GeoPT 物理模拟概览 Slides
一个结构化提示词，用于生成精美的学术 Slides，通过三列式视觉流程图解释用于物理模拟的 GeoPT 预训练方法。
gpt-image-2 · 信息图 / 教育视觉图 · 图表 · 插画 · 文本 / 排版 · 极简主义
中文提示词
目标：创建一个简洁的 16:9 学术概览 Slides，解释 {argument name="paper concept" default="GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training"}，风格采用精美的会议演示信息图表。 画布：宽屏白色幻灯片，海军蓝排版和点缀，极简阴影，高分辨率科学插图风格。底部使用一条带有小型圆形端点的细蓝色页脚线。 标题：顶部为醒目的深海军蓝大标题：{argument name="headline text" default="GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training"}。正下方为较小的蓝色副标题：{argument name="subtitle text" default="Wu 等人 (2026) 的直观概览"}。 核心理念横幅：在副标题下方，放置一个带有蓝色轮廓和极浅蓝色填充的圆角矩形标注框。左侧是一个蓝色圆形灯泡图标。文字内容为：“核心理念：”加粗蓝色字体，后接“在利用稀缺的真实模拟数据之前，使用丰富的 3D 几何图形和合成运动来预训练神经模拟器。” 主要布局：一个包含 4 行的三列式流程图，行与行之间由粗蓝色箭头从左至右连接。列标题分别为：“3D 几何图形”、“物理模拟（示例）”和“场（压力 / 速度 / 应力）”。左侧设有 4 个行标签：1) 汽车，2) 飞机，3) 船体，4) 碰撞 / 形变。 行细节：第 1 行显示一个灰色 3D 汽车模型，接着是带有蓝色流线和彩色涡流的汽车气流模拟，并附带一个小型的“t →”时间标记，然后是一个带有彩虹热力图场的汽车，以及一个标有“压力”且顶部为高、底部为低的垂直颜色图例。第 2 行显示一个灰色飞机模型，接着是带有气流流线和彩色尾迹的飞机，并附带“t →”，然后是一个带有标有“速度”高/低图例的彩虹热力图飞机。第 3 行显示一个灰色船体模型，接着是船体在海浪中移动的模拟水流，并附带“t →”，然后是一个带有标有“压力”高/低图例的彩虹热力图船体。第 4 行显示一个用于碰撞/形变的受损灰色车身，接着是带有碎片、形变和内部彩色应力/场可视化的碰撞模拟，并附带“t →”，然后是一个带有标有“应力”高/低图例的彩虹热力图形变车身。 底部说明栏：包含 3 个由虚线蓝色箭头连接的圆形图标及说明模块。模块 1 图标为线框立方体，标题为“丰富的几何图形”，说明为“来自 CAD 存储库的易获取 3D 形状”。模块 2 图标为风/流线，标题为“合成动力学”，说明为“可扩展的模拟产生多样化的运动和场”。模块 3 图标为神经网络图，标题为“更好的物理模拟器”，说明为“在有限真实数据下泛化能力更强的预训练模型”。 页脚：居中的小型灰色引用文字：{argument name="source citation" default="来源论文：Haixu Wu 等人, arXiv:2602.20399 (2026)"}。 视觉风格：清晰的矢量幻灯片构图与逼真的 3D 渲染相结合，采用蓝色学术配色，白色背景，精确对齐，无杂乱元素，无水印，除了指定的数量外，不添加额外的行或图标。
原始提示词
Goal: Create a clean 16:9 academic overview slide explaining {argument name="paper concept" default="GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training"}, styled like a polished conference-deck infographic. Canvas: Widescreen white slide, navy-blue typography and accents, minimal shadows, high-resolution scientific illustration style. Use a thin blue footer line with small circular endpoints. Header: Large bold dark-navy title at the top: {argument name="headline text" default="GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training"}. Directly underneath, smaller blue subtitle: {argument name="subtitle text" default="An intuitive overview of Wu et al. (2026)"}. Big idea banner: Under the subtitle, place a rounded rectangle callout with a blue outline and very pale blue fill. On the left is a blue circular lightbulb icon. Text reads: “Big idea:” in bold blue, followed by “use abundant 3D geometry and synthetic motion to pre-train neural simulators before using scarce real simulation data.” Main layout: A three-column pipeline with exactly 4 rows, connected left-to-right by thick blue arrows. Column headings are: “3D Geometry”, “Physics Simulation (examples)”, and “Fields (pressure / velocity / stress)”. The left side also has 4 row labels: 1) Car, 2) Aircraft, 3) Ship hull, 4) Crash / Deformation. Row details: Row 1 shows a gray 3D car model, then a car in airflow simulation with blue streamlines and colorful vortices plus a small “t →” time marker, then a car colored with a rainbow heatmap field and a vertical color legend labeled “Pressure” with High at top and Low at bottom. Row 2 shows a gray aircraft model, then an airplane with airflow streamlines and colored wakes plus “t →”, then a rainbow heatmap aircraft with a legend labeled “Velocity” High/Low. Row 3 shows a gray ship hull model, then a ship hull moving through ocean waves with simulated water flow plus “t →”, then a rainbow heatmap ship hull with a legend labeled “Pressure” High/Low. Row 4 shows a damaged gray car body for crash/deformation, then a crash simulation with debris, deformation, and internal colored stress/field visualization plus “t →”, then a rainbow heatmap deformed car body with a legend labeled “Stress” High/Low. Bottom explanation strip: Include exactly 3 circular icon-and-caption blocks connected by dotted blue arrows. Block 1 icon is a wireframe cube and title “Abundant geometry”, caption “Easily available 3D shapes from CAD repositories”. Block 2 icon is wind/flow lines and title “Synthetic dynamics”, caption “Scalable simulations produce diverse motion and fields”. Block 3 icon is a neural network graph and title “Better physics simulators”, caption “Pre-trained models that generalize better with limited real data”. Footer: Centered small gray citation text: {argument name="source citation" default="Source paper: Haixu Wu et al., arXiv:2602.20399 (2026)"}. Visual style: Crisp vector-slide composition mixed with realistic 3D renders, blue academic palette, white background, precise alignment, no clutter, no watermark, no extra rows or icons beyond the specified counts.