# AutoShortsBot Style Bible

## Positioning

**Series:** Unsichtbare Systeme in 30 Sekunden
**Niche:** Hidden Money & AI Systems
**Claim:** Wir zeigen Dinge, die dich Geld, Zeit oder Kontrolle kosten – bevor du sie bemerkst.

The channel is not an AI-avatar news channel and not a static slideshow channel. It is a semi-automatic Retention Studio for short, motion-led mechanism reveals.

## Core visual idea

```text
photorealistic macro investigation + clean mechanism overlays
```

The AI-video layer gives the real-world texture: car, phone, checkout, desk, garage, document, screen glow, hand movement, machine detail. The deterministic renderer adds truth: captions, numbers, arrows, labels, risk meters, x-ray layers and system maps.

## Palette

- Base: graphite black / dark navy / charcoal UI
- Primary text: clean white
- System accent: electric cyan for flows, data and invisible connections
- Warning accent: warm amber for hidden cost/risk
- Danger accent: red only for twist/reveal moments
- Backgrounds: realistic, slightly desaturated, high contrast, not cyberpunk neon soup

## Typography & captions

- Large kinetic captions, no paragraph cards
- 3–7 words per caption beat when possible
- Numbers and labels are renderer overlays, never AI-generated inside video clips
- Captions should puncture the pattern, not describe the obvious
- Text is added after AI-video generation to avoid hallucinated letters

## AI-video rules

Always request:

```text
no readable text, no logo, no watermark, no subtitles, no fake UI text
```

Prefer Image-to-Video for production clips:

```text
keyframe -> Wan2.2 I2V -> selected take -> renderer overlays
```

Use Text-to-Video only for rough exploration or abstract atmosphere.

## Shot families

### 1. Object Hook

Photorealistic object with immediate tension.

Examples:

- used compact car in bright inspection garage
- phone over endless feed before thumb taps
- checkout terminal close-up without readable brand text
- invoice/document stack with no readable details

Camera: slow push-in, low-angle macro, light sweep, rack focus.

### 2. Human Decision Moment

Realistic human context without avatar-hosting.

Examples:

- hand pauses before signing
- buyer looking at car keys and repair estimate
- person hovering over an app confirmation
- creator looking from AI chat to notebook

Camera: handheld micro-movement, over-shoulder, close-up hands, no uncanny faces.

### 3. Mechanism Reveal

Renderer-generated layer animation over or after real-world footage.

Examples:

- x-ray car layers: price, repairs, standdays, dealer margin, liquidity
- attention funnel: hook, variable reward, loop, habit
- AI agent control stack: task, memory, tool, approval gate, audit

Camera/motion: sequential reveals every 1–2 seconds, parallax, snap zooms, pulse on risk nodes.

### 4. Twist Frame

Wrong explanation collapses; real mechanism appears.

Examples:

- `Kaufpreis` fades out; `Zeitpunkt + Risiko` lights up
- `Du bist faul` collapses; `Design gewinnt vorher` appears
- `KI denkt für dich` collapses; `du gibst Ziele + Grenzen` appears

### 5. Follow Trigger

Series promise, not generic CTA.

Good:

```text
Morgen: der Kredit, der kostenlos aussieht.
```

Bad:

```text
Follow for more.
```

## No-go list

- static AI image slideshow as final video
- generic AI avatar as main format
- fake readable UI text
- fake charts/numbers generated by the model
- visible logos/brands unless explicitly approved
- dark underexposed clips where subject is unclear
- unreadable license plates or hallucinated text
- generic “cinematic masterpiece” prompts without specific motion
- reused-content clone formats

## Wan2.2 prompt style

Use concrete, compact prompts:

```text
[shot type] of [one subject] [one action] in [specific environment]. Camera [one motion]. [lighting/exposure]. [style/texture]. Preserve [identity/detail]. No text, no logo, no watermark.
```

Production clips should have:

- one clear subject
- one primary motion
- one camera move
- balanced exposure
- exact environment
- explicit identity preservation for I2V
- negative prompt for text/logos/deformation/flicker

## Pilot style: cheap car hidden cost

### Keyframe direction

```text
Photorealistic vertical image of a red used compact car in a bright European vehicle inspection garage, full front three-quarter view, clean empty license plate area, visible wheels and body panels, soft overhead lights, realistic reflections, shallow depth of field, balanced exposure, no text, no logos, no watermark.
```

### I2V direction

```text
Photorealistic vertical video of the same red used compact car in a bright inspection garage. Camera slowly pushes in from a front three-quarter angle. Subtle light sweep over hood and windshield, stable composition, realistic reflections, balanced exposure. Preserve the same car and garage. No text, no logo, no watermark.
```

### Renderer overlay direction

- Price tag peels away
- Repair-risk layer lights amber
- Standdays/time layer pulses cyan
- Dealer margin layer slides in
- Liquidity/risk node glows red on twist

## Quality gate

A clip is usable only if:

- subject is identifiable in first 0.5 seconds
- no unreadable generated text is prominent
- motion supports the retention beat
- enough clean space remains for captions/overlays
- no uncanny face/hand/artifact distracts from mechanism
- exposure is balanced enough for phone viewing

If the clip fails, regenerate the keyframe or change the shot. Do not polish bad source material.