AGENT SKILL

video-spec-sheet

Platform video output specs: aspect ratio, length, codec, bitrate, safe zones, ffmpeg presets for TikTok, Reels, Shorts, YouTube. Use to check an export.

使用场景图像与多媒体
License 许可
Apache-2.0
所属仓库
Muvon/octomind-tap
最近更新
2026-11-13
Stars
★ 6
Forks
0

01 前言 Front Matter

name
video-spec-sheet
rules
[ "session(video)", "session(adcraft)", "content(reels)", "content(shorts)", "content(tiktok)", "content(instagram stories)", "content(aspect ratio)", "match(\\bsafe\\s+zone\\b)", "match(\\b(video\\s+)?codec\\b)", "match(\\b(video\\s+)?bitrate\\b)", "match(\\bplatform\\s+specs?\\b)", "match(\\b(9:16|16:9|1:1|4:5)\\b)", "semantic(what dimensions should I export this video at)", "semantic(safe export settings for TikTok or Reels)", "semantic(what are the platform specs for short-form video)", "semantic(which codec and bitrate should I use for vertical video)", "semantic(why does my video get cropped on Instagram)" ]
title
Video Spec Sheet: Platform Specs and Safe Zones
domains
video
license
Apache-2.0
description
Platform video output specs: aspect ratio, length, codec, bitrate, safe zones, ffmpeg presets for TikTok, Reels, Shorts, YouTube. Use to check an export.
compatibility
Stack-agnostic.

03 SKILL.md 正文

Overview

Each social platform has its own combination of aspect ratio, max duration, codec, bitrate, frame rate, and caption / UI safe zones. Getting any of these wrong on the export means the platform will either reject, re-encode (quality drop), or crop your creative. This skill encodes the 2026 specs as a single table the agent extracts from per brief. Caps and safe zones move several times a year: treat every number marked 2026 as a starting point and confirm it against the platform's live spec page before a paid launch.

Instructions

Core Rules

  • Always pick the platform spec before generating frames — generation must match the export aspect ratio (don't generate 16:9 then crop to 9:16 unless intentional).
  • Burn captions on top of the base render, not as a sidecar. Don't assume Reels are watched sound-off: Meta reports over 75% of Reels views are sound-on (official, Meta for Developers, 2024-11-07). Captions carry the message for the muted viewer and accessibility while the audio carries it for everyone else; in TikTok's Kantar survey 88% of TikTok users said sound is essential (self-reported; official, TikTok for Business blog, 2021-06). Render them from word timings through libass with the video-captions skill; stock Homebrew ffmpeg has no libass.
  • Respect the safe zone — keep all critical text and faces away from where the platform's UI sits (top, bottom, right side for TikTok).
  • Final encode in H.264 / yuv420p, even if the source is HEVC / 10-bit. Universal compatibility beats marginal quality.
  • Audio: AAC-LC stereo, 48 kHz, 128 kbps minimum (YouTube recommends 384 kbps stereo for uploads). Mono audio reads as low-quality on most feeds.
  • Loudness: master to about -14 LUFS integrated, true peak ≤ -1 dBTP. YouTube normalizes to roughly that level; TikTok and Meta don't publish a target, so -14 is a safe common master.
  • AI-generated clips usually come out at 720p (Runway Gen-4 Turbo renders 1280×720 / 720×1280). Upscale picked shots before the final 1080p encode; never let the platform upscale.
  • Re-encoding with ffmpeg drops any C2PA Content Credentials the generator embedded, so platform auto-labelling of AI content will not fire. Disclose manually at upload (see Disclosure and Rights).

Platform Spec Table (2026 — verify live)

Platform Aspect Resolution Length cap Frame rate Codec Bitrate Audio
TikTok organic 9:16 1080×1920 10 min recorded in-app, 60 min uploaded 30 / 60 H.264 8–10 Mbps AAC 128k+ 48kHz
TikTok in-feed ad (non-Spark) 9:16 (16:9, 1:1 accepted) 1080×1920 (min 540×960) 10 min, 500 MB 30 H.264 ≥516 kbps (aim 8 Mbps) AAC 128k+ 48kHz
Instagram Reels 9:16 1080×1920 (Meta ads guide recommends 1440×2560) Organic up to 20 min since late 2025, but Reels over 3 min aren't recommended to non-followers (reported, Instagram Help via Socialinsider, 2026-07); Reels ads up to 15 min, 4 GB 30 H.264 5–8 Mbps AAC 128k+ 48kHz
YouTube Shorts 9:16 or 1:1 1080×1920 3 min; any square/vertical upload ≤3 min is classified as a Short automatically 30 / 60 H.264 8–12 Mbps AAC 384k 48kHz
Facebook video 9:16 or 4:5 1080×1920 / 1080×1350 All Facebook video uploads publish as Reels since 2025; feed ads 1s–241 min, 4 GB 30 H.264 5–8 Mbps AAC 128k+ 48kHz
IG / FB Stories 9:16 1080×1920 60s per card 30 H.264 5–8 Mbps AAC 128k+ 48kHz
Snapchat 9:16 1080×1920 60s (verify Spotlight cap) 30 H.264 5–8 Mbps AAC 128k+ 48kHz
YouTube long-form 16:9 1920×1080 / 3840×2160 12 hr or 256 GB 30 / 60 H.264 (or H.265 source) 8 / 12 Mbps (1080p30/60), 35–45 Mbps (4K30) AAC 384k 48kHz
X 16:9 or 1:1 1280×720+ 2 min 20 s standard; longer on Premium 30 / 60 H.264 5–25 Mbps AAC 128k 48kHz
LinkedIn feed 1:1, 4:5, 16:9, 9:16 1080×1080 / 1920×1080 / 1080×1920 15 min desktop upload, 10 min mobile 30 H.264 8–10 Mbps AAC 128k 48kHz
Pinterest video Pin 9:16 or 2:3 1080×1920 15 min 30 H.264 5–8 Mbps AAC 128k 48kHz

Safe Zones (keep critical text, faces and logos out of these UI bands)

Platform Top band Bottom band Side bands Source
Meta Reels ads (IG + FB) 14% (~270 px) 35% (~672 px) 6% each (~65 px) Meta Ads Guide
Meta Stories 14% (~270 px) 20% (~380 px) 6% each Meta Ads Guide
TikTok ~130–160 px ~350–440 px, grows with caption length right ~140–200 px (action rail) download TikTok's official safe-zone template
YT Shorts ~110 px ~280 px right ~180 px approximation; preview on device
YT long-form 16:9 ~60 px ~100 px (progress bar) 0 px approximation

(Pixels at 1080×1920 or 1920×1080; scale with output resolution. Ads that carry a legal disclaimer on Meta need the bottom 40% clear.) Instagram's profile grid crops covers to 3:4, so keep the cover's subject and title inside the centre 1080×1440.

Captions / On-Screen Text Rules

  • Keep the caption block out of every UI band; never bottom-aligned (bottom is platform UI).
  • 9:16 talking head: block centered at about 58% of height (Y≈1110 of 1920), just below the chin and above the bottom band. Never over the face.
  • 9:16 product or screen demo with the subject mid-frame: upper third (about 30% of height), still below the top band.
  • Font: bold sans (Poppins Black, Montserrat Black, Anton, Inter Black). Caption text roughly 80–110 px at 1080×1920, hook text slightly smaller in a box.
  • Stroke + drop shadow: about 8% of font size stroke, soft shadow — preserves readability over any background.
  • One sentence per card max. Two-sentence cards get skipped.
  • Word-by-word reveal in time with VO is the common preset look; test it against flat full-sentence captions on the account's own data (no measured source). Keep on-screen text at or under 5–10 words per second (TikTok creative guidance).
  • Instagram captions hold at most 2,200 characters (official, Meta developer docs, Instagram Graph API media reference). Instagram posts allow at most 5 hashtags (reported, Business Today, 2025-12-19; newer than the API reference, which may lag). Put the hook and topic words in the first line, before the "more" truncation.

Output Presets (ffmpeg) — copy-paste ready

# 9:16 — TikTok / Reels / Shorts / Stories
-vf "scale=1080:1920:force_original_aspect_ratio=decrease,pad=1080:1920:(ow-iw)/2:(oh-ih)/2:color=black,fps=30" \
  -c:v libx264 -profile:v high -level 4.0 -crf 21 -preset slow -pix_fmt yuv420p \
  -c:a aac -b:a 128k -ar 48000 -ac 2 \
  -movflags +faststart

# 1:1 — IG / FB feed
-vf "scale=1080:1080:force_original_aspect_ratio=decrease,pad=1080:1080:(ow-iw)/2:(oh-ih)/2:color=black,fps=30" \
  -c:v libx264 -profile:v high -crf 21 -preset slow -pix_fmt yuv420p \
  -c:a aac -b:a 128k -ar 48000 -ac 2 -movflags +faststart

# 16:9 — YouTube long-form / LinkedIn
-vf "scale=1920:1080:force_original_aspect_ratio=decrease,pad=1920:1080:(ow-iw)/2:(oh-ih)/2:color=black,fps=30" \
  -c:v libx264 -profile:v high -level 4.2 -crf 19 -preset slow -pix_fmt yuv420p \
  -c:a aac -b:a 192k -ar 48000 -ac 2 -movflags +faststart

Workflow

  1. Identify primary platform for the brief (TikTok / Reels / Shorts / etc.).
  2. Pick aspect, length cap, fps from the table.
  3. Mark safe zones when storyboarding — keep face, hook text, CTA inside the center region.
  4. At export, use the matching ffmpeg preset above.
  5. If publishing to multiple platforms, render the master in the largest aspect (e.g. 9:16 1080×1920 30fps) and downconvert to others — never upscale.

Decision Guide

Brief says Output
"TikTok ad" 9:16, 1080×1920, 30fps, H.264 8 Mbps; length is a creative choice (allowed up to 10 min)
"Instagram Reel" 9:16, 1080×1920, 30fps, ≤3 min when the goal is non-follower discovery (up to 20 min is allowed but mostly reaches followers), H.264 6 Mbps
"YouTube Short" 9:16, 1080×1920, 60fps if motion-heavy else 30, ≤3 min; keep ≤60s if it uses any claimable music
"Story (any platform)" 9:16, 1080×1920, 30fps, ≤60s per card
"IG feed" 4:5, 1080×1350, 30fps
"FB feed" 4:5, 1080×1350 (1:1 also fine), 30fps
"YouTube long-form" / "explainer" 16:9, 1920×1080, 30fps (60fps if screen-cap)
"Hero video for landing page" 16:9, 1920×1080, 30fps, muted, poster frame, MP4 (H.264) + WebM/AV1, short loop under ~5 MB
"X post" 16:9 1280×720 if widescreen, else 1:1 1080×1080
"LinkedIn post" 1:1 1080×1080 or 4:5 1080×1350
"Pinterest" 9:16 1080×1920

Disclosure and Rights (2026 — verify live)

Item Rule
YouTube Turn on "altered or synthetic content" in Studio when realistic AI makes a real person say/do something, alters a real event or place, or shows a realistic scene that never happened. Not needed for clearly unrealistic/animated content or AI used only for scripts, captions, production help.
TikTok organic Use the AI-generated content label (API field is_aigc) for realistic AI content; unlabelled realistic AI content can be removed. Paid partnerships and own-brand promotion need the commercial content toggles.
TikTok ads Ads Manager's AI-generated content disclaimer is mandatory for AI-generated or AI-significantly-altered images, video or audio; undisclosed ads are rejected or restricted.
Meta Meta auto-applies "AI info" when it detects C2PA/IPTC signals or its own gen-AI tools; political and social-issue ads must self-disclose photorealistic AI.
EU AI Act Art. 50(4): deepfake content shown to EU audiences must be disclosed as AI-generated or manipulated from 2 Aug 2026.
New York GBL § 396-b (in force 9 Jun 2026): ads that include a synthetic performer (an AI-generated person not recognizable as a real individual, e.g. an AI avatar presenter) must conspicuously disclose it when the advertiser knows. $1,000 first violation, $5,000 each after. Audio-only ads and AI used solely to translate a human performer are not covered.
Music YouTube Shorts over 1 min with any active Content ID claim are blocked worldwide. Business TikTok accounts and all ads use the Commercial Music Library or owned/licensed music; trending sounds are not cleared for brand use. AI music (e.g. ElevenLabs) is commercially usable only on a paid plan per its terms.

Examples

Example 1: TikTok ad export

Source: 4K 16:9 24fps generated at Veo. Target: TikTok ad.

ffmpeg -i source.mp4 \
  -vf "scale=1080:1920:force_original_aspect_ratio=decrease,pad=1080:1920:(ow-iw)/2:(oh-ih)/2:color=black,fps=30" \
  -c:v libx264 -profile:v high -level 4.0 -crf 21 -preset slow -pix_fmt yuv420p \
  -c:a aac -b:a 128k -ar 48000 -ac 2 \
  -movflags +faststart \
  tiktok-ad-9x16.mp4

Example 2: Cross-platform render plan

Brief: One ad, ship to TikTok + IG Reels + IG feed + YT Shorts + YT long-form trailer.

Cut Aspect Resolution Length
Master 9:16 1080×1920 45s
TikTok 9:16 1080×1920 45s (master)
IG Reels 9:16 1080×1920 45s (master)
YT Shorts 9:16 1080×1920 45s (master, captions repositioned higher)
IG feed (1:1 cut-down) 1:1 1080×1080 30s (re-edited, key beats only)
YT long-form trailer 16:9 1920×1080 60s (re-edited, b-roll padded)

Checklist

  • [ ] Aspect and resolution match the platform row; no platform-side upscaling.
  • [ ] Length is under the 2026 cap and the cap was checked live for paid launches.
  • [ ] Captions, faces and CTA sit outside every UI band of every placement the cut will run in.
  • [ ] H.264 yuv420p, AAC 48 kHz stereo, +faststart, about -14 LUFS / -1 dBTP.
  • [ ] AI-disclosure and music-rights rows resolved for each platform.

References

04 同仓库更多 Skills