vid-sizzle-reel

Generates a high-energy sizzle reel or hype video from brand assets and key messages. Pipeline: HyperFrames HTML composition (GSAP timelines) → headless Chromium → FFmpeg H.264 MP4. No Runway. No Pika. No AI video APIs. Zero runtime cost beyond Node.js + FFmpeg.

Differentiation from `vid-product-launch`:

• vid-product-launch tells a narrative story — one earned reveal, five sections, tension arc
• vid-sizzle-reel creates energy and excitement — fast cuts, music-first, no single story beat

Critical Rules (read before every generation)

1. Write the key messages yourself — or ask the user to write them. These are the exact phrases that appear on screen. "AI will figure it out" produces filler. 3-5 sharp lines is the entire content of a sizzle reel.

1. Brand name appears ONLY in the Land section. Cold Open and Build create tension without naming the brand. The Land section is the payoff.

1. Declare the rhythm pattern in a comment at the top of the script before writing any HTML. Format: // RHYTHM: flash-sequence | Cold[0-5s]: stat | Build[5-40s]: msg(1.5s)|flash|... | Peak[40-55s]: ... | Land[55-60s]: logo+CTA. Every timing decision checks against this.

1. Build end-state layout in CSS first — no GSAP yet. Position every element at its most-visible moment in static CSS. Then add GSAP entrances with gsap.from() and exits with gsap.to(). Never position at animated start state.

1. `class="clip"` on every timed element. Required by HyperFrames. Without it, the element is invisible to the player.

1. `data-start`, `data-duration`, `data-track-index` on every clip. No exceptions. Same-track clips cannot overlap — use different track indices.

1. No `Math.random()`. HyperFrames requires deterministic compositions. Use seeded PRNG (mulberry32) for any pseudo-random values.

1. All GSAP timelines start `{ paused: true }` and register to `window.__timelines["comp-id"]`. The HyperFrames player controls playback — never call tl.play().

1. Never `position: absolute` on `.scene-content` containers. Use width:100%; height:100%; padding:Npx; display:flex. Reserve position:absolute for decorative elements only.

1. Music: separate `<audio>` element with `data-track-index`. Video must be muted playsinline. Never put audio in a <video> element.

1. Read `references/cut-patterns.md` AND `references/tone-presets.md` before generating any HTML.

1. Never dump HTML in chat. Save to file. Show summary only.

Step 1: Intake

Required:

• key_messages — 3-5 punchy lines to flash on screen (user must write these)

Optional parameters and defaults:

cut_style auto-defaults: energetic → fast; professional → fast (2s); cinematic → cinematic; emotional → mixed

If `key_messages` is missing, ask exactly:

"To generate the sizzle reel, I need your key messages — the exact phrases that will flash on screen. Give me 3-5 punchy lines. These are the most important thing you're writing for this video.

>

Examples: 'Used by 500+ growth teams' / 'From days to minutes' / 'Works with Claude, Codex, Gemini'

>

Also useful: tone (energetic/cinematic/emotional/professional), duration (30/60/90s), any brand colors or logo URL."

If key_messages are present → proceed to Step 2 immediately.

Step 2: Install HyperFrames

Install HyperFrames skills (first time only — skip if already installed):

npx skills add heygen-com/hyperframes

Verify environment:

node --version   # must be >= 22
ffmpeg -version  # must be present

Scaffold the project:

npx hyperframes init sizzle-[slug] --example kinetic-type --non-interactive
cd sizzle-[slug]

Slug: kebab-case from end_card brand name or first key_message, max 25 chars.

Step 3: Internal Architecture (never shown to user)

1. Derive cut-point timing from `duration` + `cut_style`:

Convert every boundary to milliseconds. Calculate per-scene duration from cut_style:

• fast: 1.2s-1.8s per message (use 1.5s default)
• cinematic: 3.5s-5.0s per message (use 4.0s default)
• mixed: 2.0s-4.0s varying by section energy

2. Map key_messages to scenes:

• Cold Open (1 scene): one impact stat or question — NO brand name, builds intrigue
• Build: one scene per key_message
• Peak: repeat the strongest message at double scale, or combine 2 messages in rapid flash
• Land: brand name + tagline + CTA URL

3. Detect beat sync if music provided:

npx hyperframes tts --analyze-beats [music-file]

If unavailable, calculate from BPM: beat_ms = 60000 / BPM. Align cuts to every 2nd beat.

4. Select tone preset — read references/tone-presets.md for CSS tokens.

5. Declare rhythm pattern — write comment at top of script:

// RHYTHM: flash-sequence
// Cold Open [0-5000ms]: stat
// Build [5000-40000ms]: msg-1(1500ms)|flash|msg-2(1500ms)|flash|msg-3(1500ms)|flash|msg-4(1500ms)|flash|build-accum(10000ms)
// Peak [40000-55000ms]: msg-strongest(1200ms)|flash|msg-2nd(1200ms)|flash|HOLD(5600ms)
// Land [55000-60000ms]: logo+tagline+cta

6. Determine pixel dimensions:

• 9:16 → W=1080, H=1920 (default — Instagram Reels, TikTok, YouTube Shorts)
• 16:9 → W=1920, H=1080

Step 4: HyperFrames HTML Composition

Read references/cut-patterns.md and the HyperFrames skill (/hyperframes) before writing any code.

Required HTML skeleton:

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
[font CDN link from tone preset]
<style>
:root {
  [all CSS tokens from tone preset]
}

*, *::before, *::after { margin: 0; padding: 0; box-sizing: border-box; }
</style>
</head>
<body>
<div
  data-composition-id="sizzle-[slug]"
  data-width="[W]"
  data-height="[H]"
  style="position:relative;width:[W]px;height:[H]px;overflow:hidden;background:var(--bg)"
>

  [if cut-flash: true]
  <div id="cut-flash"
    style="position:absolute;inset:0;background:var(--cut-flash-bg);opacity:0;pointer-events:none;z-index:200"></div>
  [end if]

  [if film-grain: true]
  <canvas id="grain-overlay"
    width="240" height="135"
    style="position:absolute;inset:0;width:[W]px;height:[H]px;pointer-events:none;opacity:0.022;mix-blend-mode:overlay;z-index:50"></canvas>
  [end if]

  [if vignette: true]
  <div id="vignette-overlay"
    style="position:absolute;inset:0;background:radial-gradient(ellipse at center,transparent 35%,rgba(0,0,0,0.65) 100%);pointer-events:none;z-index:51"></div>
  [end if]

  <!-- Scenes — one div per scene -->
  <div id="scene-cold"
    class="clip"
    data-start="0"
    data-duration="5"
    data-track-index="0"
    style="position:absolute;inset:0;opacity:0"
  >
    <div class="scene-content" style="width:100%;height:100%;padding:120px;display:flex;flex-direction:column;align-items:center;justify-content:center;box-sizing:border-box">
      [cold open content — one stat or impact phrase, no brand name]
    </div>
  </div>

  [one div per Build scene, data-start increments by cut duration]
  [Peak scene(s)]
  [Land scene — brand name + tagline + CTA]

  [if music provided]
  <audio id="music"
    class="clip"
    data-start="0"
    data-duration="[total_duration]"
    data-track-index="10"
    data-volume="0.85"
    src="[music_path]"
  ></audio>
  [end if]

  <script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/gsap.min.js"></script>
  <script>
  // ── RHYTHM DECLARATION ──────────────────────────────────────────────────────
  // RHYTHM: [flash-sequence | cinematic-hold | beat-sync | mixed]
  // Cold Open [0-Nms]: ...
  // Build [N-Nms]: ...
  // Peak [N-Nms]: ...
  // Land [N-Nms]: ...

  // ── SEEDED PRNG (for grain only — no other random values) ───────────────────
  function mulberry32(seed) {
    return function() {
      seed |= 0; seed = seed + 0x6D2B79F5 | 0;
      let t = Math.imul(seed ^ seed >>> 15, 1 | seed);
      t = t + Math.imul(t ^ t >>> 7, 61 | t) ^ t;
      return ((t ^ t >>> 14) >>> 0) / 4294967296;
    };
  }

  [if film-grain: true]
  // ── FILM GRAIN (cinematic/emotional only) ───────────────────────────────────
  const grainCanvas = document.getElementById('grain-overlay');
  const grainCtx = grainCanvas ? grainCanvas.getContext('2d') : null;
  // Grain updated by GSAP ticker (not per-frame capture — HyperFrames handles this)
  [end if]

  // ── GSAP TIMELINE ───────────────────────────────────────────────────────────
  const tl = gsap.timeline({ paused: true });

  // ── COLD OPEN ───────────────────────────────────────────────────────────────
  [cold open entrance and exit tweens — see cut-patterns.md for exact pattern]

  // ── BUILD SECTION ───────────────────────────────────────────────────────────
  [per-message tweens — flash-sequence or cinematic-hold from cut-patterns.md]

  // ── PEAK SECTION ─────────────────────────────────────────────────────────────
  [peak tweens — same pattern as build but tighter timing]

  // ── LAND SECTION ─────────────────────────────────────────────────────────────
  [land entrances — logo, tagline, CTA. No exit — holds to end]

  // ── TIMELINE REGISTRATION ────────────────────────────────────────────────────
  window.__timelines = window.__timelines || {};
  window.__timelines["sizzle-[slug]"] = tl;
  </script>
</div>
</body>
</html>

Design quality rules:

• Cold open stat: ≥160px font size for 16:9, ≥100px for 9:16
• Key messages: ≥80px font size — readable at a glance
• Land section brand name: ≥120px for 16:9
• All colors from preset token variables — no free hex except var(--accent) literal hex inside canvas fillStyle
• scene-content always uses padding + flex — never position:absolute on content container
• class="clip" on every element with data-start

Step 5: Self-QA (fix every failure before Step 6)

Content:

• [ ] key_messages — each phrase appears exactly once, no duplicates
• [ ] Brand name appears ONLY in Land section (not cold open, not build, not peak)
• [ ] Cold open contains NO brand name and NO product name — only stat or impact phrase
• [ ] Land section has: brand name + tagline + CTA URL (three elements minimum)
• [ ] end_card CTA URL present on final frame

HyperFrames contract:

• [ ] class="clip" on every element with data-start
• [ ] data-start, data-duration, data-track-index on every clip
• [ ] data-composition-id on the root composition div
• [ ] window.__timelines["sizzle-[slug]"] = tl registered
• [ ] All timelines start { paused: true }
• [ ] No Math.random() — seeded PRNG only
• [ ] No tl.play() or tl.resume() calls

GSAP / Layout:

• [ ] gsap.from() for all entrances (FROM offscreen TO CSS position)
• [ ] gsap.to() for all exits (FROM CSS position TO offscreen)
• [ ] No position:absolute on .scene-content containers
• [ ] Rhythm pattern declared in comment at top of script

Cut timing:

• [ ] fast tone: average cut duration ≤ 2.0s
• [ ] cinematic tone: average cut duration ≥ 3.0s
• [ ] Cut flash fires at every cut boundary for energetic tone (60ms, opacity 0→1→0)
• [ ] Same-track clips do NOT overlap (check all data-track-index groupings)

SFX:

• [ ] window.__sfxTimeline present and placed before timeline registration
• [ ] word-hit event per key message
• [ ] whoosh at Build → Peak transition boundary
• [ ] cta-chime at Land start
• [ ] No SFX fires before 500ms

Step 6: Lint, Render, Output

npx hyperframes lint
npx hyperframes inspect --json

Fix all lint errors. Fix all inspect errors (text overflow, clip escaping containers). Then render:

npx hyperframes render --output sizzle/[slug]/sizzle-reel.mp4

Output:

## Sizzle Reel: [brand name from end_card]
Tone: [tone] | Duration: [N]s | Cut style: [cut_style] | Aspect: [ratio]
Rhythm: [declared pattern]

Sections
  Cold Open [0-Ns]:  [impact stat or phrase]
  Build     [N-Ns]:  [N key messages listed]
  Peak      [N-Ns]:  [strongest message at max scale]
  Land      [N-Ns]:  [brand name] + "[tagline]" + [CTA URL]

Files
  Source: sizzle/[slug]/index.html
  Output: sizzle/[slug]/sizzle-reel.mp4

SFX Timeline (embed in every composition)

The same FFmpeg-synthesized SFX system as vid-product-launch. Place window.__sfxTimeline immediately before the GSAP timeline registration.

SFX type reference for sizzle reel:

window.__sfxTimeline = [
  // one word-hit per key_message (computed from scene start times)
  { ms: MSG_1_START_MS,           sfx: 'word-hit',   vol: 0.50 },
  { ms: MSG_2_START_MS,           sfx: 'word-hit',   vol: 0.50 },
  { ms: MSG_3_START_MS,           sfx: 'word-hit',   vol: 0.50 },
  // ... repeat for each message
  { ms: PEAK_START_MS - 700,      sfx: 'whoosh',     vol: 0.55 },
  { ms: LAND_START_MS,            sfx: 'cta-chime',  vol: 0.65 },
];

Compute all ms values from declared section timing constants. Never hardcode.

Prompt Tips (show when user asks for guidance)

"Write the key messages yourself. They are the entire content of this video. 3 sharp lines beats 10 vague ones every time."

>

"Choose music before choosing visuals. The BPM determines the cut style. The cut style determines every timing decision."

>

"Cold open: one number. Not a sentence, not a tagline — one specific, surprising number. It earns 5 seconds of attention."

>

"The Land section is not a credits page. Logo + one line + one URL. Anything more dilutes the CTA."

>

"Match tone to channel. Energetic for social and Product Hunt. Cinematic for B2C premium and investor decks. Professional for enterprise conferences. Emotional for crowdfunding."

Good:

Sizzle reel, 60 seconds. Tone: energetic. Key messages: ['AI skills, ready to install' /
'52+ skills across GTM, content, research' / 'Works with Claude, Codex, Gemini' /
'Zero setup. Instant value.']. Music: 128bpm electronic. Cut style: fast.
End card: OpenDirectory + 'AI skills, ready to install' + 'opendirectory.dev'. Aspect: 9:16.

Bad:

hype video for our company