Skip to content

Explainers (3Blue1Brown style)

You get an animated concept explainer built from a topic: a script you approve, AI narration, colour-coded animated scenes, burned English and Chinese subtitles plus .srt files, transitions and a quiet music bed. Make it as a 5–12 minute 16:9 film for YouTube or B站, or as a 60–120 second vertical short for 小红书, 抖音, Shorts or TikTok.

Contact sheet of a 98-second vertical explainer about CUDA: animated diagrams with English and Chinese captions

  • A topic, and any notes you want covered. An existing SCRIPT.md works too.
  • Node with npx hyperframes working. Scenes are HTML compositions rendered to video by HyperFrames. Without it, this recipe can’t render.
  • A narration voice. Reelfold uses the voice this computer has: your persona’s, OpenAI when its key is set (about $0.015 per minute), else the free offline voice on Apple silicon. See Voices for narration.
  • Fonts for the subtitles: ./install.sh downloads open-licensed ones (Noto Sans SC, STIX Two Text).
  • An AI model. The animated scenes are drawn by your AI (Claude Code, Codex, an API key or a local model). Without one, the project waits in Needs you (“The visuals need an AI model”) rather than rendering a text-only video.
  • Optional: a music file of your own for the bed.

Starting from notes or a Notion page instead of a topic? See Course videos from notes.

  1. On Home, describe the explainer in the request box: the topic, who it’s for, long or short, and where it goes.
  2. On Autopilot it runs to a finished video, and every choice is listed under What the AI decided. With Ask me first, the plan shows the explainer recipe, the platforms (the first one sets the canvas), the voice and long or short; adjust it in plain words, and the project stops in your Inbox at each decision:
    • Lock the script. One spoken English line per scene, with its Chinese subtitle. Numbers are written the way they are spoken.
    • Approve the storyboard. One frame per scene with its timing and motion.
    • Pick the voice. Choose the voice and speed and see the cost estimate before any paid narration runs.
    • Check the subtitle pairs. English chunks paired with their Chinese, with questionable pairs flagged.
  3. Each scene is drawn by the AI on the voice’s cue times, checked and kept only if it plays inside the video. Then it is rendered and exported per platform with a cover and post copy.
  4. Follow it on the live run page: the drawn scenes show up under How it is being made as they appear.

Not happy with the scenes? Press Redo the visuals on the clip page, or tell the chat (“the visuals are too plain”, “more diagrams”). Every scene is drawn again and the video is rendered again; the voice and captions stay. A scene that fails is kept as before, with Redraw scene N. Details: Course videos from notes.

Ask for an explainer and the skill follows workflows/explainer. Claude asks once about length, destination and subtitle layout, then works in stages and stops at each one for you:

  1. Script (SCRIPT.md), with every number in the worked example checked first.
  2. Sketches: one still per scene at its most complete moment, drawn with the real fonts and colours.
  3. Voice: a 10–15 second sample, then the full narration. Every take is transcribed and compared with the script, and a take that dropped a sentence is regenerated.
  4. Bilingual subtitles aligned to the narration.
  5. Scenes in HyperFrames, checked with snapshots, then a preview link. It renders only after you approve.

To deliver one render to several platforms:

Terminal window
python3 -m vstudio.export renders/short.mp4 --platforms xiaohongshu:full,douyin,youtube-shorts \
--out exports/ --cover cover.png
16:9 explainer Vertical short
Length 5–12 min, 12–18 scenes 60–120 s, 5–8 scenes of up to about 14 s
Shape hook → intuition → one worked example → variants → recap a hook in the first 2 seconds → one core picture → one worked example → a one-line payoff
Voice about 175 words per minute about 10 % faster, shorter first sentence
Canvas 1920×1080 9:16 (xiaohongshu:full, douyin, youtube-shorts, tiktok) or 3:4 (xiaohongshu:vertical)

One 9:16 build can ship to 小红书 full-screen, 抖音, Shorts and TikTok. A 3:4 version needs its own build. The 16:9 film is never reframed into 9:16, because the maths and captions would be cut off.

Option Default What it does
Mode long long (5–12 min) or short (60–120 s).
Platforms YouTube, B站 The first platform decides the canvas and the caption box.
Voice auto The voice this computer has: persona, OpenAI with a key (cedar), else the free voice.
Voice speed 1.0 0.7–1.4. Shorts usually run at 1.1.
Render quality standard draft, standard or high.
Music bed auto A calm bed chosen for the topic, ducked under the voice, or your own file.
  • “Make a 3Blue1Brown-style explainer on how model quantization works, about seven minutes, for YouTube and B站.”
  • “Vertical short explaining how CUDA runs on a GPU, under two minutes, English narration with Chinese subtitles.”
  • “Use my notes in this file as the outline. Keep one worked example with the same numbers all the way through.”
  • “Scene 4 feels crowded. Make the graphic bigger and move the result line down.”