An Animated Explainer About Its Own Making

(inferred) A familiar viewer scene turns into an apparent large-team production, then the film reveals a single tool behind it. An illustrated explanation proves the claim through concrete metaphors, a task demonstration and a making-of montage before returning to the viewer.

Why it works

• (inferred) The question appears by shot 2, while shot 3 turns the answer into three sequential cards.
• (inferred) Shot 8 gives the longest visual comparison, 6.32 s, so the abstract middle has a simple anchor.
• (inferred) Shots 16–19 compress four production stages into 5.92 s, paying off the opening claim.

Tools you'll need: Remotion for motion graphics · ElevenLabs or a recorded voice · Claude alone: yes

Hook formula (inferred): familiar viewer → assumed production scale → overturned explanation for {{PRODUCT}}. Why it holds (inferred): The first cut at frame 72 changes the social scale of the picture. The card is spoken and visible together, while the answer waits beyond the first three seconds.

Retention map (inferred): re-hooks in shots 3, 9 and 16 · payoff in shots 16–21 · final question in shot 22 · hard stop, with no measured fade to black.

Mix (inferred): illustrated wide scenes alternate with isolated cards, stage tableaux and interface-like proof. Rhythm: median shot 2.44 s; shots 3 and 8 make room for staged reveals, while shots 16–19 accelerate the final proof.

Rules this editor follows (inferred):

• Cut at phrase boundaries; keep the narrator's stressed words clear of extra effects.
• Use a wipe of motion only when the visual world changes sharply; the five measured whips are short.
• Keep within-shot builds distinct from cuts: shot 3 contains the frame-240 reframe and the frame-528 gradual blend.

• Look (inferred): flat, rounded vector scenes with subdued background layers, warm characters, pale card surfaces and crisp dark outlines. The reference alternates darker violet stages and brighter workplace scenes; map those roles to {{BRAND_BG}} and {{BRAND_ACCENTS}}.
• Graphics and overlays (inferred): simple white cards, 2D pictograms and short word reveals sit above illustrated environments.
• Transitions (inferred): hard scene cuts plus short directional whips; the flash-like change within shot 3 is handled separately.
• Branding and end card (inferred): shot 22 returns to the viewer and holds the question. In a new version, add a small {{LOGO}} and {{CTA_URL}} text in the final composition without obscuring the question.

      0s         10s        20s        30s        40s        50s        60s
mean  ▄▅▅▄▃▄▅▄▃▃ ▂▆▆▆▅▅▅▅▃▃ ▃▃▃▄▅▄▅▄▄▄ ▄▄▄▂▂▂▄▄▅▅ ▄▄▄▄▄▅▆▆▆▆ ▆▅▅▅▅▅▅▄▄▄ ▄▄
p95   ▇██▇███▆▄▄ ▇███████▆▆ ▆▆▇▇▆▆▇▇▇▇ ▇▇▇▆██████ █▇▇▇▇█▇███ ███████▇▇▇ ██

:root { --brand-bg: {{BRAND_BG}}; --brand-accent: {{BRAND_ACCENTS}}; --brand-surface: #f6f3ee; --ink: #202434; }
.scene { background: var(--brand-bg); filter: saturate(.78); }
.card { background: var(--brand-surface); border: 2px solid rgba(32,36,52,.12); border-radius: 20px; box-shadow: 0 12px 28px rgba(0,0,0,.16); }
.character { filter: drop-shadow(0 8px 6px rgba(0,0,0,.14)); }
.stage { background: radial-gradient(circle at 50% 15%, rgba(255,255,255,.14), transparent 42%), var(--brand-bg); }
.whip { filter: blur(12px); }

Measured: 193 words cover 79% in the v2 speech grid; v1 transcription gives roughly 194 overall words/min, low median pitch and animated emphasis, with short pauses around clauses.

Voice direction (inferred): An adult explainer speaks briskly and plainly, with a curious lift for questions and a dry, compact emphasis on the correction. Record an original voice or supply a licensed TTS voice. Keep the words understandable before adding the bed.

• Music (inferred): continuous restrained tonal electronics under the voice, with light changes of texture for the stage, mechanism and making-of sections.
• Sound effects (inferred): quiet card ticks, a short rush for each measured whip, and one restrained impact on the number reveal. Do not duplicate cues labelled “voice?” as effects.
• Silence and space (inferred): duck the bed under the narrator; let the final question breathe without a large closing sting.

██████████ ▇██████▇█▇ ▇█▇████▇▇█ ████▇█████ ██████████ █▇███▇▇███ ▇▁

• Title (inferred): “How {{PRODUCT}} made {{RESULT}}” or a verified production question; avoid promising a result you cannot show.
• Cover / thumbnail (inferred): one viewer silhouette opposite one clean production icon, with a short question in {{BRAND_FONT}}.
• Post copy (inferred): one claim, one piece of proof, then {{CTA_URL}} as text.
• Audience signals: The captured post showed 21,576 views; no replay data is available.

(inferred) Keep about 193 spoken words at roughly 187 words/min overall; captions are short anchor labels that follow the spoken noun, not a full transcript. Write original, verifiable claims for {{PRODUCT}} and {{RESULT}}.

[beat 1 · shots 1–2 · about 12 words] You might think {{RESULT}} took a whole team. Here is the surprise.
[beat 2 · shots 3–7 · about 45 words] Name three real parts of the work, then identify {{PRODUCT}} and its provenance. Each card follows its noun.
[beat 3 · shots 8–9 · about 24 words] Compare three truthful modes or stages, then ask how the central mechanism works.
[beat 4 · shots 10–13 · about 45 words] Explain the mechanism with one visual metaphor, one simple rule and one verified quantity.
[beat 5 · shots 14–20 · about 48 words] Show a real task and narrate four production verbs as the corresponding stations appear.
[beat 6 · shots 21–22 · about 19 words] State the documented result, return to the viewer, and ask a closing question.

Two routes (inferred)

Minimum: Claude alone in code. Remotion with SVG/CSS vector scenes, character rigs, text cards and a synthesized original bed reaches about 80% of the look. It simplifies character acting and perspective; it gives up bespoke illustration detail and natural voice nuance. The user supplies a recorded or TTS narration file and any real evidence of {{PRODUCT}} or {{RESULT}}.

Full: with the toolkit. Write and verify the narration; generate an original voice with ElevenLabs or record it; build vector art and all 2D motion in Remotion; compose and synthesize the bed in code; mix the WAV stems; render H.264. Handoffs: script and phrase marks → voice WAV at 48 kHz; voice WAV and click/cue track → Remotion timing; SVG character assets and real evidence captures → Remotion at 1920×1080, 60 fps; Remotion frames → final MP4 and stereo WAV. Export camera keyframes as JSON so scenes can be retimed without repainting.

Asset honesty: use real captures of {{PRODUCT}} or UI rebuilt from verified product material. Never generate product UI, fake screenshots, results, metrics or reviews. The production-station montage must depict what this team actually did.

Script prompt (inferred)

Write an original ~193-word narrated explainer for {{PRODUCT}} showing a verified {{RESULT}}. Use six beats and the shot mapping in plate 4. Open on an everyday viewer, challenge an assumption, explain one real mechanism with a metaphor and a verified number, then show four actual making-of steps. End on an honest question and text CTA {{CTA_URL}}. Use short anchor labels only. Avoid reference wording and unsupported claims.

Illustration prompt (inferred): Build reusable rounded SVG figures and scene props in {{BRAND_BG}} and {{BRAND_ACCENTS}}. Keep character construction consistent across shots 1, 2, 16–20 and 22. Use actual {{LOGO}} only in the closing composition.

How it's built (inferred)

Signature move: the apparently still stage becomes a three-way scale comparison in shot 8. Build one wide vector stage with three pedestal/character groups as separate layers. Use the plate 5 camera path: its aggregate image move is x −35.7%, y +37.6%, scale ×1.97, ease-in, with the final focus on the larger figure. Retain the measured ↳ segments from the camera table instead of spreading that move uniformly across frames 1070–1448. Use cubic-bezier(.55,.05,.8,.45) for the travelling segment and a 12 f ease-out settle. Character entrances occur on their spoken names; animate pedestal scale and opacity in 10–16 f. The camera moves on ones, as the measured cadence does. Render 4 subframes per output frame during fast travel and average them for approximately 180° shutter blur; keep text on an unblurred overlay. The gotcha is clipping the stage edges after nearly doubling scale: draw a 25% overscan backdrop.

Other moves and hand-offs. Shot 3 starts at frame 225; at frame 240 jump the card canvas ×1.36 with x +10.2%, y −5.4% reframe. The strips show cards building over the following seconds; keep those as element animations, not new cuts. At frame 528, blend the two pictures over the measured 8 f with a white radial mask and opacity crossfade. (inferred) The frames suggest a bright wash; the measurement says an unknown gradual change, so the exact blend mode is uncertain. The 7 f whip into shot 4 at frame 549 carries the white wash into the darker stage. The 6 f whips at frames 1449 and 2391 bridge question/stage and task/desk; use directional position plus 12 px motion blur. The 2 f whips at frames 2751 and 3411 bridge into the production montage and final viewer; use a vertical blurred smear without adding an extra cut. Hard-cut boundaries are only those in plate 5.

For shot 10, move the tall book/code composite upward by the measured +129.5% of frame height with ×1.32 scale, using the irregular camera segments in plate 5; leave source labels on an unblurred overlay. In shots 16–19, pan across four separately drawn stations, using plate 5's measured shot 16 +43% x travel and the following pan segments; cut only at the measured boundaries. The tiles suggest brief smoke-like overlays around station hand-offs, so add a soft 8–12 f opacity mask only inside a shot. Shot 22 returns to the room and gradually reveals the question; preserve its measured final hold and do not fade to black. All moving layers render on ones at 60 fps.

Build order

1. Lock the original narration to the six beats and measured shot boundaries; mark phrase and cue frames.
2. Draw the reusable viewer, workplace, stage, card, book and desk layers; verify the evidence in beats 4–5.
3. Animate shot 8's camera path, then the shot 3 card sequence and shot 10 vertical move.
4. Add the five measured whips, the frame-528 internal blend, then the remaining cuts and text builds.
5. Mix voice, original bed and restrained accents; render and compare plate 14.

• All 22 shot starts match plate 4 within ±1 frame; frame 240 and frame 528 remain changes inside shot 3.
• Shot 8 reaches the measured plate 5 travel and ×1.97 scale without exposing scene edges.
• Moving scenes render on ones near 60 unique frames/s, with no accidental frame holds.
• Every must-read word is at least 38 px cap at 1920×1080; mobile versions use their plate 15 minimum.
• Caption nouns follow narration; no effect masks the stressed words or turns “voice?” cues into extra impacts.
• Cue accents land within ±3 f of the selected cue frames in plate 8; cuts follow phrases rather than 108 BPM.
• X/LinkedIn master targets −14 LUFS integrated and true peak ≤−1 dBTP.
• Ending remains visible through the final shot, without a fade to black; {{LOGO}} and {{CTA_URL}} are readable text.
• No reference brand colours, font, mascot, slogans, music or unsupported production claim survives.

Brand slots

What this film avoids (inferred): live-action talking head, dense UI chrome, full transcript subtitles, long feature lists, stock footage, QR codes and store badges.

Brand-distance warning (inferred): A small orange robot, violet gear cards and the same production claim together would read as the reference. Replace the character silhouette with one drawn from {{PRODUCT}}'s own visual language.

Adapt the aspect (inferred):

• 9:16: Re-lay out every card and station vertically; move side labels below their icons and place the final question above the lower platform controls. Use ≥22 px cap at 1080×1920, with more generous 32 px for primary words; keep top and bottom UI safe zones clear.
• 1:1: Re-lay out the three-card sequence as a compact grid and move the final question inward beside the viewer. Use ≥22 px cap at 1080×1080 and reserve outer margins for platform UI.

Adapt the length (inferred):

• 15 s: Keep beats 1, 2, 5 and 6 in compressed form; omit optional provenance and mechanism illustrations while retaining shot 8's signature comparison.
• 30 s: Keep all six beats, shorten explanations in beats 2 and 4, and keep the final return.
• 60 s: Use the measured structure; spend added space on verified examples and clear narration, never on longer empty holds.

The creator's signature (don't copy): the reference's robot, company naming, exact claim, palette, typography and music. The transferable format (do copy): viewer assumption, visual negation, illustrated explanation, production proof and return-question structure.

Source: https://x.com/sleepclip/status/2119975673662194100 · measured 2026-11-11 · (measured) = read off the film · (inferred) = our reading