{
  "_comment": "MasterClass 'App Sizzle (3D CGI)' \u2014 the worked example (clients/masterclass/ad-runs/run-03-run-03). Copy to config.json and edit. A ~22.6s premium 3D-CGI app-demo sizzle: a gold-trimmed phone floats in a smoky-black studio; 6 app features demo in sequence \u2014 each bursts REAL App Store UI elements outward in 3D with amaranth rim-light + bokeh, then settles back into the phone at the climax. 1080x1920 9:16. ALL on-screen UI is composited via PIL from real App Store screenshots \u2014 NEVER AI-rendered. Per beat: a nano_banana_2 CGI plate (blank-glow phone + placeholder burst-out shapes) -> PIL screen composite (real screenshot onto the bezel) -> kling3_0 i2v steady-float clip (Ken-Burns FFmpeg fallback if Kling garbles the burst-out UI). VO durations are LOCKED from the rendered ElevenLabs audio before the timeline is cut. Final is sped 1.15x + anti-AI grain.",
  "brand_name": "MasterClass",
  "width": 1080,
  "height": 1920,
  "fps": 24,
  "final_speed": 1.15,
  "target_duration_sec": 22.6,
  "studio_look": {
    "background_hex": "#0A0A0A",
    "rim_light_hex": "#E32652",
    "rim_light_direction": "upper-left",
    "phone": "single iPhone 15 Pro, gold/champagne titanium frame, floating center-frame at ~5-12 degree tilt toward camera, screen ON showing only a uniform warm glow (UI composited in post)",
    "bokeh": "warm amaranth sparkles + a few cool-white highlight orbs drifting in the empty space",
    "register": "Apple-keynote 3D CGI product render \u2014 glossy bezel reflections, soft ground shadow, premium luxury grade, no people, no hands, no faces, no English text on the plate",
    "reference_pin": "Pinterest pin 1119566788643692467 ('Ultra-Smooth 3D CGI Advertisement'), re-paletted from champagne studio -> smoky-black + amaranth"
  },
  "cgi_plate": {
    "model": "nano_banana_2",
    "provider": "fal",
    "base_prompt": "Premium 3D CGI advertising still, 9:16 vertical product render. A modern smartphone is the subject, photographed from DIRECTLY IN FRONT so we see the FULL DISPLAY SCREEN facing us. The phone floats centered in frame, tilted only 5 degrees left. The DISPLAY SCREEN is ON and shows ONLY a uniform soft warm glow \u2014 completely blank, NO icons, NO UI, NO text. Slim gold/champagne titanium edges catching warm light. SETTING: smoky-black studio backdrop (#0A0A0A), soft amaranth-red rim-light (#E32652) raking from upper-left. Warm amaranth bokeh sparkles + cool-white highlight orbs drift around the phone. Faint window-stripe shadows on the dark floor. LOOK: Apple-keynote 3D CGI product render. Glossy bezel reflections. Soft realistic ground shadow. Premium luxury feel. NEGATIVE: NO back of phone, NO camera lenses, NO people, NO faces, NO English text, NO logos, NO multiple phones, NO furniture.",
    "anchor": "beat 1 plate is generated first, then beats 2-6 are anchored on it (--anchor scene-01-plate) so the phone/studio stay identical across beats; each beat appends its burst-out element layer to base_prompt"
  },
  "screen_composite": {
    "method": "PIL \u2014 auto-detect the bright warm phone-screen bbox in the plate, resize the real App Store PNG to fit, warm-tint 6%, brightness +5%, feather into a rounded-rect mask so it sits inside the bezel",
    "hard_rule": "the on-screen UI (instructor faces, course cards, video tiles, brand poster) is ALWAYS the real App Store screenshot \u2014 never AI-rendered. Instructor portraits are the real Cuban/Ramsay/Perel App Store faces."
  },
  "beats": [
    {
      "id": "beat-01-discovery",
      "label": "Discovery \u2014 For You feed",
      "screen": "screen-01.png",
      "screen_role": "MC 'For You' tab",
      "burst_out": "3 glass-morphic instructor portrait tiles (Cuban, Ramsay, Perel \u2014 REAL faces from screen-01) 3D-pop out toward camera at three depths with parallax",
      "burst_out_source": "real_screenshot_pil",
      "plate_element_layer": "placeholder rectangular glass-morphic tiles popping out toward camera at different depths; real portraits composited in PIL post",
      "kling_motion": "Steady cinematic locked-tripod 3D camera, no shake no wobble. Phone holds centered; three instructor portrait tiles gently 3D-pop outward toward camera, settling at three depths with parallax. Warm amaranth bokeh drifts upward. Phone screen content stays exactly as shown.",
      "kenburns_fallback": "slow zoom 1.00->1.06 push-in on the full-bleed App Store screen (build_v2_clips.py)",
      "vo": "Two hundred world-class instructors. One feed.",
      "vo_actual_duration_sec": 3.71,
      "t_start_sec": 0.0,
      "t_end_sec": 4.11
    },
    {
      "id": "beat-02-library",
      "label": "Library \u2014 catalog breadth",
      "screen": "screen-02.png",
      "screen_role": "MC Library",
      "burst_out": "a deck of 5-7 course thumbnail cards fans out in 3D like a magician spreading cards, each catching the amaranth rim-light",
      "burst_out_source": "real_screenshot_pil",
      "plate_element_layer": "fan of 5-7 rectangular glass-morphic placeholder cards spread like a hand of playing cards at varied depth/rotation; real course tiles composited from screen-02 in PIL",
      "kling_motion": "Phone holds steady at center; the fan of course-thumbnail cards spreads wider across the screen plane like a magician's deck cascading, each card rotating a few degrees toward camera and catching the rim-light. Phone screen content stays exactly as shown.",
      "kenburns_fallback": "slow zoom 1.00->1.06 push-in on screen-02",
      "vo": "Cooking. Writing. Photography. Business. Strategy.",
      "vo_actual_duration_sec": 4.99,
      "t_start_sec": 4.11,
      "t_end_sec": 9.4
    },
    {
      "id": "beat-03-bite-size",
      "label": "Bite-size lessons / Up Next",
      "screen": "screen-03.png",
      "screen_role": "MC Lesson player",
      "burst_out": "one large video-thumbnail tile ejects forward toward camera + two 'Up Next' pills cascade in behind from the right edge",
      "burst_out_source": "real_screenshot_pil",
      "plate_element_layer": "one large portrait glass-morphic video tile ejecting forward + two stacked pill placeholders cascading from the right",
      "kling_motion": "Phone holds steady; the video tile eases forward toward camera with a slight grow-in; the two Up-Next pills gently cascade and settle; sparkles drift down. Phone screen content stays exactly as shown.",
      "kenburns_fallback": "slow zoom 1.00->1.06 push-in on screen-03",
      "vo": "Lessons that fit a coffee break.",
      "vo_actual_duration_sec": 1.72,
      "t_start_sec": 9.4,
      "t_end_sec": 11.62
    },
    {
      "id": "beat-04-cross-device",
      "label": "Cross-device orbit",
      "screen": "screen-04.png",
      "screen_role": "MC 'My Learning'",
      "burst_out": "small generic 3D iPad + Apple TV remote + MacBook mockups orbit the phone clockwise (one rotation across ~3.5s), parallax so iPad passes in front, remote behind",
      "burst_out_source": "ai_generic_devices",
      "plate_element_layer": "generic iPad/TV-remote/laptop mockups orbiting the phone at mid-depth with BLANK screens (no UI, no text \u2014 would conflict with the phone screen)",
      "kling_motion": "Phone holds locked at center; the three device mockups slowly rotate clockwise in orbit at constant velocity, parallaxing \u2014 iPad front-left nearer camera, TV remote behind, MacBook crossing front-right. Phone screen content stays exactly as shown.",
      "kenburns_fallback": "slow zoom 1.00->1.06 push-in on screen-04",
      "vo": "Mobile. Tablet. Web. TV.",
      "vo_actual_duration_sec": 3.97,
      "t_start_sec": 11.62,
      "t_end_sec": 15.99
    },
    {
      "id": "beat-05-audio-offline",
      "label": "Audio + Offline",
      "screen": "screen-05.png",
      "screen_role": "MC Audio player",
      "burst_out": "a large glass headphone icon + a glass download-arrow icon eject outward from the phone with soft glow trails, settling left and right",
      "burst_out_source": "ai_generic_icons",
      "plate_element_layer": "large translucent glass headphone icon (left) + glass download-arrow icon (right) ejecting outward, glowing from within, PURELY ICONIC SHAPES with NO text labels",
      "kling_motion": "Phone holds steady; the glass headphone and download-arrow icons glow brighter and pulse outward a tiny amount with warm glow trails toward the phone; sparkles drift. Phone screen content stays exactly as shown.",
      "kenburns_fallback": "slow zoom 1.00->1.06 push-in on screen-05",
      "vo": "Listen on the run. Download for the flight.",
      "vo_actual_duration_sec": 2.35,
      "t_start_sec": 15.99,
      "t_end_sec": 18.94
    },
    {
      "id": "beat-06-the-pitch",
      "label": "The pitch \u2014 CLIMAX",
      "screen": "screen-06.png",
      "screen_role": "MC brand poster ('LEARN FROM THE BEST. BE YOUR BEST.')",
      "burst_out": "NO new burst \u2014 all floating elements from prior beats gently collapse BACK into the phone screen; amaranth rim-light intensifies, bokeh thickens; optional baked burst-climax portrait tiles ring the phone (build_burst_climax.py)",
      "burst_out_source": "collapse_in",
      "plate_element_layer": "phone tilted ~9 degrees toward camera (subtle push-in), no floating elements outside the phone (collapsed in), rim-light INTENSIFIED, denser bokeh \u2014 reads as the climax hero composition",
      "kling_motion": "Phone holds in a hero composition; the amaranth halo slowly intensifies and bokeh thickens for a climax payoff; the phone subtly nudges a few degrees closer to camera (very slow push-in). Phone screen content stays exactly as shown.",
      "kenburns_fallback": "heavier zoom 1.00->1.10 push-in on the scene-06 CGI plate (visual punctuation)",
      "vo": "Two hundred plus classes. One app.",
      "vo_actual_duration_sec": 3.0,
      "vo_gain_db": 2,
      "is_climax": true,
      "t_start_sec": 18.94,
      "t_end_sec": 23.0
    }
  ],
  "end_card": {
    "method": "PIL composite (no AI)",
    "canvas_hex": "#0A0A0A",
    "amaranth_bar": {
      "hex": "#E32652",
      "width_px": 800,
      "height_px": 6,
      "y": 960
    },
    "wordmark": {
      "asset": "masterclass-wordmark.svg",
      "width_px": 480,
      "baseline_y": 900
    },
    "tagline": {
      "text": "Learn from the best.",
      "font": "Montserrat",
      "size_px": 56,
      "weight": "medium",
      "color": "#FFFFFF",
      "y": 1020
    },
    "vignette": true,
    "dwell_sec": 3.0,
    "captions_suppressed": true
  },
  "voiceover": {
    "provider": "elevenlabs",
    "voice": "Eryn",
    "voice_id": "dMyQqiVXTU80dDl2eNK8",
    "model_id": "eleven_v3",
    "voice_settings": {
      "stability": 0.45,
      "similarity_boost": 0.75,
      "style": 0.1,
      "use_speaker_boost": true
    },
    "direction": "Confident, declarative, restrained. Apple-keynote spec-sheet cadence. NOT lyrical, NOT breathy. Each line lands beat-for-beat with the on-screen feature.",
    "note": "render all 6 cues FIRST, ffprobe each MP3, then lock the beat windows from the MEASURED durations (never from planned word counts)"
  },
  "music": {
    "provider": "elevenlabs_music",
    "prompt": "A premium tech advertisement underscore, cinematic and modern. Slow build over ~27 seconds. Sparse low piano notes establish the first 4 seconds. Add soft warm synth pad at 4s. Gentle pulsing rhythm (no drums \u2014 synth pulses) enters at 12s. Subtle string swell at 18s. Full warm string + piano resolution at 22s. NO vocals, NO lyrics, NO chant. Hopeful, confident, premium. Apple keynote register. Key of D major.",
    "trim_intro_sec": 2.5,
    "force_instrumental": true,
    "no_artist_names": true
  },
  "audio_mix": {
    "vo_bus": "loudnorm -23 LUFS, de-ess, high-pass 80Hz",
    "music_bus": "loudnorm -23 LUFS, sidechain-compress with VO as key (ratio 20:1, attack 5ms, release 250ms), duck -9dB under VO",
    "climax_boost": "+2dB VO on beat-06",
    "master_bus": "loudnorm -14 LUFS (IG/TikTok safe), -1.5 dBTP true peak"
  },
  "captions": {
    "style": "klap_restrained_white_sans",
    "treatment": "line-by-line, one cue per beat, NO karaoke pills (MC premium register)",
    "font": "Helvetica Neue 44, white, thin outline, lower-third",
    "suppress_on_end_card": true,
    "timing": "fade in +0.05s after VO start, fade out +0.20s after VO end"
  },
  "finalize": {
    "speed": "1.15x (setpts=PTS/1.15 on video, atempo=1.15 on audio, pitch preserved) -> 26.0s master becomes ~22.6s",
    "grain": "anti-AI grain pass on the final master",
    "codec": "1080x1920 H.264, AAC 256k"
  },
  "post_production": {
    "music": {
      "default": "on",
      "note": "cinematic bed under the VO, default on"
    },
    "captions": {
      "default": "on",
      "note": "restrained Klap line-by-line captions, default on"
    },
    "end_card": {
      "default": "on",
      "note": "real-wordmark brand end card, default on"
    }
  }
}

# PIPELINE — config → source steps

This molecule ships the **recipe** (`config.example.json`), not a re-built runnable
pipeline. The engine is the working scripts from the source run
`clients/masterclass/ad-runs/run-03-run-03/working/` (plus its `edits/` finalize).
This maps each config field to the step that consumes it.

## Order of operations

VO first (locks the timeline) → CGI plates → PIL screen composites → burst-climax →
Kling clips (Ken-Burns fallback per garbled beat) → end card → captions → mix → 1.15x + grain.

## Field → step map

| config field | source step | what it does |
|---|---|---|
| `voiceover.*` (voice_id, model_id, voice_settings, beats[].vo) | `render_eryn_vo.sh` | One curl per beat → MP3 → `ffprobe` duration. The measured durations become `beats[].vo_actual_duration_sec` and drive the beat windows. |
| `studio_look.*`, `cgi_plate.*`, `beats[].plate_element_layer` | `gen_plates.py` | nano_banana_2: beat-1 plate first (blank warm-glow phone in a smoky-black + amaranth studio), then beats 2-6 `--anchor`ed on it, each appending its per-beat burst-out placeholder layer. Screen is left blank on purpose. |
| `beats[].screen`, `screen_composite.*` | `composite_screens.py` | PIL: auto-detect the bright phone-screen bbox in each plate, resize + warm-tint + brightness-bump + feather the **real** App Store screenshot into the bezel → `scene-NN-composite.png`. Never AI-render the UI. |
| `beats[].burst_out` where `burst_out_source: real_screenshot_pil` (climax portraits) | `build_burst_climax.py` | PIL: real instructor portrait tiles with rim-light baked BEFORE rotation (so frame + portrait rotate as one unit) + amaranth glow + drop shadow, ringed around the climax plate. |
| `beats[].kling_motion`, `cgi_plate.provider` | `gen_clips.py` | Kling 3.0 i2v (fal): each composite → a steady-float clip; burst-out pops/settles while the phone + screen stay locked (`--mode std`, no-shake negative prompt). PAID, largest spend. |
| `beats[].kenburns_fallback`, `fps` | `build_v2_clips.py` | FFmpeg `zoompan` Ken-Burns push-in per beat (heavier zoom on the climax) — the **fallback** for any beat Kling garbles. This is the path the shipped demo master used. |
| `end_card.*` | PIL end-card composite (in `build_v2_clips.py` / edits) | Smoky-black canvas + amaranth bar + wordmark SVG (480px) + tagline (Montserrat 56px) + vignette → a static `end-card.png` held for `dwell_sec`. Captions suppressed here. |
| `captions.*` | `build_captions.py` | Restrained line-by-line ASS: one cue per beat, offsets from the measured VO windows, fade in +0.05s / out +0.20s, no karaoke pills, suppressed on the end card. |
| `music.*`, `audio_mix.*` | ElevenLabs Music + FFmpeg mix (`audio/`) | Premium-tech bed → trim the sparse intro → loudnorm → sidechain-duck under VO (+2dB VO on the climax) → master loudnorm -14 LUFS. |
| `finalize.*` (`speed`, `grain`, `codec`) | FFmpeg finalize (`edits/`) | Concat the beat clips + end card, then `setpts=PTS/1.15` (video) + `atempo=1.15` (audio, pitch preserved) → ~22.6s, plus the anti-AI grain pass → `master-final-v4-compressed.mp4`. |

## Notes

- **Two model surfaces only:** nano_banana_2 (plates) + Kling 3.0 (float clips). Everything
  else (screen composites, burst-climax, end card, captions, mix, finalize) is free/deterministic.
- **The demo master is the Ken-Burns cut.** `build_v2_clips.py` (push-in on the real screens +
  CGI climax plate) is what shipped as `master-final-v4-compressed.mp4`; `gen_clips.py` is the
  intended Kling primary, kept for beats that hold up under i2v.
- **Anchor discipline:** all six plates share the beat-1 phone/studio via `--anchor`, so the
  sequence reads as one continuous product film.

# render-cgi-sizzle — FREE assembly

This capability ships the **recipe** (`config.example.json`) + a config→step map
(`PIPELINE.md`) for the 3D-CGI app-sizzle format, plus this note documenting the
**FREE, deterministic assembly** — everything between the paid model calls. The paid
steps (nano-banana CGI plates, Kling i2v float clips, ElevenLabs VO + music) are separate
capabilities; the recipe orchestrates and gates them. Everything below is $0
(PIL + FFmpeg), keeps the real UI + wordmark pixel-crisp, and never makes an AI or paid call.

## What's FREE here

### 1. PIL real-UI composite (the format's whole credibility)
The nano-banana plate renders only the **phone shell, smoky-black studio, amaranth
rim-light, bokeh, and placeholder burst-out shapes** — the screen is left a blank warm
glow on purpose. Then, deterministically in PIL:

- **Screen composite** — auto-detect the bright warm phone-screen bbox in each plate,
  resize the **real App Store screenshot** to fit, warm-tint ~6%, brightness +5%, and
  feather it into a rounded-rect mask so it sits inside the bezel → `scene-NN-composite.png`.
  The on-screen UI (feeds, cards, players, instructor faces) is ALWAYS the real screenshot —
  never AI-rendered, so no claim can be invented and nothing reads fake.
- **Burst-out overlays** — where the burst-out element is real UI (e.g. the climax
  instructor portrait tiles), bake the real portraits as glass-morphic tiles with the
  rim-light baked BEFORE rotation (so frame + portrait rotate as one unit) + amaranth glow +
  drop shadow, ringed around the plate. Where the burst-out is a generic device/glass icon,
  the AI plate's placeholder shape is used as-is (no real UI to protect).

### 2. Ken-Burns FFmpeg fallback (per beat)
Kling 3.0 i2v is the intended primary for the steady-float motion, but any beat where Kling
garbles the burst-out UI, distorts the phone, or animates the screen content drops to a FREE
Ken-Burns push-in: FFmpeg `zoompan` slow zoom (`1.00→1.06` on feature beats, heavier
`1.00→1.10` on the climax) over the composite. The shipped demo master used this path for the
feature beats — don't ship a garbled Kling beat when the free push-in is clean.

### 3. Dice + intercut concat
Normalize every beat clip to one fps/codec, then concat the beats in order + the end-card
clip. Hard assembly, no dissolves; the timeline is locked from the **measured VO durations**
(each beat window = its `ffprobe`'d MP3 length), so the cut lands beat-for-beat with the VO.

### 4. Audio mix (sidechain duck + loudnorm)
- VO bus: `loudnorm -23 LUFS`, de-ess, high-pass 80Hz.
- Music bus: premium-tech bed → trim the sparse intro (`music.trim_intro_sec`, default 2.5s)
  → `loudnorm -23 LUFS` → **sidechain-compress with the VO as key** (ratio 20:1, attack 5ms,
  release 250ms), ducking ~-9dB under VO; +2dB VO on the climax beat.
- Master bus: `loudnorm -14 LUFS` (IG/TikTok safe), -1.5 dBTP true peak.

### 5. End card (PIL, never AI)
Smoky-black canvas + amaranth bar + the brand's **real wordmark** (SVG/PNG, never
AI-rendered) + tagline (Montserrat) + vignette → a static `end-card.png` held for
`end_card.dwell_sec`. Captions are suppressed here (the composited text is the message).

### 6. Finalize — 1.15x speed + anti-AI grain
Concat done, apply `setpts=PTS/1.15` (video) + `atempo=1.15` (audio, pitch preserved) to
tighten the ~26s cut to ~22.6s, then an anti-AI grain pass → the H.264 (+ AAC) master.

## Contract

- Deterministic + FREE (PIL + FFmpeg); no paid calls, no AI-rendered UI/faces/wordmark.
- The screen is ALWAYS the real App Store screenshot composited via PIL; the plate's screen
  is left blank on purpose. If AI leaks UI/a face onto the screen, regenerate the plate with a
  blank warm-glow screen and re-composite.
- Phone/studio stay identical across beats (the paid plates are `--anchor`ed on beat 1;
  preserved here). Ken-Burns is the per-beat fallback for any garbled Kling beat.
- Timeline is locked from the measured VO durations, never planned word counts.
- The paid steps — nano-banana plates, Kling 3.0 i2v beats, ElevenLabs VO + music — are
  separate capabilities (create-image-fal, create-video-fal, create-vo-elevenlabs,
  create-music-elevenlabs); the recipe orchestrates them and gates the spend.

See `PIPELINE.md` for the full config-field → source-step map.

# Smoke Test

Copy `scripts/config.example.json` → `config.json`, follow `scripts/PIPELINE.md` for the
cgi-app-sizzle format — the FREE assembly documented in `scripts/README.md` runs the PIL
real-UI composite, Ken-Burns FFmpeg fallback, dice+intercut concat, sidechain-ducked mix,
PIL end card, and 1.15x speed + grain finalize. Output 1080x1920, ~22.6s, deterministic.

Pass when:
- **PIL screen composite** feathers the REAL App Store screenshot into each plate's phone
  bezel → `scene-NN-composite.png`; the on-screen UI + faces are the real screenshot, never
  AI-rendered, and the plate's screen was a blank warm glow.
- **Ken-Burns fallback** produces a clean FFmpeg `zoompan` push-in for any beat Kling garbles
  (heavier zoom on the climax); no garbled Kling beat ships.
- **Concat** joins the beats (order + timeline locked from measured VO durations) + the PIL
  end card into one normalized-fps master.
- **Audio mix** sidechain-ducks the music under VO and loudnorm's the master to -14 LUFS.
- **End card** is a PIL composite of the real wordmark (never AI); captions suppressed there.
- **Finalize** applies 1.15x speed (pitch preserved) + anti-AI grain → H.264 (+ AAC).

The FREE assembly (PIL + FFmpeg) makes NO paid or AI calls. The paid steps — nano-banana
plates, Kling 3.0 i2v beats, ElevenLabs VO + music — are separate capabilities the recipe
orchestrates and gates.