的东西来写主体

    年龄、衣着、姿态、视线，这些模型读得懂，是指令。「美丽」「惊艳」这类形容词读不懂——提示词太长要删的时候，先删它们。

  3. 03

    交代环境，并且说清光从哪来

    把光的方向写出来，比任何一个「高质量」词都更能让画面像照片。这一页里真能骗过眼睛的提示词，无一例外都点明了光的方向和色温。

  4. 04

    文字要一字不差地引用，还要说清位置

    画文字是 GPT Image 2.5 的强项，也是最经不起转述的地方。把字符串加引号写死，给它位置和字体，然后禁掉其他位置的一切文字。

  5. 05

    最后写清什么不能变

    结尾一小段禁止清单，专门用来堵模型的老毛病：莫名其妙的水印、自己编的招牌、边上突然多出来一个人。除了主体段落，这是这个库里出现频率最高的一段。

  6. 06

    给了参考图，就要给它一个明确的职责

    改图类提示词失败，基本都是因为模型不知道哪些东西该保留。先说清哪张是参考、它负责什么——比如「图 1 是产品，标签文字和比例保持不变」——然后只描述该变的部分。

出图不理想怎么办

改那句没生效的指令

同一条提示词重跑一遍，只会换一个种子再错一次。改一句话。

文字拼错了
把要写的字加引号写死，声明这是画面里唯一的文字，并给它一个位置。转述过的招牌，拼写也会被转述。
看起来像渲染图，不像照片
写明镜头焦段和光圈，写明光从哪来。把「photorealistic」「8K」这类词删掉——它们把画面往渲染感上推，而不是往真实感上推。
改图时把不该动的地方也改了
在描述要改什么之前，先说清参考图负责哪一部分，并列出必须保持一致的特征。
画面里出现了水印或签名
在最后一行明确禁掉。这个习惯是模型从你要的风格里学来的，改风格那一段没用。
版面完全不按要求来
用画面里的绝对位置来描述——上三分之一、画面左侧、占下半幅——而不是「在后面」「在旁边」这种相对说法。
只有一个细节不对
只改那一句再跑。其余部分一个字都别动；整段重写等于一次性改掉了所有变量。
常见问题

关于 GPT Image 2.5 提示词

在复制第一条提示词之前，大家通常会问这些。

什么样的 GPT Image 2.5 提示词算好？

把格式、主体、场景、文字、限制分开写，而不是堆形容词。看看上面那些经得起复制的提示词：它们都写明了镜头和光的方向，把要画的文字一字不差引出来，并且在结尾列出一小段不能变的东西。

Sunburst 和 Flare 有什么区别？

这是 OpenAI 给 GPT Image 2.5 的两个官方检查点。Sunburst 偏精细，Flare 偏日常快出。上面的生成器两个都有，为其中一个写的提示词在另一个上也能跑。筛选按钮可以看每条帖子当时用的是哪一个。

这些提示词是从哪来的？

来自 awesome-gpt-image-prompts，一个公开的 GitHub 语料库。它收录 X 和 Reddit 上公开发布的 GPT Image 2.5 出图，连同作者公开的提示词一起，为每个来源账号署名并链接原帖。本页每隔几小时读一次这个语料库。提示词是原文照录；卡片上的一句话概括和「为什么有效」是我们写的。

怎么在 Seadanse 上用这些提示词？

复制走，或者点「用这条」，生成器里就已经填好了。提示词点名要参考图的话，把图传上去，选 Sunburst 或 Flare，选比例和分辨率，然后出图。

GPT Image 2.5 的提示词应该写多长？

比视频提示词短。这里真正有效的，大多在一到六句话之间。再往长里写，只有在增加约束时才有用——要画的确切文字、版面位置、不能变的清单；如果加的是形容词，只会更差。

这些提示词可以商用吗？

提示词是作者本人公开发布的原话，本页为每一条署名并链接原帖。你用提示词生成出来的东西，适用你所使用的那个服务的条款。本页不对这些帖子里的图片本身授予任何权利。

这一页多久更新一次？

每周。每周一会扫一遍 X 和 Reddit，新帖按公开的入选线筛一次，通过的补进语料库和这一页。X 的互动数字跟着语料库刷新；Reddit 那部分是上一次扫描时的数值，因为 Reddit 不响应自动化请求。

来源语料库：GitHub 上的 awesome-gpt-image-prompts（MIT）。每张图都为作者署名并链接原帖；互动数字截至语料库最近一次刷新。提示词均为作者原话。 打开仓库

挑一条，用 GPT Image 2.5 跑一遍

复制提示词，选 Sunburst 或 Flare，看看它能出什么。注册即送初始额度。

打开生成器了解 GPT Image 2.5