Why it works
- Every claim is shown as its receipt: a highlighted screenshot of the source on the same evidence board, about once a minute — it reads as research, not opinion.
- Definitions as the spine: each essay pins down one contested word and judges the claim against it ('definitions determine decisions').
- Opponent-first hooks: the video opens on the other side's clip, comment or quote, then answers it in one line.
- One locked visual system — paper, grid, light-blue halftone circle, kraft paper, yellow highlighter, dark-red/royal-blue data, serif headlines — so a very fast production still looks designed.
- Fairness signals and a fixed ritual: steelman objections, credentials, the same newsletter pitch twice and the same sign-off every time.
Kaizen Asiedu (@thatsKAIZEN, 59,100 subscribers when we looked on 26 October 2026) makes long studio essays on political and cultural topics. This post is not about what he argues. It is about how the videos are made: a script that starts from a definition, and a graphic system consistent enough that an essay with 30 to 90 graphic moments still looks like one hand made it.
We read eight transcripts, watched five essays (156 minutes) at one frame per second, and measured more than a dozen animations frame by frame at full frame rate. Each claim below links to the moment in the original video so you can check it. The animated loops in this post are our rebuilds, playing our own sample content. None of his footage or screenshots appear here. Every element is in Reel Cut as the KAIZEN package.
Three formats, one voice
The channel's most-viewed video is not an essay. It is the rally vlog, shot vertically on a phone with no graphics (491,000 views). The street interviews cut handheld answers under single-word pop captions. Then came three-to-four-minute takes: selfie camera, word-pop captions, black-bordered screenshots as picture-in-picture (short take, 1:19). Since late August 2026 almost every upload is a studio essay.
| Format | Picture | Graphics |
|---|---|---|
| Phone vlogs, stunts, street interviews (from 2024) | vertical, handheld | none, or one-word captions |
| Short takes (2025–2026) | vertical selfie, 3–4 min | word-pop captions, framed screenshots |
| Studio essays (since late Aug 2026) | 16:9, 4K, 16–58 min, two cameras, chaptered | the paper board system below |
What carried over matters more than what changed. The 2025 short take and the essay of 17 Sep share a title and open the same way: an opponent's clip, a two-word pivot, a definition. The essay did not invent a new voice; it gave the old one a set and a graphic system. Some essays are literally compilations: the 38-minute essay of 17 Sep stitches five earlier short takes into five numbered chapters. If you are starting out, that is a useful model. Short takes are the test bench; the essay collects the ones that worked.
The script
By the numbers
| Measure | Essays |
|---|---|
| Speaking rate | 133–154 words per minute |
| Median sentence | 10–11 words |
| Spoken numbers | 1.4–3.9 per minute |
| Hard cuts | 3.8–6.2 per minute, median shot 4.3–6.2 s |
| Screen time | A-roll 55–65 %, graphics 12–19 %, clips and B-roll the rest |
Slow delivery, short declaratives, cuts tight between thoughts. The number rate drives graphic density: the essay of 7 Oct says 3.9 numbers a minute and spends the most time on the board (19 %).
The opening: their voice first
Every essay opens in under 40 seconds with the same pattern: the other side speaks first, a one-beat pivot, the title said as the answer, then the stakes.
| Video | Opener | Pivot | Answer |
|---|---|---|---|
| essay of 7 Oct, 0:00–0:33 | reads a top viewer comment that argues the opposite, shown as a comment card | concedes one number | says the rest of the comment is wrong |
| essay of 14 Oct, 0:00–0:17 | conference clip with a slogan | repeats the slogan as a question | the dictionary definition |
| essay of 17 Sep, 0:00–0:22 | debate clip with an accusation, burned-in captions | "Hold on." (0:21) | the title |
| essay of 19 Oct, 0:00–0:39 | two public figures on one word, red-tinted | grants the premise for argument's sake | three rhetorical questions |
| essay of 30 Aug, 0:00–0:35 | no clip: a blunt verdict | immediate nuance | the stakes |
Why it works: the viewer hears the claim in its own voice before hearing the reply, so the reply has something to land on, and the title is paid off in half a minute. The YouTube "most replayed" graph confirms where attention goes afterwards: the peaks sit on evidence moments (essay of 17 Sep 12:21 and 26:14, essay of 30 Aug 4:19), not on the hook. People rewind the receipts.
Definition first
Each essay pins down one contested word and judges the claim against a dictionary or legal definition. His own line for it is "definitions determine decisions" (essay of 7 Oct, 1:29), and it returns at 13:09. A second recurring line says a word has a meaning, not a mood. The definition is shown, not only said: a dictionary page with the key clause highlighted (1:24), a definition headline over a row of objects, a stamp over what people think the word means. Near the end the dictionary page comes back with a page curl (14:33). The definition is the ruler; everything in between gets measured against it.
The beats
The analogy beat is long and deliberate. In the essay of 30 Aug a sports hypothetical runs from 4:23 to 8:28, a quarter of the video, before he maps it back onto the real case at 8:48. The essay of 14 Oct uses a water bottle relabelled as a medicine; the essay of 17 Sep a pilot-hiring thought experiment with two systems side by side.
Voice and devices
- First person to "you", calm and prosecutorial.
- Numbered structures: three sources of confusion, five statements, three questions, objection one and two.
- Antithesis as a section closer: two short sentences that mirror each other, one word swapped.
- Fairness signals: says the piece is not an attack on the person (essay of 14 Oct, 1:19), steelmans objections, concedes the strongest counter-evidence before answering it.
- Dated sourcing in speech: month, year and author before the claim, with the same source on screen as a card.
- Credentials, used sparingly.
Rituals
The newsletter pitch is the same short script every time, two or three minutes in and again mid-video (essay of 7 Oct, 2:56 and 11:13). The sign-off is always a variant of one line, ending on the hope that the video helped you "think more clearly" (essay of 7 Oct, 17:19; essay of 30 Aug, 15:55). Same open, same middle break, same close: a 20-minute essay feels like an episode of a show.
Titles and thumbnails
Titles follow five formulas: contradiction ("No, …", "… Didn't Even …"), a two-clause turn, a conditional challenge ("If …"), a first-person stunt, and a mastery promise. A named person or news hook comes first, Title Case, 6–12 words. Thumbnails split the frame: the subject on the left (often desaturated or a symbolic collage), the host on the right mid-argument, a thin gold divider, and a 2–4 word condensed-caps verdict on a navy bottom band with one word in yellow or red.
The visual system
The board
Most graphics live on one surface. Bottom to top: cream paper with fine grain, a faint graph grid (a line every 100 px at 1080p), a small "Source: domain" label top-left. The grid is locked to the screen. When a card scales and the grid stays still, the eye reads a card placed on a desk rather than a zoomed video. Around each card sits the same set of props: a light-blue halftone circle, a dark dot matrix, an offset sheet of kraft paper, two ink dots. None of them carry information. They are the frame that makes a screenshot from any website look like it belongs to the same scrapbook.
Palette
Two families on one paper: paper, sky, kraft and highlighter for evidence, red and blue for shares, navy and gold for generated data cards.
The A-roll
Seated at a table with a boom-arm mic, a glass of water, plants and a city window; a warm-teal grade (the window samples around #378583); shallow focus. A wide master and a side close-up, recorded in 4K at 23.976 fps. Graphics are either full-frame cutaways from that set or cards laid over it.
The elements
Counted on one-frame-per-second contact sheets of the five essays. The evidence board appears about 10.6 times per ten minutes; everything else together adds about 4. The board alone is most of the graphics.
1. The evidence board
What it is. A source screenshot as a white card on the paper board, with the key line highlighted (essay of 7 Oct, 7:09).
Why it works. It turns "trust me" into "look". It costs almost nothing per use: the dressing is fixed and only the screenshot, highlight and domain change, so a single editor can produce dozens per essay and they all match.
Anatomy, bottom to top:
| Layer | Detail |
|---|---|
| Paper + grain, grid, source label | locked to the screen |
| Halftone circle | #A0D6E2, about 460 px wide, behind the card's top-left corner, about a third visible |
| Dot matrix | 6 × 6 dark dots over the circle's edge |
| Kraft paper | #E1C593, about 40 px past the right edge, about 38 px below |
| Card | white, soft shadow, about 67 % of frame width |
| Highlighter | #F5F213, multiplied behind the key line(s) |
| Ink dots | two dots under the card's bottom-left corner |
Timing (measured on the 30 fps upload):
| Event | Start | End | Curve |
|---|---|---|---|
| hard cut in, card already at 77 % scale | 0.00 s | — | — |
| card scales 0.77 → 1.00 | 0.13 s | 1.10 s | cubic-bezier(0.30, 0.45, 0.10, 0.86), ~1 % overshoot, settled by 1.4 s |
| dot matrix appears | 0.20 s | 0.23 s | pop |
| circle scales from 0 | 0.20 s | peak 0.70 s, settled 0.95 s | spring, ~5 % overshoot (stiffness 70, damping 11, mass 1) |
| kraft slides out from behind the card | 0.40 s | 0.70 s | ease-out cubic |
| highlight, line 1, left to right | 0.40 s | 1.03 s | ease-out sine |
| highlight, line 2 | 0.53 s | ≈0.90 s | ease-out sine |
| drift | 1.4 s | the cut | linear zoom-out, ≈0.5 % per second |
| exit | — | — | hard cut back to A-roll |
Two details are worth copying. The cut does the appearing and the scale does the settling: the card never grows from zero. And the motion ends on the information: the props are done by 0.95 s and the highlight finishes last. In the 30 fps upload every fifth frame repeats, so the animation was made at 24 fps and conformed to 30.
Variants: a push-in to the highlighted headline (essay of 7 Oct, 7:33), a page curl for dictionary pages (1:24), a stack of overlapping sources (essay of 14 Oct, 6:12), and a mock card with a small disclaimer underneath for hypotheticals (essay of 30 Aug, 5:27).
Rebuild it. Group the circle, dots, kraft, card, highlight and ink dots and scale them together around the card's centre; keep paper, grid and label outside that group. Define highlights as rectangles in screenshot pixels with a start time each, so swapping the screenshot keeps them aligned. In our package: screenshot, sourceLabel, highlights (X, Y, W, H, start), enter (scale, page-curl, stack), optional push-in, drift, disclaimer, and a stepped24 toggle that steps the motion to 24 fps.
2. Flag-map stat stack
What it is. A country silhouette filled with its waving flag on the left, a serif title, and a stack of dark-red plates on the right, each a percentage and a label (essay of 7 Oct, 3:18).
Why it works. Each plate lands on the word it belongs to, so the screen keeps pace with the sentence. The flag treatment does the place-setting without a caption: the country is masked over a waving flag plate, halftoned, with a cyan rim offset down-left like a misregistered print.
| Layer | Start | Duration | Motion | Curve |
|---|---|---|---|---|
| flag map | 0.00 s | 0.70 s | flies in from ≈3.5× scale | ease-out quart |
| title | ≈0.35 s | ≈0.5 s | slides down out of a blur | ease-out cubic |
| row n | on its voice cue (measured: +0.47, +1.87, +4.57, +7.23 s) | 0.73 s | wipes left to right while the percentage counts up | cubic-bezier(0.10, 0.60, 0.80, 1.07), ~1.5 % overshoot |
| flag | loop | 2.5 s period | waving shading | linear |
Rebuild note. The plate is not a bar: its width follows its text, not its value. It is a list dressed as a chart. If your labels vary a lot in length, viewers may read width as size; keep labels similar or use the split bar instead.
3. Split share bars
What it is. A flag map, a serif title, and one or two rows: a handwritten label over a horizontal bar split into two shares, both with counters, plus a legend (essay of 7 Oct, 11:41).
| Layer | Start | Duration | Motion |
|---|---|---|---|
| map | 0.00 s | 0.50 s | flies in |
| row label | ≈0.1 s | ≈45 ms per character | types on |
| title | ≈0.2 s | — | letters drop in one by one, 2 frames apart |
| full bar in share-B blue | ≈0.67 s | 0.33 s | grows from the left |
| share-A red segment + both counters | ≈1.0 s | ≈1.5 s | grow and count together |
| second row | on its own cue (+3.33 s in the measured shot) | same | same |
Why it works. The handwritten label reads like a note in the margin next to hard numbers, and both counters running at once make the split feel measured in front of you. A comparison variant puts two countries side by side over a grey halftone city collage (essay of 7 Oct, 3:48; essay of 14 Oct, 2:55).
4. Side card over the A-roll
What it is. A white card over the right 38 % of the talking head, carrying a chart, a pull quote, a list of claims or a headline (essay of 7 Oct, 6:56).
Why it works. The face stays on screen. For a number that needs eight seconds of explanation, cutting away to a full board would lose the speaker; the side card keeps both.
We tracked the card's left edge frame by frame: 1402 → 1130 px (at 1920 wide) over 1.83 s, with opacity following the same curve from about 10 %.
| Time | Share of the move done |
|---|---|
| 0.25 s | 43 % |
| 0.50 s | 68 % |
| 0.93 s | 88 % |
| 1.83 s | 100 % |
A fast start and a long, soft landing: cubic-bezier(0.22, 0.70, 0.35, 1.00) fits it. The exit is short, about a third of a second. The pull-quote variant (essay of 30 Aug, 11:58) uses a gold left rule, the last clause in bold, and the attribution in condensed caps over a small grey publication-and-date line; the claims variant (1:40) is a caps title over ruled lines.
5. Annotated line chart
What it is. The chart inside the side card, also used full size on the board (essay of 14 Oct, 12:39). It is static; the card carrying it provides the motion.
Anatomy: a caps title with a short gold underline; light grey y-ticks; a cream band (#FCF4E7) over the period that matters; a dotted baseline with a small label; a 3–4 px navy line; a gold dot on the peak with the value in bold above it; a caps annotation with a thin curved leader; x-ticks; a caps source footer bottom-left.
Why it works. One claim per chart, and no legend. The band answers when, the dot answers how much, the annotation answers why. Everything else is quiet grey. Our rebuild keeps it static by default and adds an optional draw-on.
6. Milestone timeline
What it is. A poster on plain paper: a condensed-caps title with a printed texture, a gold rule with a centre diamond, a thick navy track with gold ring markers, a year and a two-line label over each, and under it a dashed grey track with grey dots and a caption stating the usual duration (essay of 30 Aug, 1:44).
| Layer | Timing (23.976 fps source) |
|---|---|
| title | drops in with vertical motion blur, settled 0.75 s after the cut |
| gold rule with diamond | arrives with the title |
| track | draws to the first marker (0.25 s) |
| marker n | pops with a small overshoot on its voice cue (second marker at +1.1 s) |
| dashed track | follows the main track head |
Why it works. The two tracks make a comparison without a second chart, and each year appears when it is said. A card variant sits in the board dressing with the last event in red and an italic tagline underneath (essay of 14 Oct, 22:40).
7. Type moments
Definition headline with objects (essay of 7 Oct, 5:45): the word slides in letter by letter from the right out of a blur (0.67 s), the definition types in under it, and three object cutouts rise one after another about 0.6 s apart, each with a typed label.
Oversized serif quote (5:53): words rush in from the right at about 1.6× size with heavy horizontal blur and settle in 0.8 s; then a hand-drawn swash underline draws on in about 0.6 s. The blur sells the speed, so the move can be short.
Verdict stamp (0:53): a collage of what people think a word means builds over about four seconds (black-and-white cutouts, cobalt circles, small icons); a rubber stamp enters from the top-left, lands at about 4.3 s, lifts away, and a red imprint stays on the collage.
8. Single-idea cards
Stat certificate (essay of 30 Aug, 10:25): a cream card with a double gold hairline and corner ornaments pops in, then each word appears at the moment it is spoken: the number at +0.13 s, the noun at +0.5 s, the year at +0.75 s, the gold second line from +2.25 s. The layout is the transcript. Build it from word timestamps, not keyframes.
Numbered question (essay of 19 Oct, 31:38): a white card in the board dressing with a navy left rule, a small spaced-caps eyebrow, a gold kicker (FIRST, SECOND, THIRD), a heavy caps question, a short gold rule, an italic serif sub-line and progress dashes. It turns a numbered structure in the script into something the viewer can track.
Question intertitle (essay of 19 Oct, 9:12 and 10:44): a charcoal silent-film card with grain and an ornamental Victorian frame. One version sets the question in italic old-style caps with the key word in red; another sets a statement in plain white caps. In a 58-minute essay these work as act breaks.
Fallacy card (essay of 17 Sep, 13:04): a teal card on the right of the A-roll with the fallacy's name in caps, a line icon, "error type: logical fallacy" and a one-line definition. For a channel built on clear thinking, naming the argument error on screen is the most on-brand graphic in the set, and it is rare in our sample. Our package ships a set of common fallacies as presets.
Poll pie (essay of 14 Oct, 1:03): a black wedge sweeps around a light flag map while the percentage counts up (about 1.5 s), then an age pictogram row and a serif caption follow.
Concept chain (essay of 7 Oct, 6:00): two serif words joined by a rule, and below them three rounded pills (grey, mint, dashed lavender) linked by arrows, with a small caption under the last one.
9. Overlays and transitions
The newsletter break (essay of 7 Oct, 2:59):
| Step | Duration |
|---|---|
| diagonal stripe wipe: black, white and grey bars at about 65° | 0.5 s |
| back on the A-roll, which then blurs | 0.5 s |
| cream subscribe card fades and sharpens in | 1.1 s |
| hold | ≈2.8 s |
| card smears out sideways while the A-roll un-blurs | 0.33 s |
The stripe wipe is rare; almost everything else enters and leaves on a hard cut. Reserving it for the pitch marks the pitch as a break, which makes it easier to sit through twice.
Comment card (essay of 7 Oct, 0:11): the viewer comment as a white card on the lower third, brought back at 2:01 when he returns to it. Using a comment as the hook gives the essay a reason to exist.
10. Explainers and footage
Funnel and unit grid (essay of 17 Sep, 5:16–6:50): one wide paper canvas, with the camera panning between panels: a funnel, a group of pilot icons, a dot grid with a cut-off line, then two hiring systems side by side. It is the analogy beat made visual: one canvas, one camera, no cuts.
Analogy props (essay of 14 Oct, 5:15; essay of 30 Aug, 5:36): a relabelled bottle on the board, a cartoon player beside the host. Cutouts with soft contact shadows, placed like objects on the desk.
Footage treatment: opponents' clips keep their own burned-in captions; vertical phone clips sit centred on a dark textured field (essay of 17 Sep, 13:24); a red monochrome tint marks a confrontational cold open (essay of 19 Oct, 0:24).
How it is probably made
These are inferences from the frames, not confirmed by the channel.
- The board system looks like After Effects (or a similar template tool). The animation runs at 24 fps inside a 30 fps upload, the curves have the long tails of easy-ease keyframes, text has true motion blur, the stamp is a pre-rendered 3D object, and the dressing repeats exactly while only the screenshot changes. With 30 to 90 board moments per essay, a dedicated editor or a small team is likely.
- The navy-and-gold data cards look generated with code. The timelines, line charts, certificate and question cards share web-like typography, crisp vector edges and word reveals timed to the transcript, which suggests HTML/SVG output dropped into the board or side card.
- The cutouts (farms, factories, the bottle, the cartoon player) look AI-generated or stock 3D renders with contact shadows.
- The essays are scripted. The cadence and the numbered structures point to a full script written before recording; chapters are added in the description.
Start a channel in this style
Everything above is in Reel Cut as the KAIZEN package: free fonts (Newsreader, Oswald, Inter, Patrick Hand, IM Fell English, Anton, all SIL OFL), our own sample content, and every timing from the tables above as editable fields. A workable first episode:
- Write the essay before you record. In Write, set the beats as headings: a hook under 40 seconds in the other side's voice, the definition, numbered receipts, one extended analogy, objections answered with history, the verdict, your outro. Use the runtime shares above as a budget: for a 15-minute essay that is about 7 minutes of receipts and 90 seconds of analogy.
- Attach a receipt to each claim as you write it. Link the line to its screenshot, mark the lines to highlight and type the domain. Aim for about one board card a minute. If you cannot find a receipt for a line, that is a script note, not an editing problem.
- Record against the document. Your voice-over lands on lines that already know what they show.
- Place the elements in Edit. Pick them in the Motion window. Board elements hold for as long as your voice-over needs them, and every reveal (a highlight, a stat plate, a timeline marker, a certificate word) takes a start time in seconds, so it lands on the word.
- Make it yours. Change the circle colour, the data pair and the headline serif; write your own sign-off and your own ritual. Take the structure and the craft, not someone else's brand.
- Export in 4K from Render, or hand the timeline to Premiere, Resolve or Final Cut as XML.
Credits
All analysis is of public videos by Kaizen Asiedu, @thatsKAIZEN; his newsletter is ClearThinker. Watch the originals:
- Studio essay · 7 Oct 2026 · 17:31: evidence board 7:09, flag-map stats 3:18, side card 6:56, split bars 11:41, type moments 5:45–6:00.
- Studio essay · 14 Oct 2026 · 26:24: poll pie 1:03, card stack 6:12, line chart 12:39, sequence timeline 22:40.
- Studio essay · 19 Oct 2026 · 57:57: red-tinted cold open 0:24, intertitles 9:12 and 10:44, numbered questions 31:38.
- Studio essay · 17 Sep 2026 · 38:34: hook 0:00–0:22, unit grid 5:16, fallacy card 13:04.
- Studio essay · 30 Aug 2026 · 16:00: timeline 1:44, analogy 4:23, stat certificate 10:25, pull-quote card 11:58.
- Earlier formats: the rally vlog, the street interviews, the short take.
Measurements were taken from frame dumps of the published videos; colours, timings and font matches are our readings and may differ from the original project files.