One idea per scene means every scene in an explainer video teaches exactly one thing, a thing you can state in a single sentence before the scene is built. If you can't write the sentence, the scene isn't understood yet, and if the sentence needs an "and," you are looking at two scenes.
The other craft rules trace back to this one: scene length, camera moves, what gets animated, and what the voice says all follow from the scene's one idea.
The examples below come from a 90-second explainer for a feature-flag platform: a flag flips on without a deploy, a rollout ramps from 5% of users toward everyone, a metric dips, and a rollback lands instantly. Each of those is one idea, and each gets exactly one scene.
What "one idea" means
An idea is a claim: a sentence the viewer should believe after the scene that they didn't believe before it. "The flag flips without a deploy." "A rollout ramps in steps you choose." "A bad metric rolls back in one click."
A topic is not an idea. "The dashboard" and "rollouts" have no finish line, so they can absorb unlimited screen time. A claim finishes the moment it is believed, and that finish line is what makes it schedulable into 8–12 seconds.
The whole video also argues one sentence. For the flag video: "you can ship to 5% of users and take it back in one click." Each scene owns one link of that argument, and a scene sentence that doesn't push the video's sentence toward obvious marks a scene to cut, no matter how good it would look.
Why the rule holds
Viewers infer meaning from what changes on screen, and they can track one focal change at a time. In our graded builds, when three things animated at full strength at once, the viewer picked one at random, and it was usually the wrong one. The builds that passed review keep one focal element per moment and dim everything else to about a third of full strength.
A two-idea scene has two payoffs, so neither gets staged at full scale, and two claims, so the narration rushes and the voice-picture sync lands on neither. One of our worst-graded builds divided its attention this way and staged its climax so small that the reviewer reported seeing nothing happen at all (the invisible climax). The flag video's version of the failure: the rollout percentage steps up while a targeting-rules panel animates beside it, and neither change gets the frame.
Four tests for a two-idea scene
- The "and" test. Write the beat intent. "The viewer sees that the rollout ramps in steps and that a bad metric triggers a rollback" is two intents wearing one scene.
- The two-payoff test. Every scene has a moment the whole thing exists for. If you can point at two, the second is being wasted, because a payoff the viewer isn't set up for is just an event.
- The shape test. Good scenes have a nameable shape: here is a thing, watch it work, change one part, or what did it leave behind. When no single shape fits, the beat is usually two beats. Mapping a beat to a shape should feel like selection, and when it starts feeling like invention, that is the smell.
- The narration test. If the voice has to explain something the picture isn't currently doing, the scene carries an idea it doesn't show.
The most common two-idea scene is behavior plus anatomy: the machine runs, and mid-run the scene also tries to explain what's inside one of its parts. The fix is a cut we use constantly: freeze the run at its interesting moment, hold that exact state across the scene boundary, and open the next scene inside the held moment. The run is one idea, the anatomy is another, and the held frame is the seam. Never finish the run and then run it again for the close-up, because viewers give a machine's first run full attention and its third almost none.
Sequencing the beats
A video of one-idea scenes is a sequence of sentences, and the sequence has a shape: orient from what they know, then the macro claim, then the micro mechanism, then one worked example. The flag video's chain:
- You already ship features behind flags. (The orientation.)
- A flag flips without a deploy. (The macro claim.)
- Clients check the flag at runtime, so the change reaches users on their next request. (The micro mechanism.)
- One bad release: the metric dips at 10%, and the rollback lands before the other 90% ever saw it. (The worked example.)
The chain also scopes the video: an internals deep-dive on flag evaluation has no link to live on, so it stays out.
- The camera moves between ideas, and cuts happen inside one. A camera move tells the viewer "new idea," and the signal stays honest only while scene boundaries are idea boundaries. When they drift apart, viewers report the video feels jarring.
- Name what each scene refuses to teach. We write, per scene, what not to say, and keep a standing list of topics deliberately untaught, each with a destination in another video or the docs.
What the rule does not mean
One idea per scene is not one element per scene. The best-graded builds keep several surfaces alive at once, and the flag video can too: an evaluation counter climbing, the rollout bar creeping, and a metric chart ticking all serve the single focal idea of a ramp in progress. A frame can be rich as long as you can answer, at any moment, "what is this scene arguing?"
A scene also earns whatever time its one idea needs to land, usually 8–12 seconds. Cutting a scene before its idea lands loses the sentence the same way stuffing two ideas in does.
Downstream, the rule decides scene count (storyboards get one contract row per idea), video length (the runtime is the sum of what the ideas need), and narration (the voice gets one intent per scene to serve). It is the closest thing this craft has to a first principle.
FAQ
How do I know if my idea is too big for one scene? Name its payoff moment. A scene-sized idea has one moment the viewer must see; an idea that needs several moments in order is a chain, and each link gets a scene.
Can two scenes share an idea? Only as a deliberate pair: a run that freezes mid-flight, and a second scene that completes the same run after a close look inside. Two scenes that independently half-teach the same idea both fail.
Does this apply to a 30-second video? More, if anything. Thirty seconds holds three or four ideas at most, and one sentence per scene tells you, before production, whether you have a 30-second script or a 90-second script wearing a 30-second runtime.
What about the hook — is that an idea? Yes: the video's thesis stated visually. The hooks guide covers the first five seconds that carry it.
Your own product is the fastest test: send us the URL and choose from twenty short videos of it.