Software Demo Video

Create software product videos that look premium and actually convert. This skill covers the full pipeline: strategy, scripting, production, and distribution.

Quick Start

Choose video type - Match to funnel stage and goal (see Video Types below)
Write the script - Pick a framework, write to word count, read aloud (see references/script-templates.md)
Plan production - Storyboard, select music, prepare environment (see references/production-techniques.md)
Produce - Record, film, animate, edit, mix
Optimize for distribution - Thumbnail, title, SEO, embedding strategy (see references/thumbnails-and-distribution.md)

Core Principles

These separate great software videos from forgettable ones. Derived from studying Sandwich Video, Apple, Stripe, Linear, Arc, Figma, Descript, and the top SaaS video agencies.

1. Start with the viewer's problem, not your features

Every top-performing software video opens with pain the viewer already feels. The product appears as the resolution, not the subject. Slack opens with communication chaos. Salesforce opens with slumping sales. Monday.com opens with "What does your ideal work look like?" Feature lists don't create desire; felt problems do.

2. One big idea per video

If the viewer forgets everything else, what must they remember? That's your video. Not three features, not five benefits. One idea.

WRONG: "Our platform has project management, time tracking,
        invoicing, team chat, and file storage."

RIGHT: "Stop switching between apps. Everything your team needs
        is in one place."

3. Show the magic

Sandwich Video's philosophy: identify the moments that make people go "oh, that's cool" and build the video around those moments. Find the 2-3 interactions in your product that create genuine delight and give them the most screen time.

4. Hook in 5 seconds or lose them

55% of viewers leave within 60 seconds. Your first 5 seconds must earn the next 55. Never open with a logo animation, "welcome back," or context-setting. Open with the problem, the result, or a pattern interrupt. Cold opens outperform brand intros in every retention study.

5. Sound design is the most underrated technique

Figma's click sounds, Apple's foley, Arc's deliberate silence -- audio transforms how viewers perceive visual content. Add subtle sounds to UI interactions. Sync cuts to music beats. A video with good sound design feels 2x more polished than one without.

6. Match video style to product personality

Linear's minimalism works because the product is minimal. Arc's bedroom vlogs work because the product is personal. Descript's comedy sketches work because the audience is creative. Style mismatch is worse than low production value.

7. Screen recordings don't have to look like screen recordings

Every top company has a strategy: Apple re-creates UI in motion graphics. Figma adds click sounds and cursor labels. Slack builds custom idealized screenshots. Linear's product is beautiful enough to capture raw. Stripe treats code as art. Find your approach.

Video Types

Type	Length	Funnel	Goal	Audience
Explainer	60-90s	Top	"What is it?"	Problem-aware prospects
Product Tour	2-4 min	Middle	"How does it work?"	Evaluating prospects
Feature Highlight	30-90s	Middle	"What can this do?"	Users + evaluators
Customer Story	60-120s	Bottom	"Who else uses it?"	Decision-stage buyers
Launch Video	30-90s	All	"What's new?"	Users + market
Comparison	2-4 min	Bottom	"Why this over X?"	Active evaluators
Quick Tip	30-90s	Post-sale	"How do I do this?"	Current users
Brand/Culture	60-180s	Top	"Who are these people?"	Talent + curious

See references/script-templates.md for detailed script structures and fill-in-the-blank templates for each type.

Script Frameworks

Pick the framework that matches your video's job. Each has a detailed template in references/script-templates.md.

Problem-Agitate-Solution (PAS) -- Most common

Phase	Time	Words	Purpose
Problem	15-20s	30-40	Name the pain
Agitate	15-20s	30-40	Show consequences of inaction
Solution	25-30s	50-60	Introduce product as resolution
CTA	5-10s	10-20	Clear next step

Videos using problem-solving structure show 60% higher engagement.

Before-After-Bridge (BAB) -- Best for visual products

Show transformation: the viewer's world today, their world with your product, how the product bridges the gap. Best for dashboards, design tools, analytics.

Tell-Show-Tell -- Best for tours and tutorials

Set context (tell), demonstrate live (show), reinforce value (tell). Increases retention by up to 65%. Break complex products into multiple Tell-Show-Tell modules.

Hero's Journey -- Best for customer stories

Your customer is the hero. Your product is the guide. Three acts: ordinary world (the problem), the journey (discovering and using the product), the return (transformed outcomes with metrics).

Other frameworks

JTBD Narrative: "When [situation], I want to [job], so that [outcome]"
Hook-Story-Offer: 5s hook, 30s narrative, 10s CTA (great for short-form)
AIDA: Attention, Interest, Desire, Action
Golden Circle: Why, How, What (Sinek)
Pixar Story Spine: "Once upon a time... Every day... One day... Because of that..."

Opening Hooks

You have 5 seconds. Pick one:

Hook Type	Example	Best For
Problem statement	"If you've spent 3 hours debugging deployment configs..."	Tutorials, pain-aware audiences
Result reveal	Show the finished output first, then explain how	Product tours, feature highlights
Pattern interrupt	Rapid zoom, sudden visual shift, freeze-frame + text	Social, crowded feeds
Cold open tease	Start at the climax: "...and that's when the build broke"	Narrative videos
Question	"What if your deploys never failed?"	B2B, specific segments
Shocking stat	"90% of support tickets are already answered in your docs"	Educational, data-driven audiences
Direct value	"Here are 5 CSS tricks you're not using"	Bottom-funnel, action-oriented

Pattern interrupts in the first 5 seconds show 23% higher retention.

Cold opens beat brand intros. Cut logo animations. Hook first, brand second.

Narration

Pacing

Video Type	WPM	Notes
Explainer	130-150	Sweet spot for comprehension
Tutorial	120-130	Slower for complex info
Promo/launch	150-160	Faster for energy

Word count reference: 30s = ~75 words, 60s = ~150 words, 90s = ~225 words, 2min = ~300 words.

Point of View

Use second person ("you") for narration -- it draws viewers in and makes them the protagonist. Switch to first person for CTAs ("Start my free trial" converts 90% better than "Start your free trial").

Tone

Conversational: Short sentences, contractions, direct. Best for explainers, startups.
Professional: Measured, authoritative. Best for enterprise, security, finance.
Enthusiastic: Energetic, forward-looking. Best for launches.
Empathetic: Warm, understanding. Best for problem-focused intros.

Write as you speak. Read the script aloud. If you stumble, rewrite.

AI vs. Human Voice

AI voice is 60-86% cheaper and instantly updatable. But: AI narration shows 70% lower retention and 35% greater drop-off within 45 seconds. Use human voice for high-stakes content. AI is acceptable for high-volume internal/tutorial content.

ElevenLabs with-Timestamps API

Use the /v1/text-to-speech/{voice_id}/with-timestamps endpoint for word-level sync data. It returns identical audio plus character-level alignment, which you aggregate to word-level and sentence-level timing. This enables automated composition timing -- section durations are derived from measured audio, not guessed from word counts.

Per-Section Audio

Always generate one audio file per script section rather than a single concatenated narration. Concatenated files insert variable inter-section pauses (2-4 seconds each) that accumulate into 20-35 seconds of drift over 10 sections. Per-section audio, each pinned to its composition frame, eliminates cumulative drift entirely. Keep a concatenated file for preview/listening only.

Voice Clones

If a cloned voice is available (e.g., via ElevenLabs voice cloning), use it for authentic founder/developer narration at AI speed and cost. Clone quality is high enough for production use. The --voice <voice_id> flag in the narration script selects the clone.

Speed and Pacing Control

ElevenLabs voice settings that affect pacing:

stability: 0.5 -- natural variation without wandering
similarity_boost: 0.75 -- voice consistency across sections
style: 0.3 -- subtle expressiveness without over-acting
use_speaker_boost: true -- clarity enhancement

For faster pacing, increase stability toward 0.7 and reduce style toward 0.1. For warmer, more conversational delivery, decrease stability toward 0.3 and increase style toward 0.5.

Voice Casting

Brand Personality	Voice Style
Innovative/tech	Clear, confident, medium pace
Warm/approachable	Friendly, smile-in-the-voice
Enterprise	Deep, measured, baritone
Playful/creative	Energetic, expressive, varied inflection
Bold/disruptive	Fast-paced, confident, provocative

Production Quick Reference

Full details in references/production-techniques.md.

Screen Recording

Record at 4K/60fps, deliver at 1080p or 4K
Move cursor at 50% natural speed, pause 0.5-1s before clicking
Close all notifications, use clean browser profile, pre-load all screens
Screen Studio ($89) is the current gold standard for polished demos
Post-process: zoom 150-300% for callouts, ease all zooms (cubic/quintic), add subtle shadows

Transitions

Hard cuts for 90%+ of transitions
Zoom transitions for feature-to-feature navigation (0.3-0.8s)
Dissolves only for time passage (0.5-2s)
UI masking for branded polish (sidebar expands to cover frame, reveals new scene)
Never use transitions as decoration. Every transition should serve the narrative.

Music

Video Type	Mood	Tempo
Launch/sizzle	Upbeat, building	120-140 BPM
Walkthrough	Light, unobtrusive	90-110 BPM
Enterprise	Professional, warm	80-100 BPM
Consumer app	Fun, modern	110-130 BPM

Beat-sync your cuts. Major transitions on downbeats, minor transitions on off-beats. This creates the "Apple feel" -- every cut landing on a musical beat.

Music platforms: Epidemic Sound ($10-20/mo, best for volume), Artlist ($10/mo, broad assets), Musicbed (premium, cinematic).

Narration Sync (Remotion Pipeline)

For Remotion-based productions, use the automated sync pipeline:

script.md --> narrate-sync.ts --> sync-manifest.json --> composition S constants --> audit-sync.ts --> render

Key principles:

Per-section audio: One MP3 per script section, each pinned to its exact composition frame. Never use a single concatenated file as the composition audio source.
Section timing formula: ceil(narration_seconds * fps) + 15 frames padding. Always derive from measured audio, never from word count estimates.
Sync manifest: JSON file with word-level and sentence-level timing data (start/end in both seconds and frames). Generated automatically by the narrate-sync script.
Audit before render: Run audit-sync to detect dead frames, narration overruns, and visual gaps. Fix any issues before rendering.

Full details in references/remotion-production-pipeline.md.

Sound Effects

Add subtle clicks, whooshes, and pops to key UI interactions. Keep SFX -20 to -30 dB below voiceover. Not every click needs a sound -- reserve for meaningful moments.

Voiceover Recording

Minimum: Blue Yeti / AT2020 USB+ ($50-130)
Professional: Shure SM7B / RE20 ($300-450)
Record in a quiet, treated room. Smile while recording.
Post-processing chain: noise reduction > EQ > compression > de-essing > limiter > normalize to -16 LUFS

Thumbnail and Distribution

Full details in references/thumbnails-and-distribution.md.

Thumbnail Checklist

1280x720px, under 2MB, 16:9
One clear focal point
3-5 words max, high contrast, readable at small sizes
Does NOT repeat the video title
Creates curiosity or emotional response
Consistent with channel visual brand
2-3 variants ready for A/B testing

Yellow+black thumbnails yield the highest CTR at 8.2%. High-contrast thumbnails improve CTR by 20-40%.

Title Formulas

Optimal: 55-70 characters, 10-11 words. Numbers boost CTR by up to 36%.

Pattern	Example
Transformation	"How [Tool] Cut Our Deploy Time by 80%"
Unexpected comparison	"[Tool A] vs [Tool B] -- The Winner Surprised Me"
Cautionary	"Stop Using [Practice] -- Do This Instead"
Speed/result	"Build a [Thing] in [Time] with [Tool]"
Discovery	"I Found the Best [Category] Tool (It's Not [Expected])"

YouTube vs. Website

YouTube for discovery and audience building (optimize thumbnails/titles)
Self-hosted (Wistia/Mux) for conversion pages (YouTube embeds add 1.3-2.6MB, show competitor ads, and degrade LCP from 1.8s to 4.5s)

A/B test thumbnails. Median CTR uplift from testing: 33%.

What the Best Companies Do

Detailed analysis of 12+ companies in references/company-examples.md.

Company	Signature Technique	Lesson
Apple	Beat-synced editing, macro close-ups, foley sound	Music sync creates emotion that bypasses rational evaluation
Stripe	Code as art, animation hierarchy, CSS-rendered devices	Developer tools can be visually elegant without dumbing down
Linear	Product IS the visual -- minimal post-production needed	If your product is beautiful, the simplest video treatment wins
Figma	Animated multi-cursors, UI click sounds	Make invisible capabilities (collaboration) physically visible
Arc	CEO in bedroom, YouTuber energy, $500 budget	Authenticity beats production value for shareability
Descript	Comedy sketch format, fourth-wall breaks	Match creative ambition to audience expectations
Slack	Problem-first, mockumentary humor, custom screenshots	Start with viewer's pain, not your features
Notion	Scenario-based narratives, named animation system	Never show a blank canvas -- always start with a specific story
Superhuman	Restraint as technique, static over video	Sometimes the most effective strategy is barely making videos
Loom	Product IS the demo (meta-demonstration)	If your product creates the content, use it

Key Statistics

Metric	Value	Source
Landing page conversion boost from video	+86%	EyeView
Viewers convinced to buy after watching video	87%	Wyzowl 2026
Viewers who watched an explainer before buying	98%	Wyzowl 2026
Conversion lift from A/B testing thumbnails	33% median	YouTube data
First-person CTA conversion improvement	+90%	Journal of Interactive Marketing
Mid-roll CTA average conversion rate	~16%	Wistia
AI narration retention drop vs. human	-70%	Industry data
Retention for educational/how-to content	42.1%	YouTube analytics
Pattern interrupt effect on first-5s retention	+23%	YouTube data
Viewers lost in first 60 seconds	55%	YouTube analytics

Production Budget Tiers

Tier	Cost	What You Get	Examples
DIY/authentic	$0-5K	Webcam + screen recording, personality carries it	Arc, Loom
Professional screen demo	$5-20K	Clean capture, motion graphics, licensed music	Raycast, Linear
Polished animated	$20-55K	Custom animation, pro voiceover, sound design	Slack, Monday
Premium production	$55-100K+	Live action, actors, scripted narrative	Salesforce, Descript, Island
Elite	$250K-400K	Full agency, film-quality, custom scores	Apple, Sandwich Video

Sweet spot for growth-stage SaaS: $8K-$25K agency range delivers strategy + script + custom art + voiceover + sound design.

Remotion Production

For programmatic video production using Remotion (React-based video framework), see references/remotion-production-pipeline.md.

Key Patterns

Per-section audio architecture: Generate individual audio files per script section, each pinned to their exact composition frame. This eliminates cumulative drift from inter-section pauses in concatenated audio files.
Sync pipeline: script.md -> narrate-sync.ts -> sync-manifest.json -> composition S constants -> audit-sync.ts -> render. Fully automated from script changes through to timing.
Section timing: ceil(narration_seconds * fps) + 15 frames padding. Never guess. Always derive from measured audio.
Audit tool: Catches dead frames, narration overruns, and visual gaps before rendering. Run before every render.

Effective Components

Component	Purpose
ZoomPan	Subtle Ken Burns on screenshots (scale 1.0 to 1.03, never more)
FeatureCallout	Glassmorphism cards for feature highlights
SplitReveal	Before/after comparisons with animated divider
SceneTransition	Four types: morph, iris, slide-reveal, scale-fade
TypingAnimation	Simulated user input (chat, code, terminal presets)
DifferentiatorsStack	Numbered feature lists with spring animations
BrowserFrame	macOS browser chrome wrapper with traffic light dots and URL bar

Common Bugs

Double zoom: Do not use ZoomPan wrapper AND BrowserFrame animated={true} simultaneously
CTA-outro overlap: Crossfade should be max 30 frames
TeaseCard brightness: Keep at 0.7+, never add more than 0.5px blur to preview images
Concatenated audio drift: Never use narration-full.mp3 as the composition audio source

Agent Review Pipeline

For structured quality review using independent reviewer agents, see references/agent-review-pipeline.md.

Six Review Dimensions

Script Quality -- hook, pacing, visual direction, CTA, tone
Visual Design -- consistency, animation, hierarchy, transitions, component polish
Pacing & Timing -- script-audio alignment, audio-visual sync, flow, silent beats
Technical Quality -- code quality, reusability, architecture, pipeline, performance
Production Pipeline -- documentation, automation, iteration workflow
Thumbnail -- visual impact, readability, click-worthiness, professionalism

Iteration Loop

script review --> apply feedback --> generate narration --> update composition
  --> audit sync --> render --> multi-dimensional review --> iterate until all >= 7

Key Principles

Self-assessment does not count. Use independent reviewer agents -- the builder has context blindness.
Score progression tracking. Each review compares to previous version scores.
Maximum two revision passes per dimension. Diminishing returns. Ship when all dimensions score >= 7.
Gemini for thumbnails. Use the Gemini vision API for structured thumbnail scoring (readability, contrast, click-worthiness).

Anti-Patterns

Avoid these common mistakes:

Opening with logo animation -- viewers click away before content starts
Feature laundry list -- choose 2-3 "magic" moments instead
Raw, unedited screen recordings -- at minimum add zoom, cursor polish, and music
Corporate jargon in narration -- write as one human explaining to another
Saving the product for the end -- show it in the first 15 seconds
Same video for all audiences -- make role-specific variants for enterprise
Ignoring sound design -- silence feels broken, bad music feels cheap
Hardcoding UI screenshots -- design modularly so product updates don't require reshoots
YouTube embed on conversion pages -- competitors' ads will play after your video ends
Skipping the script -- unscripted demos ramble and lose viewers

Workflow Checklist

Pre-Production

Identify video type and funnel stage
Define the one big idea
Write complete script using a framework
Read script aloud and time it
Create storyboard mapping each shot
Select music and map to video structure
Prepare software environment (clean data, disable notifications)
Rehearse screen recording 3-5 times

Production

Record screen at 4K/60fps
Film talking-head with two-point lighting (if applicable)
Capture B-roll (hands, environment, reactions)
Record voiceover in treated room

Post-Production

Distribution

Create 2-3 thumbnail variants
Write title (55-70 chars, includes number or hook)
Write description with timestamps
Upload to YouTube (discovery) + self-host for website (conversion)
A/B test thumbnails after 1,000+ impressions
Track CTR (target 6%+) and retention (target 45%+)

在你喜欢的 AI 中提问

文档