The night globe behind two numbers: 7,115,706 confirmed COVID-19 deaths, and 27,000,000 estimated from excess deaths.▶ Watch the film

A documentary · COVID-19

The Pandemic

From a market in Wuhan to a world that stopped, the waves, the count of the dead, and the question nobody has answered: where it came from.

7 minutes · English with subtitles · made with the Papermorph Skill

As shorts

The same film told upright for a phone, in six episodes of about a minute and a quarter.

As a scroll story

The same film as a page you scroll: the map changes under your thumb, step by step, and scrolling back rewinds it.

A step of the scroll story: Wuhan on the rail map of China at two in the morning on 23 January 2020.
Scroll through the pandemic23 steps, from Wuhan to the waves and the count. No sound, read at your own pace.Start scrolling ↓

What we set out to do

Our earlier documentary, Madness and Civilization, is pictures and words. This one tries the other kind of explainer, the kind Vox and Johnny Harris make: maps that move, numbers that land on the word that says them, documents pulled into frame and marked up. The subject was chosen for that: the pandemic happened on a map, day by day, and almost everything about it was counted.

Before any frame was made, Claude studied three of those videos cut by cut and wrote down how each transition works: what carries the eye from one picture to the next, where the camera goes, how long a number stays on screen. Those notes, not a style sheet, were the brief.

One map, all the way through

The film has one spine: a warm paper map of the world with a date chip in the corner. It comes back in every part with something new on it: one dot in Wuhan, then the train lines out of the city before the New Year, then the countries telling people to stay home, then the deaths per million, wave after wave.

The paper map zoomed in on Wuhan, with the chip December 2019. Railway lines leaving Wuhan on the paper map, with a clock at 2:38 a.m. on 23 January 2020.
The paper map in December 2019, then the night before Wuhan closed: trains still leaving at 2:38 a.m.

When the story leaves one place for the whole world, the map becomes a night globe (three.js, NASA's Black Marble city lights). Both are drawn in the same orthographic projection, so the globe can turn, push in, and become the paper map without a cut. The countries are Natural Earth outlines, projected once with d3-geo into small script files, so the finished page needs no server and no map service.

The night globe seen from above the North Pole, red arcs from Europe to New York and a grey one from Asia, under the line Most of New York's early cases came from Europe.
The globe turned over the pole, so Asia and Europe can both reach New York in one picture.

One source for every number

Every number on screen comes from one named source, written in small type under it: Our World in Data for cases, deaths, vaccines and stay-at-home orders; Wikipedia for events and dates; the WHO's own transcript for its words. The two do not always agree. Our World in Data counts by the day a death was reported, so it lags behind the news: by its count Italy passed China a few days after Wikipedia says it did. Where they disagree, the narration was written around the date rather than mixing the two.

Counters roll up as the voice says the number and stop exactly on it. The stay-at-home map counts the countries as the date runs, so when the narration says “more than a hundred and twenty”, the counter reads 121 and the date chip shows the first day that was true.

The WHO Director-General's remarks of 11 March 2020 on a paper card, with the word pandemic and the phrase cases in 114 countries highlighted.
The WHO's words of 11 March 2020, highlighted as they are spoken. A first version stamped PANDEMIC across the screen; the user turned it down, and the transcript replaced it.

A part at a time

It began as a 107-second sample: the cold open, Wuhan, the doctors who were silenced, and the trains. The user watched it, liked it, and asked for the rest of the same film, one part at a time, under ten minutes. Every part went through the same steps:

  1. Research. The Wikipedia articles and the Our World in Data tables, saved locally with their revision numbers.
  2. Narration, written in our own words, with a mark on every word that triggers something on screen.
  3. Voice. ElevenLabs v4, the voice “Roger”, sped up 1.1 times. The word times it returns place every mark.
  4. Beats, built in code and checked on contact sheets of stills before anything moved.
  5. Music. One track per part from Google's Lyria 3, levelled to the others, crossfading over two seconds so the music never stops.
  6. Review. The part exported as a video, and a fresh Claude, with no memory of making it, watching the frames as a director.
The world map shaded by deaths per million in April 2021, with a Delta pin on India, above two curves of cases and deaths a day.
The waves: one map and two curves that keep running, with a pin for each variant where it was first found.

What the reviews caught

Three director reviews watched the film from frames alone and checked every claim against the saved sources. They found things a maker misses: two different dates on screen at once; a vaccine rewind that stopped in November while the label said 2 December; three seconds of empty paper under a sentence about money; a question mark that sat on the market's label, so it read as “the market?” instead of “which one?”. Each finding came with a time, a reason and a fix, and the next round checked the fixes.

The user's own notes changed the film more. The opening first said the pandemic was “the biggest crisis of our generation, and we are already forgetting it”; the user kept only the first half, and the line was cut from the recording rather than voiced again. The join between the approved sample and the new parts looked like two films, so the sample's last picture now fades faster. A world map that stayed white while the death toll climbed beneath it got its own colour, deaths per million.

A grid of dots on black: the counted dead in solid red, the uncounted in faint outlines, under the words about 27 million, the likeliest estimate.
The count: one dot for 100,000 people. Seven million were counted; the likeliest estimate is closer to twenty-seven.

The question left open

The first outline left out where the virus came from. Near the end the user asked for it, with one rule: “don't favour zoonosis; treat them fairly.” The two answers get the same card, the same size and colour, one fact each and about the same time. The film says who leans which way: many scientists and the WHO-convened study towards a natural spillover, the FBI and the US Department of Energy towards a laboratory accident. It ends on what is known for certain: the early patient lists, swabs and blood samples were never fully shared.

The paper map of Wuhan with the Huanan Seafood Market and the Wuhan Institute of Virology pinned, and two identical cards: Nature, and A laboratory accident.
Two answers, drawn alike.

The numbers

7.4 min30 beats, 1,052 words of narration
$0.59voice, every take included (ElevenLabs v4 via OpenRouter)
$0.48music, six tracks of Lyria 3 at $0.08
29pictures and data files, each listed with its source and licence
3director reviews, besides one for each new part
0servers: the film is one static page

Everything else, the research, the script, the code for every beat and the reviews, was done by Claude (Opus) with the Papermorph Skill, with the user watching each cut and giving notes.

Sources and credits

Narration written from Wikipedia (“COVID-19 pandemic”, “COVID-19 pandemic in Italy”, “COVID-19 pandemic in New York City”, “Origin of SARS-CoV-2”) and Our World in Data (CC BY 4.0); the WHO quotations are from the Director-General's remarks of 11 March 2020 and the statement of 5 May 2023. Voice: ElevenLabs v4. Music: Google Lyria 3, made for this film. Type: Oswald and EB Garamond (SIL Open Font License, lib/fonts/). The seven doctors besides Li Wenliang, who are not identified, are stand-ins made with an image model and tagged as such on screen.

The shorts reuse the film’s pictures, maps and music, with a new script and voice for each episode. Pictures marked “AI illustration” on screen were made with an image model (gpt-image via OpenRouter); the breadline photograph is from the U.S. National Archives (public domain); the estimate of 27 million deaths is The Economist’s excess-death model. Every file of the shorts is listed with its source in SOURCES.md.

Pictures

Maps and data