# Clean image-cut / no-overlay cinematic pilot

Use this reference when the user asks for a filmic TrueTraceShorts style test or explicitly rejects PowerPoint/mockup/UI-board visuals.

## Core rule

For this class of test, final visuals are **only premium AI images**. The renderer may add:

- motion from stills: push-in, push-out, crop move, controlled Z-axis zoom, punch-in, short hold, sparse crossfade
- color/light treatment: subtle warm/red alarm shift for danger, calmer neutral grade for resolution
- music, SFX, voice, and subtitles

For TrueTraceShorts pure-AI scam shorts after the Director Pass standard, **do not** use micro-shake, full-frame vibration, or rotation wobble unless the user explicitly requests that style. Use controlled Z-axis push/crop, holds, subtle vignette/color, and restrained audio instead. See `references/director-pass-cinematic-production-standard.md` and `references/pure-ai-cinematic-controlled-motion.md`.

The renderer must **not** create the scam screen itself. The threat must be visible inside the generated AI image.

## Hard no-go list

Do not build or render:

- fake UI screens in code/canvas/PIL/SVG/HTML
- phone/laptop mockup boards
- PowerPoint-style cards
- explanatory boxes or instruction cards
- arrows, circles, labels, callouts, diagrams
- abstract cybersecurity clouds/orbs/glass shapes
- stock video, Wan, I2V, or multiple versions in parallel for a single requested pilot

Only subtitles may be rendered as text after the image is generated.

## Good visual pattern

Choose an everyday scam with a strong emotional first frame:

- parent sees a suspicious “new number” text directly on the phone in the AI image
- person sees a fake support popup directly on the laptop in the AI image
- delivery-fee scare visible on the phone in the AI image
- payment pressure visible as an everyday panic moment

If text appears in the image, keep it short, plausible, large enough to read, and generic/safe. No real brands, domains, phone numbers, bank/card data, QR codes, or private details.

## Required internal plan before rendering

Create internally before render:

1. selected theme
2. hook concept
3. first 0–3s spoken hook lines
4. planned AI keyframes
5. motion plan
6. audio plan

Then render exactly one preview unless the user explicitly asks for variants.

## Structure for this pilot style

- 0–3s: dramatic emotional hook, e.g. “Oh no… is this really my child?”
- 3–8s: visible problem moment
- 8–18s: red flag / recognition through image sequence + voice
- 18–30s: safer move
- 30–45s: simple takeaway

Aim for 30–45s for cinematic style tests unless the user requests the shorter hook-retention production cycle.

## Voice and audio

- Use Chatterbox/Gianna, not EdgeTTS.
- First 1–2 sentences may be more urgent/panicked; then return to controlled, trustworthy delivery.
- Add subtle sound design: notification ping, vibration, tension hit, whoosh, alert pulse, and restrained suspense bed.
- Keep music below the voice; avoid cheap trailer or meme/TikTok sounds.

## Subtitles

- white text
- active word yellow
- no box, no bar, no caption card
- subtle outline/shadow
- max 2 lines
- safe zone preserved

## QA checklist

Before delivery, verify and report:

- Does the first second show the scam/emotional danger without audio?
- Are all final visuals AI images, not renderer-built screens?
- Are there no arrows/circles/boxes/cards/labels/PowerPoint graphics?
- Are fake texts generic and safe?
- Is the hook emotional enough?
- Does any scene feel too static?
- Which scene is strongest?
- Which scene needs improvement later?

Deliver: candidate ID, theme, hook line, voice, music/SFX summary, number of AI images, preview MP4, contact sheet, frame sheet, and the short self-assessment above.