SKILL.md
Install the belt CLI skill: npx skills add belt-sh/cli
Video Ad Specs
Create platform-specific video ads via inference.sh CLI.
Quick Start
Requires inference.sh CLI (belt). Install instructions
belt loginGenerate a vertical video ad scene
belt app run bytedance/seedance-2-0 --input '{
"prompt": "vertical video, person excitedly unboxing a product, clean modern room, bright natural lighting, social media ad style, authentic feeling",
"ratio": "9:16",
"generate_audio": true
}'
## Platform Specifications
### TikTok
| Spec | Value |
|------|-------|
| Aspect ratio | **9:16** (vertical) |
| Resolution | 1080 x 1920 px |
| Duration | 5-60 seconds (15-30s recommended) |
| File size | Max 500 MB |
| Format | MP4, MOV |
| Sound | On by default (design with sound) |
| Text safe zone | 150px from all edges |
| Hook window | **1 second** — first frame must grab attention |
### Instagram Reels
| Spec | Value |
|------|-------|
| Aspect ratio | **9:16** (vertical) |
| Resolution | 1080 x 1920 px |
| Duration | Up to 90 seconds (15-30s for ads) |
| Cover image | Separate upload, shows in grid |
| Sound | On by default |
| Caption area | Bottom 20% reserved for text overlay |
### Instagram Stories
| Spec | Value |
|------|-------|
| Aspect ratio | **9:16** |
| Resolution | 1080 x 1920 px |
| Duration | Up to 15 seconds per segment |
| Swipe-up/Link | Available for ads |
| Top/bottom | 14% top and 20% bottom = unsafe for key content |
### YouTube
| Format | Aspect | Duration | Skip |
|--------|--------|----------|------|
| Bumper | 16:9 | **6 seconds** exactly | Non-skippable |
| Non-skippable | 16:9 | **15 seconds** | Non-skippable |
| Skippable (TrueView) | 16:9 | Any length | Skip after **5 seconds** |
| Shorts | 9:16 | Up to 60 seconds | N/A |
Resolution: 1920 x 1080 (16:9) or 1080 x 1920 (Shorts)
### Facebook Feed
| Spec | Value |
|------|-------|
| Aspect ratio | **1:1** (square) or **4:5** (recommended for mobile) |
| Resolution | 1080 x 1080 or 1080 x 1350 |
| Duration | Up to 240 min (15-30s recommended) |
| Autoplay | **Silent** — captions are essential |
| Sound | 85% of Facebook video is watched **without sound** |
### LinkedIn
| Spec | Value |
|------|-------|
| Aspect ratio | **1:1** or **16:9** |
| Resolution | 1080 x 1080 or 1920 x 1080 |
| Duration | 3 seconds to 10 minutes (15-30s for ads) |
| Tone | Professional |
| Autoplay | Silent in feed |
## AIDA Framework for Video Ads
| Phase | Time | Goal | Technique |
|-------|------|------|-----------|
| **Attention** | 0-3s | Stop the scroll | Pattern interrupt, bold visual, question |
| **Interest** | 3-10s | Keep watching | State the problem, show relevance |
| **Desire** | 10-20s | Want the solution | Show the product/outcome, social proof |
| **Action** | Final 3-5s | Click/buy/sign up | Clear CTA, urgency, offer |
### Hook Techniques (First 3 Seconds)
| Technique | Example |
|-----------|---------|
| Bold statement | "This tool replaced my entire marketing team" |
| Question | "Why are you still doing this manually?" |
| Surprising visual | Unexpected transformation, before/after reveal |
| Pattern interrupt | Start mid-action, unusual angle, bright color |
| Social proof | "2 million people switched to this" |
| Pain point | "If you hate [common frustration], watch this" |
## Creating Video Ads
### Vertical (TikTok, Reels, Stories, Shorts)
Hook scene (0-3s)
belt app run google/veo-3-1-fast --input '{
"prompt": "vertical 9:16 video, close-up of hands struggling with tangled cables and messy desk, frustrated energy, shaky handheld camera, authentic social media style, bright lighting"
}'
Solution reveal (3-15s)
belt app run bytedance/seedance-2-0 --input '{
"prompt": "vertical video, smooth product reveal, clean wireless charging station on minimalist desk, satisfying organization transformation, bright modern room, social media ad aesthetic",
"ratio": "9:16",
"generate_audio": true
}'
Add voiceover
belt app run falai/dia-tts --input '{
"prompt": "[S1] Stop wasting time with this mess. This one product changed my entire setup. Everything charges. Everything is organized. Link in bio."
}'
Merge video + audio
belt app run infsh/video-audio-merger --input '{
"video": "solution-reveal.mp4",
"audio": "voiceover.mp3"
}'
Add captions (critical for silent autoplay)
belt app run infsh/caption-videos --input '{
"video": "ad-with-audio.mp4",
"caption_file": "captions.srt"
}'
### Square (Facebook, LinkedIn Feed)
belt app run google/veo-3-1-fast --input '{
"prompt": "square 1:1 video, professional person at desk discovering a new software tool, laptop screen showing clean dashboard, natural office lighting, corporate commercial style, satisfied expression"
}'
### YouTube Bumper (6 Seconds)
6-second bumper: one message, one visual, one CTA
belt app run google/veo-3-1-fast --input '{
"prompt": "6 second product ad, quick montage of a sleek app being used on phone, fast cuts, modern, energetic, brand logo reveal at end, punchy and dynamic, wide 16:9"
}'
Keep it tight
belt app run falai/dia-tts --input '{
"prompt": "[S1] Your reports. Automated. Try DataFlow free."
}'
## Captions Are Mandatory
85% of Facebook and 40%+ of Instagram video is watched on mute.
### Caption Best Practices
| Rule | Reason |
| --- | --- |
| Always add captions | Silent viewing is the default on most platforms |
| Large, readable font | Small text is invisible on mobile |
| High contrast | White text with dark outline/background |
| Centered or bottom-third | Standard viewing position |
| Max 2 lines at a time | More text = can't be read fast enough |
| Key words in bold/color | Draws eye to important words |
Generate captions from audio
(create SRT file from your script, then burn in)
belt app run infsh/caption-videos --input '{
"video": "ad-video.mp4",
"caption_file": "ad-captions.srt"
}'
## Ad Structure Templates
### Testimonial Ad (15-30s)
| Time | Content |
| --- | --- |
| 0-3s | Customer states the problem they had |
| 3-15s | How they discovered and tried the product |
| 15-25s | The specific result they achieved |
| 25-30s | Product name + CTA |
### Demo Ad (15-30s)
| Time | Content |
| --- | --- |
| 0-3s | The problem (text or visual) |
| 3-20s | Product demo showing the solution |
| 20-25s | Key result/benefit |
| 25-30s | CTA + offer |
### Before/After Ad (15s)
| Time | Content |
| --- | --- |
| 0-3s | "Before" state (messy, slow, frustrating) |
| 3-5s | Transition / product introduction |
| 5-12s | "After" state (clean, fast, satisfying) |
| 12-15s | CTA |
## Common Mistakes
| Mistake | Problem | Fix |
| --- | --- | --- |
| No hook in first 1-3s | Viewer scrolls past | Open with pattern interrupt |
| Landscape video on TikTok/Reels | Letterboxed, looks amateur | Use 9:16 for vertical platforms |
| No captions | Most viewers watch silent | Always add captions |
| CTA too late | Viewers already left | Clear CTA within last 5 seconds |
| Too long for platform | Forced skip or dropout | Match platform duration norms |
| Same ad for all platforms | Wrong specs, wrong tone | Create platform-specific versions |
| Logo in first 3s | Feels like a commercial, gets skipped | Save branding for the end |
| Text in unsafe zones | Cut off by platform UI | Check safe zone per platform |
## Checklist
- Correct aspect ratio for target platform
- Hook in first 1-3 seconds
- Captions added (readable, high contrast)
- CTA clear and within final 5 seconds
- Duration matches platform norms
- Text outside platform unsafe zones
- Audio designed for both sound-on and sound-off
- Platform-specific version (not one-size-fits-all)
## Related Skills
npx skills add inference-sh/skills@ai-video-generation
npx skills add inference-sh/skills@video-prompting-guide
npx skills add inference-sh/skills@text-to-speech
npx skills add inference-sh/skills@prompt-engineering