NeverSight/learn-skills.dev

Install the belt CLI skill: npx skills add belt-sh/cli

Explainer Video Guide

Create explainer videos from script to final cut via inference.sh CLI.

Quick Start

Requires inference.sh CLI (belt). Install instructions

belt login

# Generate a scene for an explainer
belt app run google/veo-3-1-fast --input '{
  "prompt": "Clean motion graphics style animation, abstract data flowing between connected nodes, blue and white color scheme, professional corporate aesthetic, smooth transitions"
}'

Script Formulas

Problem-Agitate-Solve (PAS) — 60 seconds

Section	Duration	Content	Word Count
Problem	10s	State the pain point the viewer has	~25 words
Agitate	10s	Show why it's worse than they think	~25 words
Solution	15s	Introduce your product/idea	~35 words
How It Works	20s	Show 3 key steps or features	~50 words
CTA	5s	One clear next action	~12 words

Before-After-Bridge (BAB) — 90 seconds

Section	Duration	Content
Before	15s	Show the current frustrating state
After	15s	Show the ideal outcome
Bridge	40s	Explain how your product gets them there
Social Proof	10s	Quick stat or testimonial
CTA	10s	Clear next step

Feature Spotlight — 30 seconds (social)

Section	Duration	Content
Hook	3s	Surprising fact or question
Feature	15s	Show one feature solving one problem
Result	7s	The outcome/benefit
CTA	5s	Try it / Learn more

Pacing Rules

Content Type	Words Per Minute	Notes
Standard narration	150 wpm	Conversational pace
Complex/technical	120 wpm	Allow processing time
Energetic/social	170 wpm	Faster for short-form
Children's content	100 wpm	Clear and slow

Key rule: 1 scene per key message. Don't pack multiple ideas into one visual.

Scene Duration Guidelines

• Establishing shot: 3-5 seconds
• Feature demonstration: 5-8 seconds
• Text/stat on screen: 3-4 seconds (must be readable)
• Transition: 0.5-1 second
• CTA screen: 3-5 seconds

Visual Production

Scene Types

# Product in context
belt app run google/veo-3-1-fast --input '{
  "prompt": "Clean product demonstration video, hands typing on a laptop showing a dashboard interface, bright modern office, soft natural lighting, professional"
}'

# Abstract concept visualization
belt app run bytedance/seedance-2-0 --input '{
  "prompt": "Abstract motion graphics, colorful data streams connecting floating geometric shapes, smooth fluid animation, dark background with glowing elements, tech aesthetic",
  "generate_audio": true
}'

# Lifestyle/outcome shot
belt app run google/veo-3-1-fast --input '{
  "prompt": "Happy person relaxing on couch with laptop, smiling at screen, bright airy living room, warm afternoon light, satisfied customer feeling, lifestyle commercial style"
}'

# Before/after comparison
belt app run falai/flux-dev-lora --input '{
  "prompt": "Split screen comparison, left side cluttered messy desk with papers and stress, right side clean organized minimalist workspace, dramatic difference, clean design"
}'

Image-to-Video for Scenes

# Generate a still frame first
belt app run falai/flux-dev-lora --input '{
  "prompt": "Professional workspace with glowing holographic interface, futuristic but clean, blue accent lighting"
}'

# Animate it
belt app run falai/wan-2-5-i2v --input '{
  "prompt": "Gentle camera push in, holographic elements subtly floating and rotating, soft ambient light shifts",
  "image": "path/to/workspace-still.png"
}'

Voiceover Production

Script Writing Tips

• Short sentences. Max 15 words per sentence.
• Active voice. "You can track your data" not "Your data can be tracked."
• Conversational tone. Read it aloud — if it sounds stiff, rewrite.
• One idea per sentence. One sentence per visual beat.

Generating Voiceover

# Professional narration with Dia TTS
belt app run falai/dia-tts --input '{
  "prompt": "[S1] Tired of spending hours on reports that nobody reads? There is a better way. Meet DataFlow. It turns your raw data into visual stories... in seconds. Just connect your source, pick a template, and share. Try DataFlow free today."
}'

Pacing Control in TTS

Technique	Effect	Example
Period .	Medium pause	"This changes everything. Here's how."
Ellipsis ...	Long pause (dramatic)	"And the result... was incredible."
Comma ,	Short pause	"Fast, simple, powerful."
Exclamation !	Emphasis/energy	"Start building today!"
Question ?	Rising intonation	"What if there was a better way?"

Music & Audio

Background Music Guidelines

• Volume: 20-30% under narration (duck 6-12dB when voice plays)
• Style: match the brand tone (corporate = ambient electronic, startup = upbeat indie)
• Structure: intro swell (first 3s) -> subtle loop under narration -> swell at CTA
• No vocals: instrumental only under narration

# Generate background music
belt app run <music-gen-app> --input '{
  "prompt": "upbeat corporate background music, modern electronic, 90 BPM, positive and professional, no vocals, suitable for product explainer video"
}'

Assembly Pipeline

Full Production Workflow

# 1. Generate voiceover
belt app run falai/dia-tts --input '{
  "prompt": "[S1] Your script here..."
}'

# 2. Generate scene visuals (in parallel)
belt app run google/veo-3-1-fast --input '{"prompt": "scene 1 description"}' --no-wait
belt app run google/veo-3-1-fast --input '{"prompt": "scene 2 description"}' --no-wait
belt app run google/veo-3-1-fast --input '{"prompt": "scene 3 description"}' --no-wait

# 3. Merge scenes into sequence
belt app run infsh/media-merger --input '{
  "media": ["scene1.mp4", "scene2.mp4", "scene3.mp4"]
}'

# 4. Add voiceover to video
belt app run infsh/video-audio-merger --input '{
  "video": "merged-scenes.mp4",
  "audio": "voiceover.mp3"
}'

# 5. Add captions
belt app run infsh/caption-videos --input '{
  "video": "final-with-audio.mp4",
  "caption_file": "captions.srt"
}'

Video Length by Format

Format	Length	Platform
Social teaser	15-30s	TikTok, Instagram Reels, YouTube Shorts
Product demo	60-90s	Website, landing page
Feature explainer	90-120s	YouTube, email
Tutorial/walkthrough	2-5min	YouTube, help center
Investor pitch video	2-3min	Pitch deck supplement

Transition Types

Transition	When to Use	Effect
Cut	Default between related scenes	Clean, professional
Dissolve/Crossfade	Time passing, mood shift	Soft, contemplative
Wipe	New topic or section	Clear separation
Zoom/Push	Drilling into detail	Focus attention
Match cut	Visual similarity between scenes	Clever, memorable

Common Mistakes

Mistake	Problem	Fix
Script too wordy	Voiceover rushed, viewer overwhelmed	Cut to 150 wpm max
No hook in first 3s	Viewers leave immediately	Start with the problem or surprising stat
Visuals lag narration	Confusing disconnect	Visuals should match or slightly precede words
Background music too loud	Can't hear narration	Duck music 6-12dB under voice
No captions	85% of social video watched silent	Always add captions
Too many ideas	Viewer retains nothing	One core message per video

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

Browse all apps: belt app store