Scribe v3 Proof-Card Launch Film

A spoken launch announcement moves from a tiny mark to an abstract product title, measured proof, then demonstrations inside clean interface cards. A second capability gets its own dark reset before human examples bring the promise down to individual words and numbers.

Why it works

• (inferred) The opening mark occupies shot 1 for only 38 frames, then the brand surface appears at frame 38.
• (inferred) Dark shot 3 interrupts the pale interface at frame 345; the proof number arrives after a 3-frame fade from black.
• (inferred) The two long demonstrations, shots 12 and 13, let viewers inspect results after short claim cards have set context.

Tools you'll need: Remotion for motion graphics · ElevenLabs or a recorded voice · screen recordings of the product · Claude alone: partly

Hook formula (inferred): tiny {{LOGO}} → full-field brand surface → {{PRODUCT}} name. Why it holds (inferred): the opening keeps the visual sparse while the voice supplies the subject; the cut at frame 38 enlarges the world.

Retention map (inferred): re-hooks at shots 3, 6, 11, and 16 · payoff in shots 12–17 · call to action in shot 18 · shot 19 holds without a closing fade.

Mix (inferred): title and claim cards alternate with interface evidence; the two long demonstrations consume much of the running time. Rhythm (inferred): compact opening and proof shots lead into the extended examples, then a short invitation and longer signoff.

Rules this editor follows (inferred):

• Cut or hand off when the sentence changes subject; give UI examples room to be read.
• Let high-contrast dark plates reset attention between pale interface sequences.
• Move cards and masked surfaces within a mostly locked frame; the camera table separates those layers.

• Look (inferred): near-white and charcoal reset planes bracket blurred brand-colour fields. The reference uses blue, pink and green atmospheres; map their roles to {{BRAND_BG}} and {{BRAND_ACCENTS}}.
• Graphics (inferred): flat white interface cards with fine borders, tiny functional labels, restrained shadows, and oversized isolated numbers. Critical words must be enlarged for phone delivery.
• Transitions (inferred): mostly hard cuts; the rare fade and vertical whip are recorded in plate 5.
• Branding and end card (inferred): keep {{LOGO}} and {{CTA_URL}} as plain text on the final colour surface through shot 19; no fade to black.

      0s         20s        40s        60s        80s        100s       120s
mean  ▇▅▇██▆▁▄▅█ ████▂▁▅▆▂█ ██▃▃▆▆▆▆▆▆ ▆▆██▅▅▆▆▆▆ ▆▆▆▆▃▅▇▅▅▅ ▅▅▅▅▅▅█▆▆▆ ▆
p95   █████▇▃▇██ ████▂▁▅▆▂█ ██▃▃██████ ████▇▇████ ████▃▅████ ████████▇▇ ▇

:root { --brand-bg: {{BRAND_BG}}; --brand-accent: {{BRAND_ACCENTS}}; --ink: #171717; --paper: #f7f7f7; }
.field { background: radial-gradient(ellipse at 35% 30%, color-mix(in srgb,var(--brand-accent),white 55%),transparent 58%), radial-gradient(ellipse at 78% 76%,var(--brand-accent),var(--brand-bg) 72%); filter: blur(18px); transform: scale(1.08); }
.ui-card { background: color-mix(in srgb,white 96%,var(--brand-bg)); border: 1px solid rgba(0,0,0,.08); border-radius: 18px; box-shadow: 0 18px 54px rgba(0,0,0,.10); }
.dark-grid { background: var(--ink); border-radius: 20px; box-shadow: inset 0 0 110px rgba(0,0,0,.35); }
.glass { backdrop-filter: blur(22px); border: 1px solid rgba(255,255,255,.24); }

Measured: v1 transcript estimates 253 words and 169 wpm while speaking; v2 plate 8 recognises 234 words covering 67% of the film. The difference is a transcription uncertainty, so pace the new script to recorded phrases rather than filling every second.

Voice direction (inferred): use an assured low-register narrator, warm but precise. Give numbers a clean attack; drop slightly at the end of proof lines. Leave the long example in shots 12–13 conversational, with distinct speaker archetypes and natural pauses. Record dry, then mix the bed around speech.

• Music (inferred): light tonal electronics and airy sustained chords under narration, with a darker low note at the proof resets; no steady pulse drives cuts.
• Effects (inferred): tiny logo attack in shot 1, soft card ticks for UI appearances, a short low hit at shot 3, a muted vertical swish at shot 16. Avoid adding a second effect on cue events marked voice? in the measured grid.
• Space (inferred): pull the bed down under speech and let the shot 19 outro decay without a picture fade.

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

• Title (inferred): “Introducing {{PRODUCT}}: {{FEATURE}}” is the functional pattern; keep the proof in the description.
• Cover (inferred): one centred title on a recognisable brand surface, with a small interface fragment only if readable.
• Post copy (inferred): one sentence stating the capability, one verified {{PROOF}}, then {{CTA_URL}}.
• Audience signals: the context records 2,685 views; the small comment sample is too thin to infer retention.

Narration-led (inferred): write about 230–250 words for this length; aim near 115–125 wpm overall and 150–170 wpm in spoken runs. Use the measured shot frames and spoken phrases in the JSON for exact line placement. Captions are short claim or UI text timed to the corresponding words, not a full sentence-by-sentence subtitle layer.

[beat 1 · shots 1–2] Introducing {{PRODUCT}}. One clear sentence promising {{FEATURE}}.
[beat 2 · shots 3–4] State {{PROOF}} in plain language, then specify the comparison.
[beat 3 · shots 5–6] Explain one control the user can try; introduce the second mode.
[beat 4 · shots 7–11] Explain the contextual benefit and one verified measure.
[beat 5 · shot 12] Let two role-based speakers demonstrate the result; conclude with a global-use line.
[beat 6 · shots 13–17] Show a language switch and a detail-sensitive note with real data.
[beat 7 · shots 18–19] Invite action; finish with {{LOGO}} and {{CTA_URL}} as text.

Two routes (inferred)

Minimum: Remotion with CSS 3D. Claude can build the abstract fields, chart cards, interface reconstruction, transitions, captions and synthesised original bed in one code stack. It simplifies fine fluid texture to layered gradients and gives up some organic surface motion. Supply a recorded or TTS narrator and authentic {{UI_ASSETS}}; those are not Claude-alone assets.

Full: recorded voice → HTML/CSS interface capture with Playwright → After Effects or Remotion compositing → audio edit → final encode. Record clean 48 kHz WAV; capture real {{PRODUCT}} screens or build its true UI in code at 1920×1080, 30 fps. Pass transparent UI and type as 1920×1080 PNG sequences at 30 fps, camera and element keyframes as JSON, and 48 kHz WAV stems with a click/cue track into Remotion. Render H.264 1920×1080 at 30 fps and master for the intended platform. No 3D package is needed for these flat surfaces.

Asset honesty: use real product captures or your own UI rebuilt in code; where no image exists, use a plain type tile. Never invent screenshots, claims, results or conversation outcomes. Verify {{PROOF}} and the transcript in shots 12, 13 and 17 against real product output.

Script prompt (inferred)

Write a 230–250 word launch narration for {{PRODUCT}} and {{FEATURE}}. Use only verified {{PROOF}} and authentic {{UI_ASSETS}}. Match beats 1–7 and the shot frames in the blueprint JSON. Give two conversational demonstrations with role-based speakers; keep labels short enough to read on a phone. Avoid imitating the reference copy.

Caption and music prompt (inferred)

Use {{BRAND_FONT}} for short claims at the four plate 6 type levels. Every critical word needs at least 38 px cap height at 1920×1080. Animate only at phrase or interface events. Compose an original airy electronic bed with loose 96 BPM phrasing, no beat-locked edit, leaving speech clear.

How it's built (inferred)

Signature move: evidence becomes interface. Build shot 2 as layered fluid field, centered title, circular mask, then text cards. The measured shot 2 element has a ×0.77 scale and x −10.9%, y −4.7% total over moving segments; the camera stays fixed. Use the plate 5 segments rather than one continuous zoom: a 2.33–2.93 s easing move, a 4.13–4.47 s short slide, and the 4.80–5.67 s left slide. Compose each as transform on the element group with cubic-bezier(.45,0,.55,1); render on ones at 30 fps, 8 subframes for the fast slides, shutter about 0.35 frame. At f232, hard-cut the circular claim to a white text-card state in 0 f; at f268, hard-cut again to the detected-voices layout in 0 f. Keep all copy in a safe interior mask so the later 9:16 layout can be rebuilt.

Other motion and handoffs. Shot 3 enters black at f345 with its measured 3 f fade; its number grows with the plate 5 element scale, then gives way to the chart at f399. Shot 4 chart changes at f455 with a 0 f cut; do not treat it as a long dissolve. Shot 5 cuts transcript to rule composer at f623; type the rule in place, then use its measured −23.6% slide from 21.83–22.83 s and upward move from 23.23–24.17 s. Shot 6 cuts within dark frames at f866 (0 f) and jumps its title layout left by 18.1% at f922 (0 f), followed by the 30.73–31.00 s measured slide. Shot 11 hard-cuts at f1377 from growing square to a waveform panel (0 f). Shot 12 is the second signature sequence: its element region travels x −52.9%, y −17.4% and grows ×1.34 over its measured moving segments. Build dialogue cards over separate gradient fields; use clipping masks on each panel, and preserve the plate 5 stops and starts. At f1952 pull focus out over 10 f to half edge sharpness, then change the language picture at f2029 with a 0 f cut. Shot 13 follows its own stepped card movement; hold the messages still where the element table says static. Shot 16 enters with a 4 f vertical whip at f2797, using directional blur around 24 px at peak and an ease-out exit. Shot 17 types a genuine note within a fixed card; shots 18 and 19 enter by cuts at their measured frames. Use cubic-bezier(.22,1,.36,1) for card arrivals and cubic-bezier(.55,0,1,.45) for departures. The camera rows marked ? are tracking hints: the sheets show mostly fixed frames with moving elements. Do not substitute camera pushes for those moves.

Sheet checks. (inferred) The transition strip shows a new text card at 7.733 s and the detecting-voices layout at 8.933 s; plate 5 records both as cuts inside shot 2. The shot 4 strip shows an abrupt chart style change at 15.167 s, matching f455. The dark title strip shows a layout jump at 30.733 s, matching f922. The sheets show a progressive waveform reveal around 45.9 s; plate 5 classifies its first changed frame as a hard cut at f1377.

Build order

1. Write and record beats 1–7 against the shot frames; lock the speech phrases before animation.
2. Build the brand surfaces, proof charts and truthful UI assets for shots 1–17.
3. Animate the signature handoffs in shots 2 and 12, then the internal cuts and whip listed above.
4. Add the original music bed and discrete cues around the narration.
5. Add beat 7’s text close, render, inspect at phone size, and master audio.

• Shot boundaries and internal changes match plate 4 and plate 5 within ±1 frame; only f345 fades 3 f and f2797 whips 4 f.
• Shot 2’s measured element travel and scale, shot 5’s −23.6% slide, and shot 12’s −52.9% travel are present on moving layers rather than the camera.
• Every viewer-critical word is ≥38 px cap at 1920×1080; adapted versions meet their plate 15 minimum.
• Shot 12’s f1952 focus pull takes 10 f; the end does not fade to black.
• Check selected cue frames against plate 8, keep effects off spoken stress, and avoid a forced beat grid.
• YouTube master measures near −14 LUFS integrated and ≤−1 dBTP; voice stays intelligible over the bed.
• Claims, screens and conversation results come from actual {{PRODUCT}} material.

Brand slots

What this film avoids (inferred): stock footage, talking head, elaborate device mockups, heavy interface chrome, a rapid feature list, and beat-synchronised cutting.

Brand-distance warning (inferred): the reference’s paired-bar mark plus pale blue fluid field and the same title wording would read as its brand. Change the visual mark and surface geometry first.

Adapt the aspect (inferred)

• 9:16: Re-lay out: centre crop keeps only 31.6% of width and 11/83 measured boxes. Stack cards vertically; put titles in the middle 70% of width, keep critical text ≥22 px cap, and clear the top account strip and bottom caption/action UI.
• 1:1: Re-lay out: centre crop keeps 56.2% of width and 49/83 boxes. Narrow dialogues to two stacked cards, move charts inward, keep critical text ≥22 px cap, and leave roughly 10% top/bottom clear for platform overlays.

Adapt the length (inferred)

• 15 s: Keep the hook, one proof card, the signature interface handoff, and beat 7; drop optional intermediate demonstrations and shorten scalable narration.
• 30 s: Keep one example from beats 5–6, the second-product reset, and beat 7; condense repeated proof states.
• 60 s: Keep both demonstrations but reduce their conversational turns; let additional proof live in new, verified examples rather than stretching static holds.

The creator’s signature (don’t copy): its mark, names, slogans, fluid palette and track. The transferable format (do copy): spare title → measured proof → truthful UI demonstrations → concise invitation.

Source: https://www.youtube.com/watch?v=cw_NRHMOPtw · measured 2026-11-11 · (measured) = read off the film · (inferred) = interpretation