四格 AI 风景对比图

目标：制作一张宽幅社交媒体对比图，将四张 AI 生成的动漫风景壁纸并排展示，在保持构图和图像模型不变的前提下，对比不同提示词模型等级的效果。 画布：宽幅 2:1 横向图片，尺寸约为 1200×620 像素，深炭灰色背景。顶部添加黑色标题栏，下方排列四个垂直的长方形图像面板，面板间以窄黑边分隔。 标题文字：左上角使用大号日文标题：「同じ構図・同じ画像AI（GPT Image 2）。プロンプトを“書いたAI”だけが違う4枚」。下方配以较小的灰色副标题：「変えたのは『プロンプトを書くClaudeのモデル階層』だけ（左＝最も安い → 右＝最上位）」。 布局：使用 4 个等宽的垂直面板，从左至右排列。每个面板顶部设有一个深色标签条，配以白色文字。四个标签分别为：1) 「haiku｜最も安い」、2) 「sonnet」、3) 「opus」、4) 「fable｜最上位」。 面板图像：四个面板均描绘相同的动漫电影级构图：一个孤独的小人影站在长满草的山脊上，俯瞰广阔的山谷，远处是层叠的山峦、戏剧性的云层和金色的阳光。每个面板都要精美，但在艺术风格上略有不同。 面板 1 (haiku)：温暖的日落奇幻景观，橙粉色天空，太阳接近地平线，紫蓝色山峦，山谷间有薄雾，前景有野花和草地，一个微小的深色人影位于下三分之一处居中，风格精致但较为常规。 面板 2 (sonnet)：更克制的自然光影和大气透视，蓝金色的天空中漂浮着巨大的纹理云团，山谷中弥漫着柔和的光晕，一个微小的人影站在山脊小路的尽头，具有真实的深度感和柔和的暖光。 面板 3 (opus)：更具电影感的构图，云层翻涌，低垂的太阳位于右侧地平线，山谷中有发光的湖泊或河流，一个微小的人影位于底部中心附近的山脊上，色彩平衡经过精炼，前景草地清晰，具有自然的空气透视感。 面板 4 (fable)：更明亮的动漫电影剧照感，青蓝色天空，高耸的积雨云，阳光穿透云层，郁郁葱葱的绿色山谷和丘陵，一个穿着白色连衣裙、戴着遮阳帽的小人影在草坡上移动或转身，即使在小比例下，服装和姿态也清晰可见。 视觉风格：高端动漫背景艺术，具有绘画感且细节丰富，新海诚风格的光影，电影般的氛围，柔和的辉光，优雅的色彩分级。保持对比图简洁易读，使用细边框和统一的面板尺寸。 可自定义细节：标题中使用 {argument name="image model name" default="GPT Image 2"}，对比 {argument name="first model label" default="haiku｜最も安い"}、{argument name="second model label" default="sonnet"}、{argument name="third model label" default="opus"} 以及 {argument name="fourth model label" default="fable｜最上位"}。 约束：包含 4 个面板，总共 4 个微小的人影，每个面板一个。除标题、副标题和四个面板标签外，不要添加任何徽标、水印、额外图表、图标或其他说明文字。

Goal: Create a wide social-media comparison graphic showing four AI-generated anime landscape wallpapers side by side, comparing prompt-writing model levels while keeping the same composition and image model. Canvas: Wide horizontal 2:1 image, approximately 1200×620 px, dark charcoal background. Add a black header band across the top and four tall vertical image panels beneath it, separated by narrow black gutters. Header text: Large Japanese title at the top left: 「同じ構図・同じ画像AI（GPT Image 2）。プロンプトを“書いたAI”だけが違う4枚」. Smaller gray subtitle underneath: 「変えたのは『プロンプトを書くClaudeのモデル階層』だけ（左＝最も安い → 右＝最上位）」. Layout: Use exactly 4 vertical panels, equal width, arranged left to right. Each panel has a dark label strip at the top with white text. The four labels are exactly: 1) 「haiku｜最も安い」, 2) 「sonnet」, 3) 「opus」, 4) 「fable｜最上位」. Panel imagery: All four panels depict the same basic anime cinematic composition: a tiny solitary human figure on a grassy mountain ridge, overlooking a vast valley, distant layered mountains, dramatic clouds, and golden sunlight. Make each panel beautiful but subtly different in art direction. Panel 1, haiku: Warm sunset fantasy landscape, orange and pink sky, sun near the horizon, purple-blue mountain layers, mist in the valleys, wildflowers and grass in the foreground, one tiny dark standing figure centered near the lower third, highly polished but conventional. Panel 2, sonnet: More restrained natural lighting and atmospheric perspective, huge textured cloud mass in a blue-and-gold sky, soft glowing valley haze, one tiny standing figure at the end of a ridge path, realistic depth and gentle warm light. Panel 3, opus: Stronger cinematic composition with a sweeping cloud formation, low sun at the right horizon, glowing lake or river in the valley, one tiny figure on a grassy ridge near the bottom center, refined color balance, crisp foreground grasses, natural aerial depth. Panel 4, fable: Brighter anime film still feeling, turquoise-blue sky, towering cumulus clouds, sunbeams breaking through, lush green valley and hills, one small but clearly animated person in a white dress and sunhat moving or turning on the grassy slope, clothing and pose readable despite small scale. Visual style: High-end anime background art, painterly but detailed, Makoto Shinkai-inspired light, cinematic atmosphere, soft bloom, elegant color grading. Keep the comparison graphic clean and readable, with thin gutters and consistent panel sizing. Customizable details: Use {argument name="image model name" default="GPT Image 2"} in the title, compare {argument name="first model label" default="haiku｜最も安い"}, {argument name="second model label" default="sonnet"}, {argument name="third model label" default="opus"}, and {argument name="fourth model label" default="fable｜最上位"}. Constraints: Include exactly 4 panels and exactly 4 tiny human figures total, one per panel. Do not add logos, watermarks, extra charts, extra icons, or additional captions beyond the header, subtitle, and four panel labels.