Placement questions answered from published specs

A question belongs here when a published spec and a published generator limit settle it between them. Anything that depends on an auction, a review team or creative judgement is out of scope. As of 2026-09-12.

Fig. 1 The left branch changes a production schedule. The right branch changes a media plan, and nothing on a spec sheet can inform it.

That excludes most of what an advertiser worries about — whether an ad will be approved, whether it will perform, what a review team will object to. Those are decisions made by people with access to systems no published page describes.

What is left is arithmetic, and it is arithmetic that decides schedules. Whether a placement can be filled by one generation or needs an assembly step changes the production plan, and it can be worked out before anyone opens an editor.

Every answer shows both sides of the comparison and links to the page each side came from, so a number that has since moved can be spotted rather than inherited.

1The questions

TikTok specs — TikTok publishes aspect ratios with minimum pixel sizes, a ten minute ceiling and a file size limit, which is almost the reverse of YouTube.
Shape by platform — One platform names a widescreen placement, another publishes three ratios with pixel floors, and six of the 15 tools state no shape at all.
File limits — One platform publishes a file size cap and a bitrate floor.
Ten seconds — Shorts counts a view at ten seconds or the end.
Shape first — Masthead is widescreen or 16:9.
Resolutions — Fourteen of the 15 tools in this register publish a resolution.
The first five — Skippable in-stream lets the viewer skip after five seconds and charges a view at thirty.
Past the ceiling — Three tools publish a single call longer than a minute.
The hard ceiling — Four published duration numbers: two of them can reject an upload, one is a reservation term, and one sits far above any cut on this sheet.
Emitted lengths — A placement states a range and a generator states a list of settings.
Spark Ads — One ad product states no duration restriction, because the video it runs was already accepted through the platform's organic upload path.
Where figures were read — Four entries were read from a rate card and eleven from documentation.
Length tied to a tier — Two pages make length depend on the resolution tier, and they point in opposite directions.
Who publishes a rate — Three entries state a frame rate and neither ad surface mentions one.
How long output stays — Two pages publish a retention window: two days on one and twenty-four hours on the other.
Floors, not just ceilings — Four entries state a minimum length as well as a maximum, from one second to five.
Length in frames — Two model cards state a frame count rather than a run time, and one of them adds a divisibility rule that a conversion to seconds would destroy.
Single values — Three entries state a single resolution with no cheaper rung and no higher one.
Published defaults — Four entries publish a default shape and every one of them is horizontal.
A minute in one call — Three entries clear sixty seconds without a join.
Ceilings on inputs — Two entries publish input ceilings, from five megabytes to two hundred.
Sound that arrives — One entry states that generated videos include an audio track by default.
Wider than widescreen — Four entries list twenty-one by nine and one of those also lists nine by twenty-one.
The square option — Six entries offer a square frame, and one surface here accepts square ads from 640 by 640 upward.
Identifiers and ceilings — Three entries make a ceiling depend on which model identifier or version a request names, which turns the identifier into a specification field.
Nothing about shape — Six entries say nothing about shape.
Cards, not products — Two entries are open-weight model cards.
Thirty seconds in calls — The Masthead autoplay window and the paid-view mark both land at thirty seconds.
Who publishes a minimum — Minimums appear on both sides of this register and mean opposite things: a floor a file must clear, and a floor below which nothing can be asked for.
Ad lengths — One generator here finishes ads at 15, 30 or 60 seconds.
Call to action — Platforms add call-to-action buttons and captions over an ad.
Product photo — Most generators here publish only a file ceiling for an input image.

Where no published number exists on one side of a comparison, the question is answered by naming the gap.

Elsewhere in the kit: Placements, Tools, Axes, Pairs, Terms, Learn, Data. How a spec gets onto a sheet is set out on the method page.

2A generator's ad pipeline

For ad work around “Questions”, SceneMixer's pages explain how its ad projects are produced:

An AI short drama is a serialized show of 60–120 second vertical episodes whose cast, sets and footage are generated by AI. Making one takes six steps: analyze the script into episodes, lock a consistent cast with reference art, localize for your market, write and edit storyboards, generate each 4–15 second shot, then composite and publish.
SceneMixer, How to Make an AI Short Drama, Step by Step (read 2026-10-13)

A serialized vertical-video show (typically 60–120 seconds per episode, 9:16) where the cast, scenes, and footage are AI-generated.
SceneMixer, The AI Micro Drama Production Pipeline — “What is an AI micro drama?” (read 2026-10-13)