Skip to content

gooseworks-ai/goose-skills / skills/video/render-cgi-sizzle

render-cgi-sizzle

Verified

Assemble a 3D-CGI app sizzle — nano-banana CGI plates (blank-glow floating phone + a CGI studio with rim-light + placeholder burst shapes; studio look and phone finish come from the recipe choices) plus PIL compositing of the REAL App Store screenshots onto the bezel + burst-out overlays, driven by Kling 3.0 i2v steady-float per beat with a per-beat Ken-Burns FFmpeg push-in fallback when Kling garbles the UI, then VO/music mix (sidechain duck, loudnorm) + 1.15x speed + anti-AI grain finalize. The on-screen UI, any faces in it, and the wordmark are ALWAYS real assets composited via PIL — never AI-rendered. The paid steps (plates, i2v clips, VO, music) are separate capabilities; this ships the config + PIPELINE + FREE assembly and the recipe orchestrates the spend. Use for the cgi-app-sizzle video format.

FileKindSize
SKILL.mdskill md5.5 KB
scripts/PIPELINE.mdmarkdown4.1 KB
scripts/README.mdmarkdown4.4 KB
scripts/config.example.jsonconfig13.2 KB
skill.meta.jsonconfig172 B
tests/smoke-test.mdmarkdown1.4 KB

scripts/README.md

View on GitHub
# 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 free
(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, studio, rim-light, bokeh, and
placeholder burst-out shapes** (look + phone from the recipe choices; demo: smoky-black studio,
amaranth rim-light, gold phone) — 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, any 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
  portrait tiles), bake the real portraits as glass-morphic tiles with the
  rim-light baked BEFORE rotation (so frame + portrait rotate as one unit) + rim-light-colour 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: the bed from the music choice (demo: premium-tech) → 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)
Studio-background canvas + accent-colour bar (demo: smoky-black + amaranth) + 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.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall