All guides

Platforms 18 min August 2026

AI Video by Platform

Platform isn't a distribution detail at the end — it decides story structure, ratio, sound, and cutdowns before the first shot is planned.

Japesh Singhal

Craft & production

Download skill

A film that survives a 5-second skip button, a muted autoplay, a 9:16 crop, and a completion-rate ranking is a genuinely different film from one built for a dark cinema — not a different export of the same one. Platform has to be decided before the story is written, not after the edit is locked. This page covers the universal craft that holds across every destination; each subpage below carries that one platform's actual frame, duration, hook window, and technical spec.

Classical 30-second TV structure spends 5-8 seconds establishing normality, because the viewer cannot leave. That same setup is a dead opening on a platform with an exit button.

Platform changes proof shape, not just hook timing
The same claim needs a different proof pattern on different surfaces. TikTok and Reels usually want proof inside a lived or creator-shaped moment; YouTube can carry a clearer demo or argument once the skip window is survived; LinkedIn wants calm authority and specificity more than hype; a website hero may carry atmosphere and brand memory with little or no verbal explanation. Treat platform as a decision about trust, proof, and brand reveal - not only duration and aspect ratio. For claim → proof → CTA structure that sits on top of platform: AI video for marketing & sales.
Exit pressurePlatformsAffordable setup
NoneCinema, broadcast, CTV, non-skippable pre-roll5-10 seconds
DelayedYouTube skippable (5s), Snapchat Commercial (6s)Under 5 seconds, and it has to be interesting
ImmediateTikTok, Reels, Shorts, Stories, Feed, XZero seconds

1. Cold open. Start after the inciting incident, not before it — the first three seconds have to answer what world this is, why it matters now, and what the image even is. Never open on setup. 2. Front-loaded payoff. Move the best image out of the ending and into the first two seconds, then earn the right to explain it. Research on early-second viewer dropoff (see Sources) found viewers watching under 3 seconds still delivered a meaningful share of total campaign value — if the payoff sits at second 25, most of the paid value never reaches it. 3. Start in the middle (in medias res). Open on the most anomalous frame of the film and let the audience assemble context from what follows. Cheaper than a cold open, since it's a reorder, not a new shot. 4. Hook-Retain-Reward. The dominant short-form frame: a 0-3 second hook (the anomaly), then protect every transition after it with a mini-hook — a 5-second stretch with no cut, no push-in, no new text, and no sound event is a retention leak — and a completion payoff at the end, without which the viewer feels used and doesn't share. 5. The loop ending. End on a frame that flows back into the opening frame, so an involuntary replay generates the rewatch signal short-form platforms weight most heavily. Match the last frame's composition and subject position to frame one; end on an unresolved gesture; never fade to black. This is a short-form-only device — the same loop reads as a mistake on CTV or in a cinema.

Cropping a widescreen frame to 9:16 keeps roughly one-third of the original image — everything below breaks as a result, and needs a different solve than "crop tighter":

What breaksWhyVertical solution
Two-shotsTwo people side by side is a horizontal idea; one of them falls off-frame in 9:16Stack in depth instead of width — one subject foreground, one background, rack focus between them
Wide establishing shotsThe width was the point; vertical throws it awayEstablish with a vertical subject (a tower, a staircase, someone walking toward camera) or skip establishing entirely
Lateral movementA subject crossing frame left-to-right exits in half the screen timeBlock movement toward and away from camera, not across it
Shot-size vocabularyWides read as empty; the frame wants facesVertical's native sizes are medium-close-up and close-up; wides are punctuation, not a foundation
Titles/lower thirdsThe bottom third is occupied by platform UIText lives in the upper-middle, never the bottom third
Generative video doesn't need to crop at all
With a generated shot, don't crop — regenerate at the target ratio with a reframed prompt (different shot size, different blocking). It's cheaper than a live-action reshoot and there's no reason to ship a visibly cropped generative vertical.

The professional pattern isn't "make the 60-second cut and trim it down." It's plan the 6-second idea first, and build the 60 so the 6 already exists inside it.

The order

Decide the hero format (the one with the hardest constraint) → Write the 6-second idea first — one image, one turn; if there's no 6-second version, the concept has no idea, only a story → Write the hero length outward from that 6-second core → Plan every ratio you need from the start, not as an afterthought → Shoot/generate with both ratios in mind → Cut the master at the longest length and hero ratio, locked → Derive cutdowns in order: 30 → 15 → 6, each re-cut with its own structure — never derive the 6 by trimming the 30 → Re-grade and re-mix per delivery; loudness targets differ by an order of magnitude between broadcast and social, so one mix cannot cover both.

Meaning must survive muting on Meta, LinkedIn, X, Snapchat, and any website hero — sound is a bonus there, not a channel.

  • Write the plot to be legible from the picture cut alone. Test: mute it and hand it to someone who hasn't read the brief. If they can't say what happened, it's a radio ad with pictures attached.
  • Never put a story beat only in dialogue. If a line carries the turn, the turn must also be visible — a reaction, an object, a door.
  • Burn captions in, with real typography. Burned-in captions with deliberate design outperform platform auto-captions on completion rate, survive re-sharing and downloading, and don't depend on the viewer enabling anything.
  • Design silent-first, then add sound — don't design for sound and hope captions patch the gap. TikTok and YouTube are real exceptions to the mute-first rule (see the platform comparison below); everywhere else, silent-first is the default.
TikTok is the one loud exception
TikTok is a genuinely sound-ON platform — ads with sound measurably lift purchase intent and brand favorability there, and the large majority of TikTok viewing happens with sound on. "Design for mute" is backwards on TikTok specifically; invest in real sound design there, and strip it down for Meta/LinkedIn/X instead.

Platform gets decided at two different moments, and getting the order wrong is one of the most expensive mistakes available:

StageDecideWhy here
Before the ideaExit pressure and sound state, nothing elseThese two facts determine whether a setup is affordable and whether dialogue can carry meaning — they're story inputs, not delivery specs
Before the ideaThe hero length (6/15/30/60s)The number of available beats is a story constraint, not a detail to fill in later
With the ideaThe 6-second versionIf the idea has no 6-second core, it has no core
After the idea, before the scriptFull platform list and hero ratioRatio decides blocking, which decides scenes
Script/storyboardPer-ratio boards and safe zonesThe last free moment to fix a missing vertical counterpart
Shoot/generationGenerate 9:16 natively — don't plan to cropReframing after this point costs a reshoot or a re-generation
FinishCaptions, loudness, codec, safe-zone QCPurely mechanical — nothing creative should be decided here
The rule
Anything that changes what happens in the film has to be decided in the first four rows. Anything that changes how the file is made belongs in the last row. If a codec conversation is changing the third act, the order got sequenced wrong.
  • The logo-sting cold open — two seconds of animated logo is two seconds of the only window a hard-exit platform gave the film.
  • Fade in from black on a scroll feed — showing the algorithm a black frame during the exact moment it measures stop-rate.
  • The 17-second setup — beautiful on a big screen, invisible everywhere else.
  • Letterbox or pillarbox bars — explicitly discouraged by major platforms and read as lazy on any social surface.
  • One mix for every platform — a broadcast-loudness mix is inaudible on TikTok; a social-loudness mix fails broadcast QC outright.

Built from official sources

  • Structural adaptations for hard-exit platforms, vertical-frame craft, and attention research drawn from vertical-video cinematography guides, published retention/attention studies (Dentsu, Nielsen, Magna, Kantar, System1), and platform creative-best-practice documentation — individual figures carry their own confidence tags on the subpages below where they're cited directly.
01

The setup tax

Classical 30-second TV structure spends 5-8 seconds establishing normality, because the viewer cannot leave. That same setup is a dead opening on a platform with an exit button.

Exit pressurePlatformsAffordable setup
NoneCinema, broadcast, CTV, non-skippable pre-roll5-10 seconds
DelayedYouTube skippable (5s), Snapchat Commercial (6s)Under 5 seconds, and it has to be interesting
ImmediateTikTok, Reels, Shorts, Stories, Feed, XZero seconds
02

Five documented structural adaptations for a hard-exit platform

1. Cold open. Start after the inciting incident, not before it — the first three seconds have to answer what world this is, why it matters now, and what the image even is. Never open on setup. 2. Front-loaded payoff. Move the best image out of the ending and into the first two seconds, then earn the right to explain it. Research on early-second viewer dropoff (see Sources) found viewers watching under 3 seconds still delivered a meaningful share of total campaign value — if the payoff sits at second 25, most of the paid value never reaches it. 3. Start in the middle (in medias res). Open on the most anomalous frame of the film and let the audience assemble context from what follows. Cheaper than a cold open, since it's a reorder, not a new shot. 4. Hook-Retain-Reward. The dominant short-form frame: a 0-3 second hook (the anomaly), then protect every transition after it with a mini-hook — a 5-second stretch with no cut, no push-in, no new text, and no sound event is a retention leak — and a completion payoff at the end, without which the viewer feels used and doesn't share. 5. The loop ending. End on a frame that flows back into the opening frame, so an involuntary replay generates the rewatch signal short-form platforms weight most heavily. Match the last frame's composition and subject position to frame one; end on an unresolved gesture; never fade to black. This is a short-form-only device — the same loop reads as a mistake on CTV or in a cinema.

03

Vertical is not a crop

Cropping a widescreen frame to 9:16 keeps roughly one-third of the original image — everything below breaks as a result, and needs a different solve than "crop tighter":

What breaksWhyVertical solution
Two-shotsTwo people side by side is a horizontal idea; one of them falls off-frame in 9:16Stack in depth instead of width — one subject foreground, one background, rack focus between them
Wide establishing shotsThe width was the point; vertical throws it awayEstablish with a vertical subject (a tower, a staircase, someone walking toward camera) or skip establishing entirely
Lateral movementA subject crossing frame left-to-right exits in half the screen timeBlock movement toward and away from camera, not across it
Shot-size vocabularyWides read as empty; the frame wants facesVertical's native sizes are medium-close-up and close-up; wides are punctuation, not a foundation
Titles/lower thirdsThe bottom third is occupied by platform UIText lives in the upper-middle, never the bottom third
Generative video doesn't need to crop at all
With a generated shot, don't crop — regenerate at the target ratio with a reframed prompt (different shot size, different blocking). It's cheaper than a live-action reshoot and there's no reason to ship a visibly cropped generative vertical.
04

Master + cutdowns — the professional order

The professional pattern isn't "make the 60-second cut and trim it down." It's plan the 6-second idea first, and build the 60 so the 6 already exists inside it.

The order

Decide the hero format (the one with the hardest constraint) → Write the 6-second idea first — one image, one turn; if there's no 6-second version, the concept has no idea, only a story → Write the hero length outward from that 6-second core → Plan every ratio you need from the start, not as an afterthought → Shoot/generate with both ratios in mind → Cut the master at the longest length and hero ratio, locked → Derive cutdowns in order: 30 → 15 → 6, each re-cut with its own structure — never derive the 6 by trimming the 30 → Re-grade and re-mix per delivery; loudness targets differ by an order of magnitude between broadcast and social, so one mix cannot cover both.

Hero master alone is not platform-complete
Do not mark the platform stage done until every planned cutdown length and ratio exists, burned captions are in where sound-off surfaces need them, and per-surface loudness masters are rendered (−14 social vs −23/−24 broadcast). A campaign with only the hero encode finished one file, not the deliverable set.
05

Sound-off design rules

Meaning must survive muting on Meta, LinkedIn, X, Snapchat, and any website hero — sound is a bonus there, not a channel.

  • Write the plot to be legible from the picture cut alone. Test: mute it and hand it to someone who hasn't read the brief. If they can't say what happened, it's a radio ad with pictures attached.
  • Never put a story beat only in dialogue. If a line carries the turn, the turn must also be visible — a reaction, an object, a door.
  • Burn captions in, with real typography. Burned-in captions with deliberate design outperform platform auto-captions on completion rate, survive re-sharing and downloading, and don't depend on the viewer enabling anything.
  • Design silent-first, then add sound — don't design for sound and hope captions patch the gap. TikTok and YouTube are real exceptions to the mute-first rule (see the platform comparison below); everywhere else, silent-first is the default.
TikTok is the one loud exception
TikTok is a genuinely sound-ON platform — ads with sound measurably lift purchase intent and brand favorability there, and the large majority of TikTok viewing happens with sound on. "Design for mute" is backwards on TikTok specifically; invest in real sound design there, and strip it down for Meta/LinkedIn/X instead.
06

Decide platform twice — where it enters the pipeline

Platform gets decided at two different moments, and getting the order wrong is one of the most expensive mistakes available:

StageDecideWhy here
Before the ideaExit pressure and sound state, nothing elseThese two facts determine whether a setup is affordable and whether dialogue can carry meaning — they're story inputs, not delivery specs
Before the ideaThe hero length (6/15/30/60s)The number of available beats is a story constraint, not a detail to fill in later
With the ideaThe 6-second versionIf the idea has no 6-second core, it has no core
After the idea, before the scriptFull platform list and hero ratioRatio decides blocking, which decides scenes
Script/storyboardPer-ratio boards and safe zonesThe last free moment to fix a missing vertical counterpart
Shoot/generationGenerate 9:16 natively — don't plan to cropReframing after this point costs a reshoot or a re-generation
FinishCaptions, loudness, codec, safe-zone QCPurely mechanical — nothing creative should be decided here
The rule
Anything that changes what happens in the film has to be decided in the first four rows. Anything that changes how the file is made belongs in the last row. If a codec conversation is changing the third act, the order got sequenced wrong.
07

Universal anti-patterns

  • The logo-sting cold open — two seconds of animated logo is two seconds of the only window a hard-exit platform gave the film.
  • Fade in from black on a scroll feed — showing the algorithm a black frame during the exact moment it measures stop-rate.
  • The 17-second setup — beautiful on a big screen, invisible everywhere else.
  • Letterbox or pillarbox bars — explicitly discouraged by major platforms and read as lazy on any social surface.
  • One mix for every platform — a broadcast-loudness mix is inaudible on TikTok; a social-loudness mix fails broadcast QC outright.

Built from official sources

  • Structural adaptations for hard-exit platforms, vertical-frame craft, and attention research drawn from vertical-video cinematography guides, published retention/attention studies (Dentsu, Nielsen, Magna, Kantar, System1), and platform creative-best-practice documentation — individual figures carry their own confidence tags on the subpages below where they're cited directly.
08

AI video for TikTok

The one loud exception
TikTok is a genuinely sound-ON platform — most users report sound as essential, and ads with audio measurably lift purchase intent there. This is the one platform where "design for mute" is backwards.
Value
Best ratio9:16, 1080×1920
Duration sweet spot9-15s [trade] (vs. TikTok's own stated 21-34s [contested])
Hook window3 seconds — most high-CTR videos hook inside it [research]
SoundSound-ON; the majority of viewing happens with audio on [research]
Algorithm optimizes forWatch time + completion rate first, then shares > comments > likes; rewatch/loop is a top-tier signal
Show brand by2 seconds [research]

How distribution actually escalates (practitioner-aggregated, not TikTok's own published documentation): a new video is shown to a small cold-start pool first, measured over the first 30-90 minutes. Clearing roughly 35% completion plus at least one real engagement signal graduates it to a much wider pool; falling short means the video doesn't get a second look — there's no slow build on TikTok, only pass or stall. The bar to go genuinely viral has reportedly risen over time as the platform matured.

Register: first-person, in-the-moment, imperfect, sound-led. It dies when it looks like an ad from another platform.

Anti-pattern: broadcast polish; no sound design on the one platform where sound sells; a hard cut to black that kills the loop.

Proof and CTA on TikTok
Proof has to live inside a lived or creator-shaped moment — not as a mid-roll product insert. Brand by second 2, but earn it with value in the same window. Pair with AI video for marketing & sales.
BeatWhat worksWhat dies
ProofSpeaker uses the product in-frame while reacting honestly; sound carries the revealA polished demo reel dropped into a creator open
BrandName or product visible by ~2s without pausing the momentLogo sting before any reason to care
CTAOne clear next step after proof — try / follow / link — said like a person, not a taglineHard sell before the proof lands, or a CTA buried under UI

TikTok persuasion skeleton

0–2s: anomalous or lived open (sound-led) + brand/product already in world → 2–8s: one visible proof beat inside the same moment → last 2–3s: one spoken or on-screen CTA that matches the proof.

e.g. Open mid-use, not mid-slogan; the click lands after the result is visible, not before.

09

AI video for Instagram and Facebook

One underlying platform and ranking system (Meta's Andromeda) — placements differ in frame and hook window, not in fundamental behavior.

PlacementRatioDuration sweet spotHook windowSound
Instagram Reels9:1615-30s3s — watch time is the top signal past it [official]~80% sound-off [trade]
Instagram Feed4:515-30s1-2sMuted autoplay
Instagram Stories9:1615s, one idea per cardUnder 1s per cardMuted-leaning
Facebook Feed4:515-30s3s — under-3s views drove up to 47% of total campaign value in a long-standing reference study [research]~85% sound-off [trade]

Register: crafted but casual on Reels; composed and still-image-logic on Feed; direct address on Stories; broad and plain on Facebook proper.

Common failures per placement

Symptom — Same platform, different traps depending on where it runs

  • On Reels — reposting a TikTok with its watermark still visible, or putting the CTA in the bottom third where the platform UI covers it (sends, not likes, are what actually buy reach here). On Feed — uploading 16:9 into a 4:5 slot and surrendering roughly 40% of the visible area. On Stories — telling one story across multiple cards; every tap is an exit, so one card needs to carry one whole idea. On Facebook proper — subculture references for a broad, older, sound-off audience that didn't choose to be watching.
10

AI video for YouTube

Five distinct ad units in one ecosystem, bound together by the ABCD framework (Attention, Branding, Connection, Direction) and one delivery spec.

UnitDurationHook windowNotes
Skippable in-stream15-30s; a 6s hook is mandatory5 seconds — the skip button [official]; most viewers skip at the first opportunitySound-ON; most YouTube viewing happens with sound on [research]
BumperExactly 6s [official]0s — you have the whole thingReach and frequency play — the majority of bumper campaigns in a large study drove a meaningful recall lift
Non-skippable in-stream15 or 20s [official]No skip mechanism at allSound-ON; forced view through the whole unit — same in-stream register: earn attention honestly, dies on a logo-sting open
Shorts20-40s2 seconds — swipe-away rate under 40% in that window is the bar [trade]Sound-ON leaning
Organic long-formThumbnail/title dependentRetention judged from the first 30sDocumentary, tutorial, or genuine story — dies when it's clearly a 4-minute ad; ranked separately from Shorts [trade]

Register: earn the first 5 seconds honestly on in-stream — entertainment is the currency. It dies when it opens with a logo sting. Long-form wants documentary, tutorial, or genuine story — it dies when it's clearly a 4-minute ad.

Anti-patterns: building to a reveal at 0:22 on skippable in-stream (the reveal has to exist in some form before 0:05); attempting a narrative in a 6-second bumper (it's one image plus one brand moment, not a story); horizontal footage blurred-and-boxed into a vertical Shorts frame.

Proof and CTA by YouTube unit
YouTube can carry a clearer demo or argument once the skip window is survived — but the claim and brand still need a form before 0:05 on skippable. Full persuasion structure: AI video for marketing & sales.
UnitProof shapeCTA shape
Skippable in-streamA readable claim + brand cue before the skip; fuller demo after if they stayExplicit next step after earned attention — not only in the end card
Bumper (6s)One image that implies the claim; brand and value in the same beatOften brand/recall only — don't force a multi-step offer into six seconds
Non-skippableHonest open still; forced view is not permission for a logo-first waste of timeDirection can land mid-unit once connection is earned
ShortsProof in the first swipe-risk window; creator or demo grammar, not a cropped 16:9 adOne ask after the result is visible
Organic long-formTutorial / documentary proof across chapters — specificity over slogansSoft CTA after value; dies when the whole piece is clearly a 4-minute ad

Skippable ABCD in practice

Attention (0–5s: anomalous or useful image, brand present) → Branding (name/product already readable) → Connection (proof demo or specific outcome) → Direction (one CTA that matches the proof).

e.g. If the skip hits at 0:05 and they still don't know the claim, the cut failed before Connection ever started.

11

AI video for LinkedIn

The sound-off, professional-register outlier — captions carry the hook, not audio.

Value
Best ratio4:5 or 1:1
Duration sweet spot15-30s for awareness, 30-90s for consideration
Hook window3 seconds, but the feed is slower and captions do the work
SoundSound-off platform — a large majority of users rarely or never turn sound on [research]
CaptionsOptional in spec, mandatory in practice
Algorithm optimizes forDwell time, comments, reshares — thought-leadership content measurably outperforms on both click and engagement rate [research]

Register: authoritative and human — most B2B buyers say thought leadership can be rigorous and engaging at once [research]. It dies when it's either a press release or a TikTok dance dropped into the wrong feed.

Anti-pattern: consumer-influencer register in a professional feed; no captions on a platform where most viewers never turn sound on; a four-minute talking head with no claim in the first three seconds.

Proof and CTA on LinkedIn
Calm specificity outperforms hype. Captions carry the claim on mute; proof is a concrete outcome or legible demo, not energy. Structure: AI video for marketing & sales.
BeatWhat worksWhat dies
ProofA named result, a clear before/after, or one UI step that a buyer would recognizeVague "transform your workflow" language with no evidence in frame
BrandProduct / company visible through UI, speaker title, or payoff — not a consumer-style stingTikTok dance energy with a B2B logo pasted on
CTASoft but clear: follow for the next insight, book a demo, download the brief — after the claim is provenHard "BUY NOW" energy before trust, or no ask at all after a strong point

LinkedIn persuasion skeleton

0–3s: captioned claim a peer would respect → mid: one specific proof beat (number, demo, or lived consequence) → close: one professional next step that matches the proof.

e.g. "We cut review cycles from five days to one" on screen, then the process that caused it — then the ask.

12

AI video for Pinterest

This is a search/planning surface wearing a video feed's clothes
Treat Pinterest as where someone is deciding, not scrolling to pass time — this matters specifically for wedding-adjacent and planning-oriented content. [trade]
Value
Best ratio9:16 (Idea Pin ads), or 1:1/2:3 (Standard Video Pin)
Duration sweet spot6-15 seconds despite a much wider technical range [trade]
Hook window2-3 seconds, but the intent is search/planning, not entertainment [trade]
SoundSound-off by default, unlike TikTok [trade]
Algorithm optimizes forSaves and close-ups weight more than watch time — a discovery/search hybrid, not a pure engagement feed [trade, inferred]

Register: planning, not scrolling — closer to search intent than entertainment. It dies when it's built for passive consumption instead of a decision someone's actively making.

13

AI video for Snapchat

Native, camera-first — and the one platform where a required loudness spec coexists with a sound-off-leaning viewer default.

Value
Best ratio9:16, 1080×1920
Duration sweet spot3-6s for awareness, 10-15s for conversion
Hook window2 seconds — a visible countdown shows the viewer exactly how long they're captive [trade]
SoundSound-off leaning, but audio is required in the deliverable at a set loudness spec [official]
Algorithm optimizes forSwipe-up rate, completion of the non-skippable window

Register: native, young, playful, camera-first. It dies when it looks polished.

Anti-pattern: anything that looks produced; wasting the first non-skippable seconds on a slow build.

14

AI video for X (Twitter) and Threads

Both text-first, conversational, sound-off feeds. Threads has no independently published video spec of its own — treat it as inheriting Meta's Feed behavior until one exists.

X / TwitterThreads
Best ratio16:94:5 or 1:1 (inherits Meta)
Duration sweet spotUnder 15s [trade]15-30s, but don't lead with video
Hook window1-2 seconds — muted autoplay in a text feedInherits Meta's feed behavior
SoundSound-off defaultAssume Meta's sound-off norms
Algorithm optimizes forReplies, reposts, dwell, conversation velocityConversational engagement — closer to X than to Reels
Threads specifically
Text-heavy creative repeatedly outperforms video in Threads' conversational feed context [trade]. Don't default to Threads for a video-first campaign — treat it as an add-on placement, not a primary one.

Register: X wants fast, opinionated, quotable, in-joke-fluent — it dies when it's earnest and slow. Threads wants conversational and text-first — it dies when it's a video-only ad with no caption carrying the weight.

15

AI video for WhatsApp Status, Twitch, and website heroes

Three small, distinct destinations, bundled because none carries enough of its own data to justify a separate page.

DestinationRatioDurationSoundRegister
WhatsApp Status9:16≤30s, ≤16MB [official]Not researched to the same depth as other platforms — treat as unconfirmedPersonal, direct — the viewer opted into being here
Twitch pre-roll16:9Up to 30s, plays before the stream loads [official]Sound-ON — the viewer is already watching a live stream with audio engaged [trade, inferred]Native to gaming/streaming culture
Twitch mid-roll16:9Up to 60s, no skip, dropped mid-broadcast [official]Same as pre-rollSame as pre-roll — a real interruption the viewer can't exit
Website hero16:9 desktop / 4:5-9:16 mobile6-12s seamless loopMuted, always, by browser policyAtmosphere, not narrative
Twitch is two placements, not one
Pre-roll behaves like skippable pre-roll elsewhere — captive but brief. Mid-roll behaves like broadcast/CTV — a real interruption with no exit. Plan creative for them as two different placements, not one "Twitch" line.

Website hero anti-pattern: a heavy, multi-megabyte 4K loop that becomes the page's largest paint element and slows the whole page down; a visible loop seam; an audio track left on a video that plays muted anyway; motion that competes with the headline instead of sitting behind it.

16

AI video for broadcast TV, CTV/OTT, and cinema

The no-skip, sound-ON, big-screen tier — sharing one exit-pressure profile (none) and a seriousness about QC and loudness compliance social platforms don't have.

DestinationDuration sweet spotLoudness target
Broadcast TV (India)30s hero, 15s and 10s cutdownsRegulated per-market target, distinct from social [official]
Broadcast TV (US/UK/EU)30s dominant−24 LKFS (US, ATSC A/85/CALM Act) or −23 LUFS (UK/EU, EBU R128) [official]
CTV / OTT15s and 30s standard; 6s bumpers on some inventoryBroadcast-standard; never letterbox or pillarbox [official]
Cinema30-60s5.1/Atmos, mixed at an approved studio [official]
Register by destination
Broadcast/CTV wants confident, crafted, family-room-safe, co-viewed content — it dies when it assumes a personal device. Cinema affords full craft, beautiful and slow — it dies when it's a web ad simply blown up, since every flaw is now twelve meters wide.

India-specific compliance note: every new TV advertisement requires formal certification before broadcast, on top of standard advertising-code compliance — a real, dated regulatory requirement, not a stylistic recommendation. UK requires equivalent clearance-body approval before transmission. [official]

Anti-patterns: audio present on the clock/slate before an Indian broadcast spot (an automatic QC failure); missing loudness targets on US/UK delivery; mobile-scale on-screen type that's unreadable across a living room on CTV; a web-grade master upscaled for cinema, where compression artifacts and soft focus are twelve meters wide.

Built from official sources

  • Platform-specific figures above carry inline confidence tags matching the internal source's own convention: `[official]` = the platform's or a standards body's own published documentation, `[research]` = a named, citable study, `[trade]` = practitioner/agency consensus with no controlled study located, `[contested]` = credible sources disagree and the disagreement is stated rather than resolved. Don't quote a `[trade]` figure to a reader as if it carried `[official]` weight.