Explainers & Education · Nov 6, 2026
i recreated a video from 1.4 billion view youtube channel with one claude code prompt
i recreated a video from 1.4 billion view youtube… by Bajan. Watch the Explainers & Education video, explore the shared prompt and visit the original X post.
Bajan@Bajanpunkw
View on Xi recreated a video from 1.4 billion view youtube channel with one claude code prompt it's oversimplified, history cartoons 33 videos in 10 years and 10M subs mine took ~20 minutes and $0 prompt ↓ make me a 45–60 second animated video in the style of oversimplified about [your topic], built 100% in code. no ai images, no ai video 1. script - 10–12 short lines, one idea per line, a joke or a twist every 2 lines - line 1 is a hook that sets up a question or a myth. the last line pays it off, so the ending lands instead of just stopping - every fact must be true. check anything you're not sure about 2. voice - pip install edge-tts numpy. voice en-US-AndrewMultilingualNeural, rate +18% - generate each line separately with boundary="WordBoundary", cut the silence edge-tts adds at the start and end of each clip, join lines with 0.1–0.2s gaps - save voice.wav and timeline.json with every word's start/end. retry with backoff if a request fails 3. visuals - one html file, 1920x1080…
The prompt behind it.
Original prompt
make me a 45–60 second animated video in the style of oversimplified about [your topic], built 100% in code. no ai images, no ai video 1. script - 10–12 short lines, one idea per line, a joke or a twist every 2 lines - line 1 is a hook that sets up a question or a myth. the last line pays it off, so the ending lands instead of just stopping - every fact must be true. check anything you're not sure about 2. voice - pip install edge-tts numpy. voice en-US-AndrewMultilingualNeural, rate +18% - generate each line separately with boundary="WordBoundary", cut the silence edge-tts adds at the start and end of each clip, join lines with 0.1–0.2s gaps - save voice.wav and timeline.json with every word's start/end. retry with backoff if a request fails 3. visuals - one html file, 1920x1080 canvas, a function render(t) that draws frame t. one scene per line - things appear on the exact word that names them (use the word timings). every line needs at least one action, not a still picture with a caption - the camera never stops: slow push-in and drift in every scene, a quick punch-in on each scene change, zoom toward whatever the key word is about, a small shake on impacts. keep titles inside the frame when zoomed - characters always idle (breathing, blinking). nothing on screen stays frozen for more than ~2 seconds - the first 0.5s is an event (something drops in, explodes, slams), not a calm intro - style: flat colors, round-headed characters with dot eyes and thick dark outlines, oversized hats, a yellow year badge in the corner, big bold labels that pop in, simple maps from world-atlas topojson with countries filling in color, quick pop-in animations, a comic military march under the voice 4. captions - word by word at the bottom, current word highlighted, big bold font, max ~20 characters per chunk 5. sound (all synthesized in python, no samples) - a quiet music bed that fits the style, ducked under the voice, stops on a short final chord - sound effects on the actions: pops when things appear, a low thump for landings, a cannon/boom for explosions, a soft note on scene changes - use clean tones. if you need noise (explosions, wind), run it through proper low-pass filters. no raw white noise, no buzzy saw waves - the voice is always the loudest thing: voice peaks at -3 dBFS, every effect is set relative to it and sits under it, the final chord no louder than the voice. low-pass music+effects around 4.5 kHz 6. render - puppeteer loads the page, calls render(i/30) for every frame, pipes screenshots into ffmpeg (libx264, 30fps), then muxes the mixed audio 7. check before you show me - render ~10 stills from different moments, look at them and fix overlaps, cut-off text, empty space and anything unreadable - after the full render: no silence longer than 0.5s, no part of the video where nothing moves for 2+ seconds. fix and re-render if there is
Video and prompt by Bajan. Original post ↗


