# Pure-AI cinematic controlled-motion polish

Use this reference when the user asks for a filmic TrueTraceShorts / AutoShorts style test using existing AI keyframes only, especially after feedback that a prior version felt too shaky, mockup-like, or effect-heavy.

## Trigger signals

- User says the video should feel like a film/editor/producer cut, not a UI design or PowerPoint explanation.
- User asks for `FINAL VISUALS = only high-quality AI images`.
- User explicitly forbids renderer-built screens, mockup finals, UI boards, arrows, circles, boxes, labels, or instruction cards.
- User asks for an editing/polish pass using existing images only.
- User criticizes artificial shake, vibration, rotation wobble, alarm flashes, or cheap panic effects.

## Hard visual policy

For this mode:

```text
final visuals = existing/premium AI keyframes only
renderer = motion + color/light + music/SFX + voice + subtitles only
```

Do not generate or render:

- new KI images unless explicitly requested
- new scenes
- stock video
- Wan/I2V/T2V
- renderer-built fake UI/screens
- phone/laptop mockup boards
- PowerPoint cards or diagrams
- arrows, circles, boxes, labels, callouts, connector lines
- instruction cards or explainer panels

Only subtitles may be added as renderer text.

## Controlled motion rules

Preferred:

- controlled push-in / Z-axis zoom
- focus-guided X/Y crop toward the important part of the AI image
- ease-out into the scam screen or message
- short holds on readable information
- sparse crossfades
- subtle vignette in danger moments
- mild warm/red alarm grade in the first third
- calmer neutral/warm grade for safe-move scenes

Forbidden unless the user explicitly reverses this preference:

- full-frame camera shake
- artificial screen/frame vibration
- rotation wobble
- frantic zoom jumps
- repeated alarm flashes
- aggressive red filters
- effect-heavy panic edits

Rule of thumb: the camera should guide the eye, not shake the viewer.

## Hook pattern: phone-message scam

When the first frame contains a phone message, make the message dominate before captions compete with it:

1. `0.0–0.4s`: calm wider crop; person + phone visible; no shake.
2. `0.4–1.5s`: smooth ease-out Z-push toward the phone/message.
3. `1.5–2.4s`: tighter crop and short hold on the message; subtle vignette/warm alarm grade.
4. `2.4–3.0s`: controlled hold, only minimal film drift.

Delay or shorten hook captions if they compete with the phone text. The viewer should read/understand the scam screen first.

## Scam pressure scenes

For visible texts such as `Please don't call`, `Can you send money now?`, or `I can't talk. Just send it`:

- crop toward phone/face/finger, not the full room
- use a controlled punch-in at the red-flag moment
- add a short hold so the viewer understands the risk
- keep the tension from the almost-tapped / almost-paid moment, not from chaotic movement

## Safe move scenes

When showing the solution:

- slow the motion down
- remove alarm red
- use warmer/neutral color
- hold longer on the verification action or takeaway note
- make the viewer feel: `Okay, now I know what to do.`

## Voice and audio polish

For a cinematic polish pass, split voice when useful:

- Hook clip: separately rendered Chatterbox/Gianna, slightly more concerned/emotional, short pause after the opening phrase.
- Main narration: calmer, clearer, trustworthy delivery.

If testing hook voice, limit to at most two variants, choose the more natural one, and do not turn the whole video into panic delivery.

Audio should be restrained:

- low tension bed under voice
- one notification ping
- optional single subtle whoosh for the push-in
- one low hit at the red-flag reveal
- one resolve ping near the safe move/end

Avoid repeated vibrations, aggressive alarm sounds, whoosh swarms, horror/trailer music, or anything that makes the piece feel cheap.

## QA checklist

Before delivery, verify:

- exact intended existing keyframes were reused
- no new KI images / scenes were generated
- no renderer-built UI or mockup screens exist
- no arrows/circles/boxes/labels/cards appear
- no shake, frame vibration, or rotation wobble was used
- first phone/scam text dominates the first 1–2 seconds
- captions are subtitle-only, boxless, max 2 lines, safe-zone compliant
- audio true peak is safely below -1 dB, preferably around -3 to -5 dB for previews
- frame sheet has no black/empty end tile from sampling exactly at duration end

## Common pitfall

A clip can be emotionally clear but still feel cheap if the editor adds full-frame shake, repeated vibration, alarm flashes, or too many whooshes. Prefer fewer, precise editorial moves: proximity, legibility, hold, restrained sound, and calm resolution.