# Detailed AI Screen Proof Shorts — new visual standard

Use this when producing TrueTraceShorts / AutoShorts proof shorts where the mechanism is a concrete digital action, scam, invite, login, calendar, email, payment screen, or workflow failure.

## Session learning

The user rejected abstract/premium mood visuals and also rejected added renderer focus lines/boxes/arrows as not adding value. The accepted direction was a clean AI-image cut where the image model itself carries the visible mechanism through detailed, realistic screen/UI imagery.

Accepted example pattern:

```text
B3 Detailed AI Invite Cut
= detailed AI-generated real UI/screen images
+ readable concrete text inside the AI image
+ active-word subtitles
+ no extra focus lines/arrows/boxes
+ verified 9:16 SAR/DAR export
```

For the calendar invite/login proof short, effective prompts asked for a realistic Outlook-style-but-unbranded meeting invite with concrete readable text such as:

- `Quarterly Review Meeting`
- `Today, 10:00 AM`
- `Join meeting`
- `Sign in to continue`
- `Email` / `Password`
- `Calendar`
- `Open meeting from calendar`
- `Meeting verified`

## Prompting rule

When the image model is reliable enough, do not force all screen/UI text to be renderer-owned. Instead, write very detailed image prompts that specify the exact visible UI text, screen composition, and why the viewer should understand the problem in the first second.

Prompt requirements:

- vertical 9:16 / portrait
- realistic phone/laptop/desktop scene with screen sharp enough to read
- exact UI text lines listed in the prompt
- no brand logos, real company marks, QR/barcodes, watermark, people, hands, faces
- rich UI details: sidebars, cards, buttons, attendee dots, event lists, login panel, calendar grid
- clean lower safe zone for active-word subtitles
- premium cinematic lighting, but the screen mechanism must be clearer than the mood

## Overlay rule

For this accepted style, do **not** add decorative explanation layers unless explicitly requested:

- no focus lines
- no arrows
- no boxes
- no circles
- no connector lines
- no diagram markers

Subtitles are allowed and should use the static phrase + active spoken word highlight pattern.

## QA rule

Before delivery:

1. Create a contact sheet of AI keyframes and rendered video frames.
2. Inspect for readable intended text and reject obvious gibberish, logos, QR codes, fake brand marks, people/hands/faces, or cluttered subtitle zones.
3. Verify the MP4 with `ffprobe` for `1080x1920`, `SAR=1:1`, `DAR=9:16`.
4. Extract a real frame from the final MP4; do not judge aspect ratio from a square contact sheet.
5. Deliver the corrected final media, not older B1/B2 or unrelated car clips.

## Pitfall

If the user references “the new standard” or “the high-quality detailed AI images,” they mean the B3-style detailed screen-image proof short, not earlier clean blank-screen AI keyframes, focus-line variants, or unrelated generated car/Wan clips.