三栏式 AI 图像质量对比
创建一个演示文稿风格的对比 Slides，展示三张带有模型或质量设置标签的逼真纽约街头广告牌照片。
gpt-image-2
中文提示词
目标：创建一个简洁的并排对比图，展示三张生成的城市广告牌街头照片样本，对比 ChatGPT Images 2.0 的质量设置与名为 {argument name="model codename" default="mona-lisa-1"} 的 checkpoint。 画布：宽屏 16:9 横向画布，采用灰白色/浅暖灰色背景，并留有充足的边距。横向排列 3 个垂直照片面板，每个面板均由淡米色边框包裹，如同演示 Slides 上的打印照片。 布局：使用三个间距均匀的纵向图像面板。每个面板包含一张逼真的纽约市街景，拍摄视角为人行道或路缘高度，背景为多云的日光、高楼、停放的车辆、交通流、店面招牌，以及场景上方的大型高架广告牌。在每个面板下方，添加一行深灰色/黑色无衬线字体的居中标题。 面板数量与标题：共 3 个面板。左侧标题：第一行“ChatGPT Images 2.0”，第二行加粗引号内的“MEDIUM”。中间标题：仅加粗引号内的“{argument name="model codename" default="mona-lisa-1"}”。右侧标题：第一行“ChatGPT Images 2.0”，第二行加粗引号内的“HIGH”。 照片面板细节：左侧面板显示一辆黑色 SUV 停在前景的城市街道上，左侧为类似唐人街的店面，上方有一个大型黑白广告牌。中间面板显示一辆黑色 Toyota RAV4 停在路边，背景为曼哈顿中城大道，远处可见帝国大厦，砖砌建筑上方安装有一个深色广告牌。右侧面板显示一辆黑色轿车停在路边，背景为时代广场风格的商业建筑和招牌，包括一个明亮的“I ♥ NY GIFTS”店面，以及一个带有青色发光边框的大型深色广告牌。 广告牌内容：每个广告牌应类似于 ChatGPT 的社交引用卡片，带有圆形头像、标题“What do you think?”、简短引文、小字号句柄行以及底部的 ChatGPT 标志/文字标识。保持广告牌文字基本可读，但符合街头实拍的自然感。使用引文主题 {argument name="billboard quote theme" default="opinions about AI and the future"}。右侧广告牌应呈现最清晰的高对比度外观并带有青色轮廓，中间广告牌应看起来适度逼真，左侧广告牌应看起来稍显柔和/带有噪点，以暗示质量差异。 视觉风格：嵌入极简演示对比 Slides 中的逼真智能手机街头摄影。自然的透视、轻微的镜头畸变、城市杂乱感、阴天、真实的道路污垢、停放的车辆和逼真的阴影。整体 Slides 应看起来像模型评估截图，而非精修的广告。 约束条件：使用 3 个图像面板和 3 个标题。不要添加额外的面板、图表、图标、水印或解释性段落。保持标题简洁并居中于照片下方。
原始提示词
Goal: Create a clean side-by-side comparison graphic showing three generated street-photo samples of urban billboard ads, comparing ChatGPT Images 2.0 quality settings with a checkpoint named {argument name="model codename" default="mona-lisa-1"}. Canvas: Wide 16:9 landscape canvas with an off-white/light warm gray background and generous margins. Arrange exactly 3 vertical photo panels in a row, each framed by a pale beige border like printed photographs on a presentation slide. Layout: Use three evenly spaced portrait-oriented image panels. Each panel contains a realistic New York City street scene photographed from sidewalk or curb level, with cloudy daylight, tall buildings, parked cars, traffic, storefront signage, and a large elevated billboard ad near the top of the scene. Beneath each panel, add a centered caption in dark gray/black sans-serif type. Panel count and captions: There are exactly 3 panels. Left caption: “ChatGPT Images 2.0” on the first line and “MEDIUM” in bold quotation marks on the second line. Center caption: “{argument name="model codename" default="mona-lisa-1"}” in bold quotation marks only. Right caption: “ChatGPT Images 2.0” on the first line and “HIGH” in bold quotation marks on the second line. Photo panel details: Left panel shows a black SUV parked in the foreground on a city street, Chinatown-like storefronts on the left, and a large black-and-white billboard above the storefront. Center panel shows a black Toyota RAV4 parked at the curb, a midtown Manhattan avenue with the Empire State Building visible in the distance, and a dark billboard mounted above a brick building. Right panel shows a black sedan parked at the curb, Times Square-style commercial buildings and signs including a bright “I ♥ NY GIFTS” storefront, and a large dark billboard with a cyan glowing border. Billboard content: Each billboard should resemble a ChatGPT social quote card with a circular avatar, the heading “What do you think?”, a short quote, a small handle line, and a ChatGPT logo/wordmark near the bottom. Keep the billboard text mostly legible but natural for a photographed street ad. Use the quote theme {argument name="billboard quote theme" default="opinions about AI and the future"}. The right billboard should have the clearest high-contrast look with a cyan outline, the center should look moderately realistic, and the left should look slightly softer/noisier to imply quality differences. Visual style: Realistic smartphone street photography embedded in a minimal presentation comparison slide. Natural perspective, mild lens distortion, urban clutter, overcast sky, authentic road grime, parked vehicles, and realistic shadows. The overall slide should look like a model evaluation screenshot rather than a polished advertisement. Constraints: Use exactly 3 image panels and exactly 3 captions. Do not add extra panels, charts, icons, watermarks, or explanatory paragraphs. Keep the captions clean and centered under the photos.