GPT Image 2 —
提示词模板，复制即用

参考图片1的苏打水，帮我给这只苏打水生成一张15秒的8个分镜的故事脚本图。创意核心是图2的女生从冰箱里拿出苏打水真实画面，分镜镜头分别是冰箱打开展示苏打水正面图（分镜以正面视角切入）、图2女生打开瓶盖看到冒泡的苏打水、女生45度仰头喝苏打水的侧面照等场景。每张图包含故事分镜图，分镜示意文字以及其他特殊需要标注的内容，分镜图上不要有文字，文字放分镜图下方。图片内容里的分镜效果图的比例16:9。

请根据中国西游记神话【主题：六耳猕猴】轮廓自动生成一张高审美的 “中国神话轮廓宇宙 / 收藏版叙事海报” 风格作品，角色面部要稍稍写实。不要将画面局限于固定器物或常见容器，不要优先默认瓶子、沙漏、玻璃罩、怀表之类的常规载体，而是由 AI 根据主题自行判断并选择一个最契合、最有象征意义、轮廓最强、最适合承载完整叙事世界的主轮廓载体。不要完全的人物正侧面，这个主轮廓可以是与角色相关的神器法术、坐骑、器物、建筑、门、塔、拱门、穹顶、楼梯井、长廊、雕像、侧脸、眼睛、手掌、头骨、羽翼、面具、王座、圆环、裂缝、光幕、阴影、几何结构、空间切面、舞台框景、抽象符号或其他更有创意与主题代表性的视觉轮廓，要求合理布局。优先选择最能放大主题气质、最能形成强烈视觉记忆点、最能体现史诗感、神秘感、诗意感或设计感的轮廓，而不是最安全、最普通、最常见的容器。画面的核心不是简单把世界装进某个物体里，而是让完整的主题世界自然生长在这个主轮廓之中、之内、之上、之边界里或与其结构融为一体，形成一种 “主题宇宙依附于一个象征性轮廓展开” 的高级叙事效果。主轮廓必须清晰、优雅、有辨识度，并在整体构图中占据核心地位。轮廓内部或边界中需要自动生成与这个人物在西游神话中强绑定的完整叙事世界，内容应当丰富、饱满、层次清晰，包括最能代表主题的标志性场景、核心建筑或空间结构、象征符号与隐喻元素、角色关系或文明痕迹、远景中景近景的空间递进、具有命运感和情绪张力的氛围层次，以及门、台阶、桥梁、水面、烟雾、路径、光源、遗迹、机械结构、自然景观、抽象形态、生物或道具等叙事细节。所有元素必须统一、自然、有主次、有层级地融合，像一个完整世界真实孕育在这个轮廓结构之中，而不是简单拼贴、裁切填充、素材堆叠或模板化背景。整体构图需要具有强烈的收藏版海报气质与高级设计感，大结构稳定，主轮廓强烈明确，内部世界具有纵深、秩序和呼吸感，细节丰富但不拥挤，内容丰满但不杂乱，可以适度加入小比例人物剪影、远处建筑、光柱、门洞、桥、阶梯、回廊、倒影、天光或远景结构来增强尺度感、故事感与史诗感。整体画面要安静、宏大、凝练、富有余味，不要平均铺满，不要廉价热闹，不要无重点堆砌。风格融合收藏版电影海报构图、高级叙事型视觉设计、梦幻水彩质感与纸张印刷品气质，强调纸张颗粒感、边缘飞白、水彩刷痕、轻微晕染、空气透视、柔和雾化、局体积光、光雾穿透、大面积留白与克制版式，让画面看起来像设计师完成的高端收藏版视觉作品，而不是普通 AI 跑图。整体气质要高级、诗意、宏大、神圣、怀旧、安静、具有传说感和叙事感。最终要求：第一眼有强烈的主题识别度和轮廓记忆点，第二眼有完整丰富的叙事世界，第三眼仍有细节和余味。轮廓选择必须具有创意和主题匹配度，尽量避免重复、保守、常见的容器套路，优先选择更有象征性、更有空间感、更有设计潜力的轮廓形式。不要普通背景拼接，不要生硬裁切，不要模板化奇幻素材，不要游戏宣传图感，不要过度卡通化，可以偏写实一些但不要失去艺术感，不要形式大于内容。如果合适，可以在画面下方正中少量自然加入低调克制的标题或落款，让它更像收藏版海报设计的一部分，但不要喧宾夺主。

Editorial-style infographic poster titled 'SOLAR SYSTEM GUIDE'. Vertical magazine layout, retro science textbook aesthetic. Side-view illustration of the solar system showing all eight planets along an orbital arc (Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, Neptune — not to scale). Each planet has a small ID card next to it showing: name (English), diameter, distance from sun, rotation period, surface temperature, known moons count, and one short humanized caption. Dense but legible serif typography on a dark navy background with metallic gold and cream accents. Print-magazine quality.

Vertical infographic poster titled 'A CUP OF COFFEE: FROM SEED TO SIP'. Centered around a giant cross-section of a coffee cup. Layered scenes inside the cup: bottom layer shows a miniature Ethiopian highland coffee plantation labeled 'Altitude 1500-2200m · Arabica'; middle layer shows a roasting temperature curve labeled 'Light 195°C · Medium 210°C · Dark 225°C'; upper layer shows espresso extraction with a 25-second timer; foam layer shows latte art patterns. Around the cup are small icons of brewing devices (V60, espresso machine, French press, AeroPress) and tasting notes (chocolate, citrus, berry, floral). Warm beige background, hand-drawn illustration style with detailed annotations in editorial serif font.

Data visualization poster titled 'GLOBAL WARMING DATA REPORT 2026'. Bold sans-serif title with text color gradient from icy blue to fiery red, symbolizing rising temperature. Subtitle in monospace lists key data: 'CO₂: 428ppm · Sea level: +3.7mm/year · Arctic ice loss: -13%/decade'. Body of poster contains 6 data charts: a line chart of global average temperature 1880–2026, a bar chart of top emitters by country, a melting-glacier silhouette infographic, a sea-level rise diagram, a CO₂ concentration curve, and a heatmap of regional temperature anomalies. Environmental editorial style, clean grid layout, off-white background with red and blue accent colors. Print-quality typography.

请创作一张关于「【面试综合分析题答题结构】」的公考可视化信息图，目标是帮助读者直观理解：它是什么、核心考点有哪些、常见出题方式是什么、容易出错的地方在哪里、正确的分析路径和解题/答题方法是什么、在不同题型或场景下如何应用。画面要像高质量公考讲义 + 手绘教育海报，专业、清晰、信息丰富，但不要杂乱。视觉风格：竖版或横版均可，干净的浅色纸张背景，深蓝主标题，黑色/深灰正文线条，搭配少量优雅的蓝色、青绿色、金色、红色作为重点标记。使用圆角卡片、细线边框、编号标签、手绘箭头、局部放大框、流程图、小结栏和易错提醒栏。整体要美观、平衡、有“上岸讲义”质感，让人一眼看懂这个知识点的结构、规律和使用方式。内容结构建议包括：顶部标题区：知识点名称 + 一句话定位核心定义区：这个知识点“到底在考什么”题型识别区：常见命题方式/题干特征/关键词信号解题路径区：标准步骤或思考流程，用箭头串联易错点区：常见误区、陷阱选项、错误思路对比区：和相近考点的区别例题演示区：给一个简短典型例子，展示正确拆解方式记忆总结区：口诀、结论或高频提醒版面要求：信息层级清晰，重点醒目，避免堆砌大段文字；尽量把抽象方法转化为结构图、流程图、对比表、关键词标签和视觉提示；整体适合用于公考备考笔记、社媒图卡或讲义封面。

Coffee brand visual identity mockup for 'GROUNDED'. Logo: minimal coffee-bean silhouette merged with the letter G in negative space. Brand palette: deep brown (#3C2415), cream (#FFF8F0), gold accent (#C9A84C). Scene: 45-degree overhead flat-lay photography of a dark walnut wood desk in soft morning light. Items arranged neatly: stack of business cards (front shows logo, back shows full coffee-bean pattern), kraft paper takeaway coffee cup with embossed logo, retail coffee bag, menu card, linen apron with embroidered logo, brass branding stamp. Editorial advertising photography, high detail.

Health-food brand visual identity mockup for 'GREEN BOWL', a fresh salad brand for office lunch. Bright fresh aesthetic: light beige background (#FFF9E8) with neat product layout. Logo on top-left: an inverted bowl outline + a lettuce leaf inside + rounded sans-serif wordmark. Color system shown as four circular swatches: grass green, tomato red (#E54A47), avocado yellow-green (#B5C77E), butter cream. Items displayed: takeaway kraft paper bowl with transparent lid, paper napkin sleeve with logo, cotton tote bag, glass dressing bottle with custom label, menu card showing 5 salad bowls, and a compostable cutlery set. Soft natural light, editorial flat-lay style.

Visual identity mockup for 'NEXTLAND', a K-12 quality-education brand for children aged 6-12. Lively but not childish color palette: sunshine yellow (#FFC93C), ocean blue (#4A90E2), coral orange (#FF7E5F), grass green (#7ED321), cream base. Logo: circular icon combining island silhouette + lighthouse + small sailboat, with rounded wordmark 'NEXTLAND' below. Items shown on a creamy desk: enrollment brochure, stickers (lighthouse, island, sailboat, compass), notebook with embossed logo, pencil set, canvas tote bag, branded water bottle, classroom poster, name-badge lanyard. Cheerful overhead flat-lay photography, soft natural light.

A modern e-commerce website homepage screenshot. The brand sells pink glittering fairy milk and deep blue cosmic milk. Dark-mode UI with subtle glassmorphism panels, minimalist and clean layout. Hero section shows two tall product bottles side by side with crisp pricing labels. Top navigation: Shop / About / Cart. Realistic browser-frame styling, sharp typography, soft glow accents on the bottle highlights.

UI design screenshot showing a complete bank app transfer flow. Four phone screens arranged horizontally with arrows connecting each step. Design language: financial-grade trustworthy feel, deep navy primary (#1A2A6C) with white cards and gold accent. Screen 1 'Account Home': balance card on top + recent transactions list + bottom 4-grid quick actions (Transfer / Deposit / Bills / Statements). Screen 2 'Transfer Input': recipient input + contact list + large amount keypad + memo field. Screen 3 'Confirm': summary card with all transfer info, recipient avatar, amount in large text, fee breakdown, large 'Confirm' button. Screen 4 'Success': green checkmark animation, transaction ID, share/save receipt buttons. Realistic iOS-style status bar on each screen.

A speculative screenshot of 'GPT in 2028': a futuristic AI assistant interface, dark glassmorphic UI with neural-network style background animation suggested by static. Top bar shows model selector 'GPT-7-omni · 50T params'. Left sidebar shows multimodal capability tabs: Chat, Voice, Image, Video, Code, Agent, Memory, Brain Link. Main chat area shows a conversation about quantum computing with rich inline visualizations (3D molecular model, real-time stock chart, generated video thumbnail). Right panel shows live agent task tracker with 12 background tasks. Realistic browser-window framing.

Open-world RPG character design sheet (CharacterSheet) for a 20-year-old female swordsman. Light gray grid paper background, formal character design document style. Center: standard three-view (front / side / back) of the character wearing light leather combat armor (deep brown body + silver shoulder guards + dark red cape + a longsword and potion vials at the waist), long hair tied into a battle braid, leather boots, archer-style fingerless gloves. Surrounding panels: weapon close-ups (longsword detail, throwing daggers, spell scroll), facial expression sheet (default / determined / surprised / smirk / battle cry / wounded), height comparison chart, color palette swatches. Anime concept-art quality, clean linework, soft cel-shading.

Full-body 3D render of a near-future cyberpunk-style delivery courier character. Asian female face, short hair with one side shaved into circuit-board patterns, the other side has a purple dyed streak falling to the chin. Right eye has a HUD cybernetic implant (pupil glows soft cyan), left ear wears a translator earpiece. Wearing a customized work jacket with the holographic delivery company logo 'FLASHPOINT' on the chest, jacket interior shows a heat-vent mesh with blue LED breathing lights. Carries a scanner gun and a folded drone (the size of a lunchbox) at the waist. Backpack contains a temperature-controlled food container. Standing pose, neon-lit rainy alley background. Photoreal Octane render, cinematic lighting, sharp detail.

3D photorealistic close-up render of an Edomae sushi platter. A rectangular dark slate stone tray (approx 30cm x 15cm, visible natural stone texture and rough edge) holds 9 pieces of sushi neatly arranged: from left to right — fatty tuna otoro (vivid red with marbled fat, glossy oil sheen), salmon (visible orange flesh fibers with a few salmon roe), uni gunkan (bright yellow uni grains with seaweed wrap), grilled eel unagi (caramel-colored teriyaki glaze on shiny skin), hokkigai (red-and-white gradient surf clam), shrimp ebi (curled cooked prawn), tamago (golden sweet egg slice tied with seaweed band), maguro lean tuna, ikura gunkan (glistening salmon roe pearls). Dark mood lighting from above-left, shallow depth of field. Hyperreal Octane render, food photography style.

Vertical Taobao-style e-commerce product detail page (long-image format). Product: 'AIXZD Cyan Cap'. Top section: hero shot of a teal duckbill cap with embroidered 'AIXZD' logo, white background, centered, highlight reflections. Second section: 4-grid lifestyle photos of a young woman wearing the cap in different scenes (street style, cafe, gym, road trip). Third section: feature callouts with icons (Quick-dry fabric · UV protection UPF50+ · Adjustable strap · Embroidered logo · Unisex). Fourth section: size chart and color options (4 swatches). Fifth section: customer review screenshots with 5-star ratings. Bottom: 'Add to cart' CTA banner in red. Tightly packed editorial e-commerce layout, Chinese typography mixed with English, clean and professional.

Realistic screenshot of a YouTube livestream page. The video shows an OpenAI livestream where they introduce their first humanoid robot. The robot stands center stage on a minimalist white podium under bright stage lighting, with the OpenAI logo on a giant LED backdrop. Sam Altman stands to the right of the robot in a dark fleece, presenting. The video player shows live indicator (red 'LIVE' badge), 247K viewers count, progress bar, fullscreen and settings icons. Below the video: title 'OpenAI Unveils First Humanoid Robot — Live Demo', channel name 'OpenAI', verified checkmark, like button (84K), share, comments. Right sidebar shows recommended videos. Realistic Chrome browser frame on top.

Stylized satirical illustration of three fictional tech-and-politics-inspired characters working behind a busy movie theater concession stand, red-and-white striped uniforms, popcorn, soda fountain, bustling lobby, cinematic humorous poster style, clearly fictional, not a real photo.

High-quality scientific schematic diagram in Science/Nature journal style. Topic: PD-1/PD-L1 immune checkpoint pathway in the tumor microenvironment. Light gray-white background. Left side: a simplified T-cell cross-section, cell membrane drawn as a phospholipid bilayer (light blue), with PD-1 receptor protein (purple) protruding from the membrane; PD-1 intracellular tail connects via ITSM motif to downstream signaling molecule SHP-2 (green). Right side: a tumor cell with PD-L1 ligand (orange) on its surface. Center: PD-1 binding to PD-L1 with arrows showing inhibition signal flow. Below: alternate panel showing anti-PD-1 antibody (e.g., pembrolizumab) blocking the interaction, restoring T-cell killing function. Labeled annotations in clean sans-serif font, scale arrows, small molecular structure inset. Educational textbook clarity.

Clean technical diagram explaining the InstructGPT training process via RLHF (Reinforcement Learning from Human Feedback). Three-stage pipeline shown left to right with arrows. Stage 1 'Supervised Fine-Tuning (SFT)': pretrained GPT model + human-written demonstration data → fine-tuned SFT model. Stage 2 'Reward Model Training': SFT model generates multiple outputs → human labelers rank them → train Reward Model (RM). Stage 3 'PPO Reinforcement Learning': SFT model interacts with prompts → RM scores outputs → PPO algorithm updates the policy → final InstructGPT model. Use boxed model icons (rounded rectangles with model name labels), arrows with stage transitions, small data icons for input/output. Soft pastel color coding by stage (blue, green, orange). Style: research paper figure, minimal, axis-grid background, English labels.

Vertical 9:16 social-media long-image post in Xiaohongshu (Little Red Book) style. Topic: 'Niacinamide for Brightening — Does It Really Work?' Cream-beige background with deep green primary color and red highlights for emphasis. Top: serif bold title in deep green '成分党 / 烟酰胺真的能美白吗?' with 'really' crossed out by a red wavy line for tension. Subtitle: '5 studies tell you the truth · No traps'. Below are stacked card sections: (1) 'What is Niacinamide' — white rounded card with ~50 char explanation + small molecular icon. (2) '5 Key Studies' — table with year, journal, conclusion. (3) 'Real Results vs Marketing' — before/after comparison. (4) 'Best Concentration' — bar chart 2%/5%/10%. (5) Conclusion box. Editorial knowledge-card layout with consistent rounded rectangles, dense Chinese typography, professional yet approachable.

Vertical 3:4 Xiaohongshu-style outfit guide poster. Theme: 'Pear-Shape Body / 7 Outfit Formulas That Never Miss'. Cream background + deep brown text + olive green accents. Top: rounded handwritten title in deep brown with '7 Formulas' circled in olive green. Subtitle small text: '175cm slightly chubby blogger · 3 years of real-world testing'. Body: 7 outfit formula cards stacked in single column (each card: left = cropped outfit photo of a tall slim-faced model, right = formula text like 'V-neck top + high-waist A-line skirt + pointed-toe boots'). Each card has a small numbered tag. Bottom: 'Save this list' CTA in olive green. Soft warm photography, editorial blog feel, mixed Chinese/English typography.

A creative mashup screenshot titled 'SONG DYNASTY SOCIAL MEDIA FEED'. Mimics a modern smartphone social media interface (Twitter/X style) but every post is from a famous Song Dynasty Chinese figure. Avatars are classical Chinese ink paintings of the figures. Sample posts: User '苏东坡 SuShi_Official' avatar shows Su Shi portrait, post text: 'Just arrived in Huangzhou. Got demoted but feeling okay. Made Dongpo pork myself today, recipe in next post 🐷' with 4.2K likes. User '李清照 LiQingZhao' avatar of the female poet, post text: 'Drinking alone again tonight. Husband is away. New ci poem coming soon...' User '王安石 WangAnShi_PM' shares a policy infographic. User '岳飞 General_YueFei' posts a battle update. UI elements: like, retweet, comment counts in modern style. Status bar shows 'Bianjing 4G' instead of normal carrier. Whimsical anachronistic blend, photoreal mobile UI.