How these explainers are made: each film is a function of time on a web canvas, captured frame by frame in a headless browser and encoded with ffmpeg; the music is synthesised with numpy from sine waves and filtered noise, its sections and hits aligned to the caption timetable. The animation is written by Claude Code, and each episode passes four steps: animate, three reviewers, fix, verify. The Chinese vertical series first used a diorama look; because it was not engaging enough, three new samples were made and the night model was chosen. Two sets of style trials are included.
Where the pictures come from
Each film is a canvas on a web page, and any frame depends only on time, with no state carried between frames. For rendering, a headless browser captures the frames in parallel chunks and ffmpeg encodes them once, with key frames placed where the music changes section so that dark pictures do not flicker.
Captions and pacing run from the same timetable: the music is at 112.5 BPM, one beat is 16 frames, each caption stays at least 4 + characters/2.6 beats (half a beat more with a number), only one thing appears or changes on screen during a caption, and no word appears on screen that the caption has not said.
The music
There are no samples: piano, plucks, pads and drums are synthesised with numpy from sine waves and filtered noise. Each series gets its own piece: a train-like groove for the regional economics vertical films, felt piano for the Japanese classroom versions, an electronic pulse for the policy series. Sections and hits are read directly from each episode’s timetable, so the beat on which a title lands is a drum hit.
How each episode is made and checked
The animation, build scripts and checking tools are written by Claude Code. An episode goes through four steps: animate the whole episode, have three reviewers watch it (a director and pacing checker, a first-time viewer, and a reviewer set up to reflect the creator’s taste), apply the required fixes, then verify that every frame draws without error, captions match the script word for word, every number comes from the script’s number table, elements stay within the phone’s safe area, and nothing moves faster than the limit.
Style trials
The first regional economics compilation began by trying eight ways of drawing:
The Chinese vertical regional economics series then used a diorama look: small isometric models seen at an angle, information panels floating above them, a new board for each section. The creator found it not interesting or engaging enough, and pointed instead to the project’s first film, a game-theory short about two ice-cream stands on a beach: one continuous little world in which, when a stand moves, the dividing line and the scoreboard change with it. So three samples were made of the same stretch of the first episode, with the same captions and music, alongside the original diorama version:
Two of the three reviewers chose the night model: it carries over the game-theory short’s approach, with one glowing railway running through the whole film and a taxi-meter readout that reacts live to the picture. The final design uses the night model as the base, borrows the weighted red chips and the pinned scoreboard from the board game, and the two-column before-and-after comparison from the doodle. The five episodes rebuilt on these rules are the vertical films of the regional economics series in the Research section.
Rules kept from feedback
- One point per episode, with one surprise; skip what viewers already know.
- Spoken wording in captions, no clipped machine-like phrasing, and group nouns rather than gendered pronouns for groups.
- One thing on screen at a time, and slower rather than faster.
- No narrator avatar or mascot.
- Numbers only from the script’s number table; caveats and sources go in the pinned comment or the article.