uttamongithb1/Premium Video Skill — v3.3 (Anchor Grid + Sound)

What does Premium Video Skill — v3.3 (Anchor Grid + Sound) do?

==================================================

A senior-agent orchestration skill that converts a voice script into a premium HyperFrames short-form video with:

• Anchor grid (kicker → hero → body → chip → presenter, bottom-fixed)
• Kinetic typography (per-word staggered animation)
• SVG animations (technical diagrams with moving elements)
• Glassmorphism evidence cards + lower thirds + end screens
• Motion-blur wipe transitions
• Complete sound system (SFX synthesis, cue sheet, audio mixing)
• Design-contract validation (prepare.py gate)
• Reproducible pipeline (scenes.sh → prepare.py → build_index.py)

---

Pipeline (order is not negotiable)

voice script
    │
    ▼
generate.py  →  new_scene.py  →  prepare.py  →  build_index.py
  (parse)      (scene files)   (validate)    (root HTML + wipes)
    │                                         │
    ▼                                         ▼
npx hyperframes lint  →  npx hyperframes render
  (0 errors gate)         (MP4 output)
    │
    ▼
python3 scripts/gen_sfx.py       # synthesize SFX (one-time)
python3 scripts/mix_audio.py     # VO + cues + bed → master.wav
ffmpeg mux → final.mp4

Step-by-step

# 1. Generate scene compositions + root HTML from voice script
python3 generate.py --script my_script.txt --topic Tor --output index.html

# 2. Validation gate (contrast, type, safe zone, field rhythm)
python3 scripts/prepare.py

# 3. Lint (must be 0 errors)
npx hyperframes lint

# 4. Preview and render
npx hyperframes preview
npx hyperframes render

# 5. Audio pipeline
python3 scripts/gen_sfx.py              # synthesize 14 SFX from scratch
python3 scripts/mix_audio.py            # VO + cues + bed → master.wav
ffmpeg -i rendered.mp4 -i renders/master.wav \
  -c:v copy -c:a aac -shortest final.mp4

---

Anchor Grid (bottom-fixed, invariant)

Every scene uses the same bottom half — only the top half re-composes:

Anchor	Element	Position	Animation
1	Kicker	Top left	Slide up from y:-26, fade in 0.32s
2	Hero	Top band (1–3 words)	Scale 1.3→1, drift 44px→0, slow drift up
3	Body	Card / odometer / lines	Slide up 56px→0, scale .95→1
4	Chip	Bottom area (3–6 words)	Slide up 30px→0, fade in 0.26s
5	Presenter	Bottom right, cropped by frame	Scale .92→1, y:46→0, fade in 0.34s

Frame 0 is a poster — scene 1 (data-cover="1") opens fully composed.

---

Six-Color Field Cycle

No two neighbor scenes share a background field:

accent → ground → ink → secondary → tertiary → quaternary

prepare.py fails the build if two scenes in a row share a field.

All text/background pairs meet WCAG ≥ 4.5:1 contrast.

---

Sound System

SFX Synthesis (no downloads, no licensing)

python3 scripts/gen_sfx.py creates 14 sound effects:

SFX	Use
pop.wav	Every scene cut
click.wav / key.wav	UI elements, list ticking
chime.wav	Reveal / score
ping.wav	Notification
notify.wav	CTA
riser.wav	Big turns before cut
impact.wav	Strong visual hit
whoosh.wav	Fast movement
error.wav	Wrong / rejected

Cue Sheet — assets/sfx/cues.json

Section-relative offsets (survive VO re-records):

{ "at": "scenes", "file": "pop.wav", "gain": 0.5 }
{ "at": "vo2-0.30", "file": "riser.wav", "gain": 0.3 }
{ "at": "vo9+0.55", "file": "notify.wav", "gain": 0.4 }

Audio Mixing — scripts/mix_audio.py

• VO + SFX cues + music bed → renders/master.wav
• Measures loudness with ebur128, corrects to −14 LUFS
• True peak ≤ −1 dBTP
• alimiter with level=false (avoids auto-level trap)
• Bed at 0.15 volume → −29 LUFS (~13 LU under −16 LUFS VO)
• No sidechain ducking — the level gap is the duck

---

Technical Topics (auto animation + evidence)

When the script contains a recognized keyword, the orchestrator inserts:

Component	Treatment
SVG diagram	Node-link flow with moving element + animateMotion
Evidence overlay	Glassmorphism card with source badges
Transition	Motion-blur wipe before animation
Audio cue	"whoosh" SFX at animation start

Recognized: tor, onion, vpn, proxy, isp, relay, network, packet, encryption, cybersecurity, dns, cdn, blockchain, ai, machine learning, and more.

---

Input Script Formats

Plain text (auto-timed ~145 wpm)

Same resume everywhere? That's why you're rejected.
Tailor it per company.
Read the **job description**.
One resume does not fit every company. Customize and get shortlisted.

Timestamped segments

[0-4s] Same resume everywhere? That's why you're rejected.
[4-10s] Tailor it per company.
[10-14s] Read the **job description**.

Emphasis: **word** → right-aligned, brand color, weight 950, pop animation.

---

Files

File	Purpose
generate.py	CLI entry point
orchestrator.py	Pipeline: parse → detect → decompose → generate scenes → build root
design_system.py	Design tokens, alignment rules, validation
PARAMS.json	The ONLY design file you edit
scripts/prepare.py	Validation gate + token injection
scripts/new_scene.py	Anchor-grid scene composition generator
scripts/build_index.py	Root HTML with contiguous timing + adaptive wipes
scripts/gen_sfx.py	SFX synthesis from scratch
scripts/mix_audio.py	VO + cues + bed → master.wav
assets/sfx/cues.json	Section-relative cue sheet
scenes.sh	Reproducible scene list template
compositions/_scene-base.html.tmpl	Anchor-grid skeleton
compositions/wipe.html	Motion-blur transition
skill.md · skill.json	Documentation + schema

---

Requirements

• Python 3 (standard library)
• HyperFrames CLI: npx hyperframes init / preview / render / lint
• Node.js ≥ 22, FFmpeg
• ffprobe (for audio measurement)

---

License

MIT — build on it, share it, sell videos with it.