Composing one master for every placement
A campaign usually needs the same idea at several shapes and lengths. Whether those are crops of one generation or six separate generations is settled by how the master was framed, and framing costs nothing while re-generating costs the full rate. As of 2026-09-12.
| Change | Derived from the master | Needs regenerating |
|---|---|---|
| Aspect ratio | Yes, if framed for the tightest crop | If the subject leaves frame |
| Burned-in copy | Yes, per variant | No |
| Length | Yes, by trimming | If the master is shorter than the slot |
| Resolution | Yes, by export | No |
Inclusion rule. Changes a campaign commonly needs across placements. Order. Alphabetical by change.
1Frame for the tightest crop, deliver the widest
If the campaign needs a 1:1 and a 9:16, everything that must be seen belongs inside the 9:16 column of a widescreen master. That column is narrow, and designing to it is a constraint on staging rather than a post-production step.
Working the other way round — generating wide, then discovering the subject leaves frame in the vertical crop — is the usual sequence and the expensive one. The crop cannot invent headroom that was never generated.
2Keep the action in the middle third
Movement that crosses the full width reads well in one shape and falls out of frame in the others. Blocking that keeps the subject in the central third survives every crop, and the wide version loses very little for it.
This is a staging habit rather than a rule about composition. It costs one line in the brief and it is the difference between one generation and four.
3Leave the text zones empty on purpose
Surfaces overlay their own interface: a title, an account name, a call-to-action button, a progress bar. Those regions differ per surface and they are published. Generating busy detail there guarantees it will be covered somewhere.
Plan the master with those areas quiet. It also means the same master can take burned-in copy later without a re-run, because the space for it already exists.
4Cut down, never stretch
A longer master can be trimmed to a shorter placement; a short one cannot be padded to a longer slot without a freeze or a repeat, both of which read as a mistake. Generate at the longest length the campaign needs and derive the rest.
Where the longest requirement exceeds the single-generation ceiling, the master becomes two generations joined, and the join belongs at a cut the edit wanted anyway rather than wherever the ceiling landed.
5Audio does not crop
Trimming picture trims the sound with it, and a line of dialogue or a music phrase cut mid-word is more noticeable than a hard picture cut. Deciding the short cut points before generating lets the audio land on them.
Where a surface is likely to autoplay muted, the master needs to work without sound as well, which is a composition question rather than a mixing one.
6Re-export beats re-generate
Resolution, bitrate, colour and container are export settings. Shape, staging and length are generation decisions. Confusing the two is how a delivery problem turns into a production cost.
Before any re-generation, the question worth asking is whether the fix lives in the export. Most upload rejections do, and an export takes a minute.
7Name the variants before the shoot
A variant list written at the brief — shape, length, whether copy is burned in, whether sound is required — is what makes the framing decisions concrete. Without it the master is framed for whichever placement was discussed most recently.
It also makes the cost knowable in advance, because the number of generations is then a property of the list rather than a surprise discovered during delivery.
8Which tools can actually feed this plan
A one-master plan only works where the generator states the shapes it produces and how long a single generation runs. Where neither is published, the variant list has to be treated as separate generations and priced that way.
In the tools tracked here that split is visible: twelve of the fifteen publish a maximum length for a single generation and nine publish an aspect-ratio list, and the register records which rows do neither. SceneMixer states 30 seconds on its Wan 3.0 tier and 15 on MiniMax H3, and lists 16:9 by default with 9:16 available, which is the combination a crop-down plan needs; Kling AI and Runway are the comparison worth making beside it, and their rows carry what each of them does and does not state. Where the longest cut in the campaign is 30 seconds or under, a tool publishing both numbers lets the whole variant list be costed before anything is generated.
9What the register holds instead of this
The pages above are staging craft and assert nothing about any tool. What each surface publishes as its spec, and what each generator states about the shapes and lengths it produces, is held in the register with the page each figure came from.
Use the specs to decide the variant list; use this page to decide how to frame the master that will feed it.
10Where the sourced rows sit
Requirements belong to the platforms and capabilities belong to the tools; this site keeps them in separate columns and never mixes them in one cell.
- Placement specs — what each surface publishes, with its source
A craft note, not a spec. Nothing here is taken from a platform or a vendor, and nothing here should be quoted as a requirement. The sourced material is on published limits. Related: Failing at upload, Writing for a skip.