Explainer videos look cheap for mechanical, nameable reasons: frames that freeze while the voice keeps talking, sentences on screen restating the narration, interface the product doesn't actually have, and one rhythm repeated until the next transition is predictable. Viewers almost never name these. They report that the video feels "off," or they stop watching, and the maker is left guessing.
Every failure below is a rejection that actually happened in our production records, most more than once. Budget was rarely the variable: when the same topics were built twice with the same tools, once rejected and once accepted, effort and code volume didn't separate the takes (the census).
Picture an explainer for Sift, an error-triage tool. Errors stream in from production, Sift groups the duplicates into a single incident, and an engineer resolves that one incident once. A strong ninety-second video of that product exists. Each failure below is a way to ruin it.
The screen restates the voice
The narrator says "Sift groups duplicate errors into one incident," and the screen prints GROUPS DUPLICATE ERRORS INTO ONE INCIDENT while he says it. The narration already owns the words, so on-screen prose competes with the voice, and a viewer forced to read and listen at once does both badly. The voice-side version of the same failure is narration drafted as copywriting, which sells while the picture is trying to teach.
Watch the video with the sound off. If you can read the whole script from the screen, you have a slide deck with a fade. Three fixes:
- Let the narration own words and the screen own motion and state. A failed group is a red mark and a ring, never the word FAILED printed on it.
- Delete every on-screen sentence the voice already says. A label can name a thing. Explaining it is the voice's job.
- Write narration that explains what the viewer is looking at. If the voice is selling while the picture teaches, rewrite the voice (how).
The frame goes still while the voice keeps talking
In the cheap Sift video, everything animates in a scene's first three seconds — errors pour in, the group forms, the count settles — and the frame then sits frozen for five more while the voice finishes. Usually the animation was authored first and the narration bolted on, or every element was allowed to settle early. Either way the picture stops carrying information, and viewers register the mismatch even when they can't describe it.
Pause five times at random. If more than a third of your landings are static frames, the video is a slideshow with a soundtrack; the measurement, the hold caps, and the repair rules live in animation timing and easing. Two fixes:
- Re-pace the beats so they land as the voice names them. The group forms when the narrator says it forms.
- Keep a pulse in every hold. A slow drift, a ticking counter, or a working dot keeps a resting frame alive.
The interface on screen was invented
Invented interface does the most damage. Wanting spectacle, the maker opens the Sift video on a "triage command center" — a wall of gauges Sift doesn't have — and floats error payloads as chips beside a diagram instead of putting them in rows. Viewers who know Sift flinch immediately. Viewers who don't still sense it through texture, because data in real software lives inside things (rows, panels, fields), and information floating without a container reads as marketing. Painting the invented panel in the product's real colors does not rescue it; we tried exactly that once, and the verdict is quoted in show the real product.
Invented color belongs to the same family: when every concept gets its own hue and glow is applied as ambience, the video has invented a visual system the same way it invented the dashboard. Three fixes:
- Get spectacle from real surfaces visibly running. Queues flipping, the group filling, parallel errors landing. If a concept can't be shown with real surfaces, the concept is wrong.
- Put every value in a container. Values resolve inside rows, panels, and fields; the only thing that travels along a connector is light.
- Give color a stated meaning or cut it. One accent, semantic states (live, done, error), neutral everything else. If you can't say what blue means, blue goes.
The climax happens where nobody can see it
Every explainer is built around a money moment; in the Sift video it is two hundred duplicate errors becoming one incident. The cheap version shows that as a counter changing inside a small box on a wide, motionless frame. The maker knows where the payoff is and can't not see it. The viewer sees a frame where approximately nothing happened.
One rejected build of ours hinged on exactly this. Its money moment was a one-word text change occupying about 1/40 of the frame, and the reviewer's whole verdict was "don't see anything cool." The event occurred on screen, but a payoff the viewer can't see is a payoff that didn't happen. Two rules:
- Stage payoffs at frame scale. Push the camera in, scale the element up, or restage the event (camera moves).
- Hold the 1/40 floor. A payoff smaller than a fortieth of the frame did not happen, whatever the narration claims.
Every scene has the same shape
In the Sift video: every feature gets the identical beat — card slides in, text fades up, card slides out — with only the words changing. In stock-template explainers: characters shrugging at laptops and icon rain. Both come from a mold, and a mold has one rhythm. Halfway through, you can call the next transition before it happens, and nothing accumulates because every scene resets the world to zero. Two fixes:
- Vary the beat shape. A wide pull, a fan-out, a close-up, a scrambled finish (the mechanics of rhythm).
- Let the world accumulate. The incident list keeps its resolved rows instead of wiping clean each scene.
The world doesn't hold together between scenes
The grouping panel is one size in scene two and a different size in scene three, twenty-four errors become six with no visual account, and a black flash sits between scenes. Each scene owned its own copy of the layout, and nobody compared one scene's exit frame with the next scene's entry. Viewers report "jarring" without being able to say why.
The linear pan belongs to the same family: a chain of steps laid out too wide for any frame, with the camera panning along it as compensation. Camera work can't fix a layout problem, and a linear pan reads like a sentence one word at a time. Two disciplines:
- Build one layout and shoot it. Scenes differ only in state; persistent objects interpolate instead of teleporting (one camera over one world).
- Fix layouts by staging, never by panning. A set piece that has to be cut in half is a scene carrying too many ideas (one idea per scene).
What the failures share
The rejected videos in our records were invented: ad-hoc diagrams, made-up interface, a visual system improvised per scene. The accepted ones were composed: real surfaces, one fixed layout, one accent used with discipline, and motion as the only truly original layer. Creative energy belongs in staging — story, time structure, camera, causality — and never in surface invention.
A bad frame costs minutes to catch as a still and hours to catch as a finished render. The batch that taught us to review stills before animating is in how to make an animated explainer video, and the positive version of everything above is what makes a good explainer video.
FAQ
Is a cheap look the tool's fault? Mostly no. Identical topics built twice with the same tooling differed on surface honesty, camera discipline, and timing, not effort (the census).
Does spending more money fix it? Not by itself. Every failure above has shipped at every budget (what video actually costs).
What's the fastest way to audit a video I already have? Watch it muted and see whether you can read the script off the screen. Pause at random five times and count the frozen frames. Show it to someone who knows the product and watch for a flinch.
What about bad music and sound effects? They fail the same way. We once rejected a full sound pass of 25 carefully mapped cues because the assets were stock UI-demo clicks, and perfect integration of cheap assets still reads cheap. Silence beats bad sound.
The fastest way to test any of this is on your own product. Send us the URL and choose from twenty short videos of it.