Animated explainer videos
Apply these rules when you plan, write, animate or review a video. Story rules come first: a beautiful video with a flat script still fails.
Making the video
When the user wants an actual video file, build it with Remotion (and Three.js for 3D) from the template in this skill.
- Story first. Write the script as beats (one idea per beat) using the Story rules below. Check the first sentence and the opening seconds before anything else.
- Project. Copy
template/to a new folder, runnpm install. Readreference/remotion.mdbefore writing code, andreference/threejs.mdbefore any 3D scene. - Voice. If the user gives a voiceover (their recording or any text-to-speech file), put it in
public/and set"voice"inpublic/plan.json. If they don't, put each line in its scene's"text"and runnpm run voice(free Kokoro voice, English): it writespublic/voice.wavand sets every scene's start and end. Pick a voice withnode scripts/voice.mjs am_michael(male) or the defaultaf_heart(female). - Timing. Run
npm run captionsfor word timestamps. With your own voice file, start each scene at the first word of the line it shows; withnpm run voicethis is already done. - Scenes. Write one component per scene in
src/scenes/, register it, and list the scenes inplan.json. Every spoken line gets its own visual. - Check. Render stills (
npx remotion still Main out/still.png --frame=N) at the start, middle and end of each scene, look at them, and fix anything that breaks the rules below. - Render.
npm run renderwritesout/video.mp4. Watch the first 3 seconds again before calling it done.
Requirements: Node.js 18+ and ffmpeg. Remotion downloads its own headless Chrome on the first render.
Story
- Every video is a small story: a character wants something, something blocks it, a twist happens, there is a payoff. A list of facts is not a story.
- The first sentence is a bold claim, a contradiction of something people believe, or a character with a problem.
- Good: "Something in your kitchen is killing you." / "I always thought computers were better at math than people." / "This is Uncle Don, and he's got a huge problem."
- Bad: "How does X work?" / "Here are 3 tools that…" / "Today we'll look at…". If the video answers a "how" question, move that question to the second sentence.
- Open a question or a tension in the first 3 seconds and keep the answer for later. Don't spend the twist in the hook.
- Pick broad or timely topics. A niche topic needs an everyday angle that anyone can feel.
- A strong opener doesn't save a weak topic. Decide the topic first, then write the hook.
- Every video stands on its own. No "Part 2", no openers that continue an earlier video, no generic utility lists.
- Show, don't tell: every spoken line has its own visual. If the video makes sense on mute, the scenes work.
- One idea per scene. If a line holds two ideas, split it into two scenes.
- Escalate: each beat raises the stakes or the surprise a little more than the one before.
- End on a line the viewer can repeat or use (a question to ask, a rule of thumb), not on a summary.
- Length is a free choice: use the length the story needs. In long videos, give a small payoff at least every 60-90 seconds.
Framing (vertical 1080x1920 unless stated)
- Keep the main subject in the upper-middle band, about y = 500-1100.
- The bottom ~20% (y > 1500) sits under the platform's UI: no key visuals there. Put captions just above it.
- Keep ~60 px margins on the sides. Keep the right edge (x > 960, y 900-1700) free of key information, because the like and comment buttons sit there.
- Center framing usually works best in vertical video. Use thirds for scenes with two subjects.
- One dominant element per frame. Everything else is smaller, darker or less saturated.
- A character's face should be at least ~15% of the frame width in key moments.
Opening seconds
- Start speaking at frame 0. No silent intro, no logo.
- Something visibly changes within the first 1-2 seconds: a new object, a move, a transformation. No static title card.
- Big text in the first frame is optional. If you use it, it states the claim; it isn't decoration.
- A face in the first frame is optional. A strong character or image works just as well.
Camera (2D and 3D)
- Every camera move has a motive: it follows a character, reveals information or adds emphasis. No decorative motion.
- Never move linearly: always ease in and out.
- Reveal = slow push-in (scale 1.0 to 1.1-1.25 over 1-3 s). Shock = fast punch-in (0.15-0.25 s) with a small shake.
- New location: wide shot first, then closer.
- Low angle = power or threat, high angle = small or helpless, eye level = neutral.
- If two characters face each other, keep each one on its own side for the whole scene.
- In 3D: FOV 35-50 degrees, depth from foreground, middle and background layers. Move the camera (dolly) rather than changing the FOV (zoom).
- In 3D cartoon scenes, use linear tone mapping and a slight saturation boost; filmic tone mapping washes cartoon colours out.
Editing and pacing
- After the opening, pacing follows the style. A fast first moment matters more than a fast whole video.
- Eye trace: the point of interest in a new shot starts where it was in the previous shot.
- Match cuts: a shape or colour from the end of one scene becomes the start of the next.
- After a joke or a punchline, hold 0.3-0.6 s before cutting.
- Slow down before the key moment, speed up after it.
- Cut on action: change shots in the middle of a movement, not after it stops.
- No white flashes or strobe transitions. Use whip pans, match cuts and wipes.
Character animation
- Squash and stretch on every impact and jump (scale 0.8-1.25, volume roughly kept).
- Anticipation: a small move the opposite way before a big move (0.1-0.2 s).
- Follow-through: hair, antennas and arms keep moving and settle after the body stops.
- Limbs and thrown objects move on arcs, never on straight lines.
- Blinking, breathing and idle sway never stop, even while a scene waits.
- Timing sells weight: heavy things move slowly and land hard, light things move fast.
- Exaggerate emotions: huge eyes, big mouth shapes, a flying sweat drop.
- One clear silhouette per pose. If the pose doesn't read as a black shape, change it.
Text and captions
- Captions: 2-5 words per chunk, synced to the voice, popping in, high contrast (white with a dark outline).
- Big kinetic words only at punch moments, and only one big word group on screen at a time.
- Numbers on screen exactly match the narration and the source.
Voice
- Short sentences, a spoken rhythm, contractions ("it's", "didn't").
- About 170-200 words per minute.
- 0.2-0.4 s of silence between beats, 0.6-1.0 s before the final line.
Honesty
- Every number on screen has a real source and matches it.
- Never put first-person claims into the narration that the creator didn't actually do.
- Parody headlines use clearly made-up names, never imitations of real outlets.