<inputs>
Ask me for: my product, the exact request a user types into it, 6 to 8 photos for the result cards, a royalty-free song around 130 BPM with a clear drop (Mixkit, free for commercial use), and my brand mark for the end card. If I skip anything, use the defaults: the food-ordering story below, food photos from Pexels (fire noodles, birria tacos, hot chicken, buffalo wings, smash burger, spicy ramen), and Mixkit "Cat Walk".
</inputs>
<direction>
A 15-second product launch film, 1920x1080 at 60fps. 2D only, one continuous take. No fades, blurs or cuts: each scene grows from the last. Text rises from a mask line, buttons expand into pages, and color floods from one object before shrinking into another. Give the camera one eased move per scene, and never zoom in and out consecutively.
Look: warm white canvas (#f5f5f2), black UI (#0e0e10), lime accent (#cdf24f), orange (#ff6a2a), and a blue send button (#2f6bff). Add an Apple Intelligence-style iridescent glow around the input: a blurred conic gradient rotating through pink, orange, lime, cyan and violet. Use real food photography on dark cards, Geist for UI and a serif for the product name on the end card.
Banned: crossfades, blur-ins, 3D, particles, camera moves that reverse direction, holds longer than 1s, anything that feels like a template.
</direction>
<structure>
130 BPM, 32 beats (14.8s). Start the song 12 beats before its drop.
Beats 0-6, ask: push slowly into the glowing composer, about 3x zoom. Type the request over three fixed lines with a natural keystroke rhythm:
"I'm really hungry, I don't know
what I want but something
spicy, crispy and filling"
Beat 6, send: press the send button. A lime ripple spreads from it and transforms the message into a chat bubble. Pull back once to reveal the phone around it (428x900 screen, 11px black bezel, 9:41 status bar).
Beats 6-12, think: the send button grows into an orange wok, a half-disc with a thin handle. It tosses 6 colorful ingredients on beats 8, 9, 10 and 11 beneath the shimmering text "Finding something for you…".
Beat 12, the drop: a lime circle expands from the wok to fill the frame in about 0.35s. Dark food cards (330x440, radius 28, photo on top, name, one-line description, price and rating) burst in from the center. Slide the strip across three cards and land on "Fire Noodles", which lifts with a "Best match" pill.
Beats 16-21, chat: the lime contracts back into the user's bubble, carrying its words as the chosen card flies into its slot. "Found it. Fire Noodles from Wok & Fire, smoky, spicy and seriously filling." rises in, followed by "Want me to order it for you?" Then show two pills: "Order now" in black and "More options" outlined. The cursor clicks Order now → spinner → "✓ Ordered".
Beats 21-32, outro: the black button expands into the page, staying pill-shaped as it reaches the edges while the camera pushes into it. Its label grows to headline size, then slides away. Bring in "You were craving noodles." and "Now it's on the way." word by word (92px, white on black). A courier dot follows a route to a pin. The pin becomes the brand mark, and the wordmark wipes out from behind it, with "made with" above and the product name below.
</structure>
<build>
1. One HTML file, 1920x1080. Compute every style from time inside seek(t): no CSS transitions, timers or state carried between frames.
2. Use one camera transform on a container, keyed by [time, zoom, x, y] with eased segments and zoom interpolated in log space.
3. Use closed-form step-response springs for every pop and settle.
4. Make every handoff a shared element: the lime flood carries a copy of the bubble's words, and the button carries its label into the page.
5. Sound: use a downloaded Mixkit SFX for each event - typing, send pop, each toss, drop impact, card swoosh, click, success tone, page-fill whoosh and mark sparkle. Place each by its measured peak and loudnorm to -14 LUFS.
6. Render with Playwright at 60fps using 8 subframes per frame, blended with ffmpeg tmix. Review every fast moment frame by frame and scan the full video frame by frame.
</build>
中文译文
<inputs>
向我询问：我的产品、用户输入到产品中的确切请求、用于结果卡片的 6 到 8 张照片、一首约 130 BPM 且带有明确 drop 的免版税歌曲（Mixkit，可免费用于商业用途），以及用于结尾卡片的我的品牌标识。如果我跳过任何内容，请使用默认值：下方的点餐故事、来自 Pexels 的食物照片（火鸡面、birria 塔可、辣炸鸡、水牛城鸡翅、smash 汉堡、辣味拉面），以及 Mixkit 的 "Cat Walk"。
</inputs>
<direction>
一支 15 秒的产品发布短片，1920x1080、60fps。只用 2D，一个连续长镜头。不要淡入淡出、模糊或剪切：每个场景都从上一个场景生长出来。文字从一条遮罩线上升起，按钮扩展成页面，颜色从一个物体涌入另一个物体再收缩进去。每个场景给相机一次缓动的运动，并且绝不要连续地拉近再拉远。
外观：暖白色画布（#f5f5f2）、黑色 UI（#0e0e10）、青柠色强调色（#cdf24f）、橙色（#ff6a2a），以及一个蓝色的发送按钮（#2f6bff）。在输入框周围添加 Apple Intelligence 风格的虹彩光晕：一个模糊的锥形渐变，依次旋转经过粉色、橙色、青柠色、青色和紫罗兰色。在深色卡片上使用真实的食物摄影，UI 使用 Geist 字体，结尾卡片上的产品名称使用衬线字体。
禁止：交叉淡化、模糊进入、3D、粒子、反向移动的相机运动、超过 1s 的停留、任何像模板一样的效果。
</direction>
<structure>
130 BPM，32 拍（14.8s）。在 drop 前 12 拍开始播放歌曲。
Beats 0-6，提问：缓慢推近发光的输入框，约 3 倍缩放。以自然的按键节奏，在三条固定的行上输入请求：
"I'm really hungry, I don't know
what I want but something
spicy, crispy and filling"
Beat 6，发送：按下发送按钮。一圈青柠色涟漪从它扩散开，并把消息变成一个聊天气泡。拉回一次，露出它周围的手机（428x900 屏幕、11px 黑色边框、9:41 状态栏）。
Beats 6-12，思考：发送按钮长成一口橙色炒锅，一个带细手柄的半圆盘。它抛起 6 种色彩丰富的食材，分别在第 8、9、10 和 11 拍，在闪烁的文字 "Finding something for you…" 下方。
Beat 12，drop：一个青柠色圆圈从炒锅扩散开来，在约 0.35s 内充满画面。深色食物卡片（330x440、圆角 28、照片在上方，包含名称、一行描述、价格和评分）从中心迸出。将卡片条横向滑过三张卡片，停在 "Fire Noodles" 上，它随即抬起，并带有一个 "Best match" 胶囊标签。
Beats 16-21，聊天：青柠色收缩回用户的气泡中，带着它的文字，同时选中的卡片飞入它的槽位。"Found it. Fire Noodles from Wok & Fire, smoky, spicy and seriously filling." 升起，接着是 "Want me to order it for you?"。然后显示两个胶囊按钮：黑色的 "Order now" 和描边的 "More options"。光标点击 Order now → 加载转圈 → "✓ Ordered"。
Beats 21-32，尾声：黑色按钮扩展成页面，在抵达边缘时仍保持胶囊形，同时相机向它推近。它的标签放大到标题尺寸，然后滑走。逐词引入 "You were craving noodles." 和 "Now it's on the way."（92px，黑底白字）。一个快递员圆点沿路线移动到一个图钉。图钉变成品牌标识，字标从它背后擦除显现，上方是 "made with"，下方是产品名称。
</structure>
<build>
1. 一个 HTML 文件，1920x1080。在 seek(t) 内根据时间计算每一个样式：不使用 CSS 过渡、定时器，也不在帧之间保留状态。
2. 在一个容器上使用单个相机变换，以 [time, zoom, x, y] 作为关键帧，使用缓动分段，并在对数空间中插值 zoom。
3. 对每次弹出和回稳使用闭式阶跃响应弹簧。
4. 让每次交接都成为共享元素：青柠色漫流携带着气泡文字的一个副本，按钮把它的标签带入页面。
5. 声音：为每个事件使用下载的 Mixkit 音效——打字、发送弹出、每次抛起、drop 撞击、卡片嗖声、点击、成功提示音、页面充满的呼啸声和标识闪烁声。按各自测得的峰值放置，并用 loudnorm 归一化到 -14 LUFS。
6. 使用 Playwright 以 60fps 渲染，每帧使用 8 个子帧，并用 ffmpeg tmix 混合。逐帧检查每个快速瞬间，并逐帧扫描整个视频。
</build>
