Video Ad Specs

Video ad creation with exact platform-specific specs for TikTok, Instagram, YouTube, Facebook, LinkedIn. Covers dimensions, duration limits, AIDA framework,...

Security Scan

The skill claims to provide platform-specific video ad specs, but its runtime instructions push installing and using a third‑party CLI (via curl | sh) and logging into a remote inference service without declaring any credentials — an incoherent and elevated network/install footprint compared to the stated purpose.

Purpose & Capability

The name/description promise static platform specs and creative guidance, which the SKILL.md does include, but the runtime instructions go beyond 'specs' and instruct the agent to install and use an external inference CLI to generate, merge, caption and upload media. A pure 'ad specs' helper would not need a remote model‑execution CLI or login, so the requested actions are disproportionate to the stated purpose.

Instruction Scope

SKILL.md explicitly tells the user/agent to run 'curl -fsSL https://cli.inference.sh | sh', 'infsh login', and multiple 'infsh app run' commands that will send prompts and media to remote services. The instructions therefore cause network calls, credential usage, and remote model execution not declared elsewhere; they also assume local file reads/writes for media merging. The skill does not explain what data will be transmitted or what credentials are required.

Install Mechanism

Although the registry lists no install spec, the README directs running a piped install from cli.inference.sh which downloads binaries from dist.inference.sh. This is a moderate-to-high risk pattern because the host is not a well-known vendor in the manifest, the installer is executed via shell pipe, and the binary download/extract steps occur outside the package manager model. The SKILL.md claims checksum verification exists, but fetching and running a remote installer remains risky and should be verified independently.

Credentials

The skill declares no required environment variables or credentials, yet instructs an 'infsh login' and uses various remote models (bytedance/..., google/..., falai/...) that likely require provider credentials or accounts. This mismatch (instructions requiring authentication but no declared primaryEnv/required env vars) is a clear incoherence and increases the chance of unexpected credential prompts or misconfiguration.

Persistence & Privilege

The skill does not request permanent presence (always:false) and has no declared system-level config changes in the manifest. However it directs installing a third-party binary and logging into a remote service, which grants that external tool ongoing network capability from the host; autonomous invocation is allowed by default and combined with the above concerns increases blast radius if the external CLI is trusted without verification.

What to consider before installing

This skill mixes benign documentation (ad specs, AIDA guidance) with instructions to install and use an external CLI (inference.sh) via a piped shell installer and to 'infsh login' — but it doesn't declare any required credentials. Before installing or using: (1) confirm you trust https://cli.inference.sh/dist.inference.sh and independently verify checksums; avoid piping unknown scripts to sh if unsure; (2) expect that prompts/media will be transmitted to remote services — don't include sensitive data in prompts; (3) check what account/credentials the CLI requires and whether you are willing to provide them (the skill manifest doesn't declare them); (4) if you only want static specs/guidance, consider using a safer, instruction-only resource and skip the installer; (5) run the installer in an isolated environment (container/VM) first to validate behavior.

Create platform-specific video ads via inference.sh CLI.

Quick Start

curl -fsSL https://cli.inference.sh | sh && infsh login

# Generate a vertical video ad scene
infsh app run bytedance/seedance-1-5-pro --input '{
  "prompt": "vertical video, person excitedly unboxing a product, clean modern room, bright natural lighting, social media ad style, authentic feeling, 9:16 format"
}'

Install note: The install script only detects your OS/architecture, downloads the matching binary from dist.inference.sh, and verifies its SHA-256 checksum. No elevated permissions or background processes. Manual install & verification available.

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)
infsh 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)
infsh app run bytedance/seedance-1-5-pro --input '{
  "prompt": "vertical video, smooth product reveal, clean wireless charging station on minimalist desk, satisfying organization transformation, bright modern room, social media ad aesthetic"
}'

# Add voiceover
infsh 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
infsh app run infsh/video-audio-merger --input '{
  "video": "solution-reveal.mp4",
  "audio": "voiceover.mp3"
}'

# Add captions (critical for silent autoplay)
infsh app run infsh/caption-videos --input '{
  "video": "ad-with-audio.mp4",
  "caption_file": "captions.srt"
}'

Square (Facebook, LinkedIn Feed)

infsh 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
infsh 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
infsh 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)
infsh 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

Browse all apps: infsh app list