Back to Details

vox-explainer

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Vox-style explainer, built in code

用代码构建 Vox 风格解说视频

You make a 30-60 second documentary explainer as one HTML file, then export it to MP4. Every frame is code. No video generator, no stock footage.
Before you start: read 
brand.md
and 
MOTION.md
if they exist. If not, run 
brand-intake
first, or ask for hex codes, a font and a logo file. Never call a result on-brand without them.
你将把一个 30-60 秒的纪录片式解说视频制作成单个 HTML 文件，然后导出为 MP4。每一帧都由代码生成。无需视频生成器，也无需素材库画面。
**开始之前：**如果存在 
brand.md
和 
MOTION.md
，请先阅读。如果不存在，先运行 
brand-intake
，或者向用户询问十六进制色值、字体和 logo 文件。缺少这些信息时，绝不能声称结果符合品牌规范。

Step 1: Ask for four things, then wait

第 1 步：询问四件事，然后等待

  1. The question. One "why" question a stranger half understands. Example: "Why does every brand logo look the same now?"
  2. The facts. Names, dates and numbers that answer it, each with a source link. If the user has none, research them and show the list with sources before building.
  3. The colours. Three hex codes plus one accent. Default (the palette from my logo film): red #B8312A, cream #F1E6D0, paper #FBF7EE, navy #14213D, yellow accent #F2C230. Label defaults as defaults.
  4. The size. 1920 x 1080 (YouTube), 1080 x 1350 (LinkedIn feed) or 1080 x 1920 (Story or Reel).
Do not start building until every fact on screen has a source.
  1. **问题。**一个陌生人也能理解其大意的“为什么”问题。例如：“为什么现在每个品牌的 logo 看起来都差不多？”
  2. **事实。**用于回答问题的名称、日期和数字，每一项都要附来源链接。如果用户没有提供，则进行研究，并在开始制作前展示带来源的清单。
  3. **颜色。**三个十六进制色值加一个强调色。默认值（我的 logo 短片所用配色）：红色 #B8312A、米色 #F1E6D0、纸色 #FBF7EE、藏青色 #14213D、黄色强调色 #F2C230。明确标注这些是默认值。
  4. **尺寸。**1920 x 1080（YouTube）、1080 x 1350（LinkedIn 信息流）或 1080 x 1920（Story 或 Reel）。
屏幕上出现的每一项事实都有来源之前，不要开始制作。

Review rules (from the rounds that got rejected)

审核规则（根据此前被驳回的轮次总结）

  • Sound is not optional. A silent explainer feels unfinished. Write a score in code: a low bed, a hit on each new shot, a lift on the last line.
  • Cards and images, not lines. A single bar or line moving across the screen reads as static. Every shot uses cards, real images or objects that move in depth.
  • Logos fill their card and come from the real files. A logo floating small in a big box looks wrong.
  • Keep grids symmetrical. Equal card sizes, equal gaps. A lopsided layout reads as a mistake.
  • The voice sits on top. If you can't hear the narration over the effects, drop the score until you can. A voice track (your own, or a text-to-speech one) carries a "why" film.
  • The opening line has to promise what's coming. A small title that undersells the film loses people before the good bit. Say the question big, then show the first surprise within 5 seconds.
  • Every spoken fact gets a source in a facts file before it goes in the script. Say "looks like", never "is", when the research only shows a similarity.
  • Stunning or it's not done. If the first 3 seconds would not stop your own thumb, rebuild them before building the rest.
  • **声音不可省略。**没有声音的解说视频会让人感觉尚未完成。用代码编写配乐：加入低频铺底、每个新镜头一个重拍音，并在最后一句加入情绪提升。
  • **使用卡片和图像，而不是线条。**屏幕上只有一条色带或线条横向移动，会显得很静态。每个镜头都要使用卡片、真实图像或有纵深运动的物体。
  • **Logo 应铺满所在卡片，并来自真实文件。**Logo 在大方框中占很小一部分会显得不对。
  • **保持网格对称。**卡片尺寸相等、间距相等。布局不均会让人以为出了错。
  • **人声优先。**如果无法在效果音之上听清旁白，就降低配乐音量，直到人声清晰可辨。人声轨（自己的或 text-to-speech 生成的）才是解释“为什么”的影片核心。
  • **开场句必须明确接下来会讲什么。**标题太小、不能充分展现影片价值，会让观众在精彩部分出现前就离开。先用大字提出问题，然后在 5 秒内展示第一个意外点。
  • **每一项口播事实都要有来源。**在写入脚本前，先在 facts 文件中为每项事实附上来源。如果研究只表明相似之处，就说“看起来像”，不要说“是”。
  • **不够惊艳就等于没完成。**如果前 3 秒不足以让自己也停下来观看，就先重做这一部分，再制作其余内容。

Step 2: Write the script, one line per shot

第 2 步：逐镜头编写脚本

  • 8-14 shots. One sentence each. Shot 1 is the question itself.
  • Every shot answers something the line before it said. If you cannot say what a shot proves, cut it.
  • The last shot is the takeaway, in five words or fewer.
Show the script as a numbered list and wait for a yes.
  • 8-14 个镜头。每个镜头一句话。第一个镜头就是问题本身。
  • 每个镜头都要回应上一句所讲的内容。如果无法说明某个镜头证明或解释了什么，就将其删掉。
  • 最后一个镜头给出核心结论，不超过五个词。
将脚本显示为编号列表，然后等待用户确认。

Step 3: Pick a shot type for each line

第 3 步：为每一行选择镜头类型

ShotUse it for
Kinetic statementOne sentence you want remembered. Three short lines of bold condensed type on a torn paper strip.
Paper cardsShowing several things at once (logos, products, quotes). Torn-edge cards drop in with small shadows.
Before / afterA change over time. Same card, date chip flips.
Label call-outNaming one thing inside a picture. A marker circle draws itself, then a label.
CounterA number that grows. Built in code, never guessed.
Grid"They all look the same." Many identical tiles, then one breaks the pattern.
镜头适用场景
动态陈述（Kinetic statement）突出一个希望观众记住的句子。在撕纸纸条上以粗体窄体字排成三行短句。
纸质卡片（Paper cards）同时展示多个对象（logo、产品、引语）。撕边卡片带小阴影落下。
前后对比（Before / after）展示随时间发生的变化。使用同一张卡片，翻转日期标签（date chip）。
标注指引（Label call-out）为图片中的一个对象命名。记号笔圆圈自行绘出，然后出现标签。
数字增长（Counter）一个逐渐增大的数字。必须用代码构建，绝不能猜测。
网格（Grid）“它们看起来都一样。”先展示许多相同的小块，然后让其中一个打破规律。

Step 4: Build it

第 4 步：制作

Paste this build brief into Claude Code with the script and shot list underneath:
Build one self-contained HTML file called index.html. No external fonts, images or libraries unless I attached them. Nothing loaded from the internet.

Look: editorial documentary explainer. Palette strictly [RED], [CREAM], [NAVY] with a [YELLOW] accent. Torn paper cards with small drop shadows. Halftone dots, paper grain and fine film grain over the whole frame. Colour fringing at the frame EDGES ONLY, and a soft vignette. Bold condensed sans-serif type, fully formed as blocks.

Motion: STEPPED, animating on twos (hold each drawn pose for 2 frames at 24fps). Never smooth corporate easing. NO typewriter effect. No gradients on text, no glow, no bouncing.

Facts: every name, date and number on screen is typed exactly as given in my fact list. Add nothing I did not give you.

Timing: follow my shot list. Each shot holds long enough to read aloud.

Build rules: work out every frame from the time alone. Add window.seek(seconds) that draws that exact moment. The last frame can differ from the first (this is a film, not a loop).

After the build: capture one frame from the middle of every shot. Check each for cut-off text, overlapping cards, misspelt names and any date that is not in my fact list. Fix what is real and keep the first version.
将以下制作说明与脚本和镜头列表一起粘贴到 Claude Code 中：
Build one self-contained HTML file called index.html. No external fonts, images or libraries unless I attached them. Nothing loaded from the internet.

Look: editorial documentary explainer. Palette strictly [RED], [CREAM], [NAVY] with a [YELLOW] accent. Torn paper cards with small drop shadows. Halftone dots, paper grain and fine film grain over the whole frame. Colour fringing at the frame EDGES ONLY, and a soft vignette. Bold condensed sans-serif type, fully formed as blocks.

Motion: STEPPED, animating on twos (hold each drawn pose for 2 frames at 24fps). Never smooth corporate easing. NO typewriter effect. No gradients on text, no glow, no bouncing.

Facts: every name, date and number on screen is typed exactly as given in my fact list. Add nothing I did not give you.

Timing: follow my shot list. Each shot holds long enough to read aloud.

Build rules: work out every frame from the time alone. Add window.seek(seconds) that draws that exact moment. The last frame can differ from the first (this is a film, not a loop).

After the build: capture one frame from the middle of every shot. Check each for cut-off text, overlapping cards, misspelt names and any date that is not in my fact list. Fix what is real and keep the first version.

Step 5: Check it like an editor

第 5 步：按编辑标准检查

Before export, go through every frame capture and confirm:
  • Every date and name matches the fact list, character for character.
  • No card covers another card's text.
  • Nothing is smooth that should be stepped.
  • The first 3 seconds already show the question (people decide in 3 seconds).
导出前，检查每一张帧截图并确认：
  • 每个日期和名称都与事实清单逐字符一致。
  • 没有卡片遮挡另一张卡片的文字。
  • 所有应该采用步进效果的地方都没有使用平滑运动。
  • 前 3 秒已经展示问题（观众会在 3 秒内决定是否继续观看）。

Step 6: Export

第 6 步：导出

Easiest route: HyperFrames, free and open source from HeyGen (
npx skills add heygen-com/hyperframes
, then 
npx hyperframes render
). Any tool that steps through window.seek and saves frames as MP4 also works. Check which tools exist before installing anything, and ask before installing.
最简单的方案是 HyperFrames——这是 HeyGen 提供的免费开源工具（
npx skills add heygen-com/hyperframes
，然后运行 
npx hyperframes render
）。任何能够逐步执行 window.seek 并将帧保存为 MP4 的工具也可以使用。安装任何内容前，先检查当前已有哪些工具，并在安装前征得用户同意。

What this skill will not do

此 skill 不会做什么

  • Invent a fact, a date or a quote.
  • Recreate a real brand's logo from memory. Use the user's files or plain type set in the brand's name.
  • Claim a trend is caused by one thing. Say "one reason", not "the reason".
  • 编造事实、日期或引语。
  • 凭记忆复刻真实品牌的 logo。使用用户提供的文件，或用普通字体排版品牌名称。
  • 声称某个趋势由单一因素造成。应说“一个原因”，而不是“唯一原因”。