Generate vertical natively, or reframe afterwards
A 9:16 master versus generating wide and cropping down, compared on framing, wasted resolution and credits per ratio delivered. With the threshold rule.
Ask five people how to get a campaign out in three aspect ratios and you'll get two answers. Generate each ratio natively, one master per destination. Or generate one wide master and crop it down. The argument usually gets settled by whoever feels strongly, because the two obvious inputs — quality and cost — both look like ties from a distance. They aren't ties. One of them is decided by a fact you can check in the catalog, and the other by arithmetic that most people never do.
Aspect ratio is not a billing dimension
Start with the cost question, because it has a clean answer. Across every model in Versely's catalog that publishes a price matrix, not one prices by aspect ratio. The dimensions models bill on are duration, output resolution, whether a native audio pass runs, or a flat per-generation rate. Shape is not among them.
The consequence: a 9:16 generation and a 16:9 generation of the same length, at the same resolution, on the same model, cost the same credits. There is no vertical surcharge and no widescreen discount. Anyone arguing for a wide master on the grounds that vertical costs more is arguing from a premise that isn't true.
Availability is a tie too. Across the whole catalog, 193 models support 16:9 and 193 support 9:16; narrow it to the 146 video endpoints and it's 89 either way. Exact parity, not a compromise. Native vertical isn't the constrained option it was two years ago.
So the real decision is about how many generations you pay for, and about resolution.
The crop math nobody runs
Reframing 16:9 to 9:16 with a full-height centre crop keeps 81/256 of the original width. That's 31.6% — you are discarding roughly two-thirds of the frame, horizontally. (The reverse, 9:16 up to 16:9 at full width, keeps the same 31.6% of the height. The geometry is symmetric.)
Put pixels on it:
| Source | Full-height 9:16 crop | Usable? |
|---|---|---|
| 1920 × 1080 (16:9 HD) | 607 × 1080 | No. 607 px wide is below every vertical spec |
| 2560 × 1440 | 810 × 1440 | Marginal |
| 3413 × 1920 (16:9 at 1920 tall) | 1080 × 1920 | Yes — exactly 1080p vertical |
That last row is the one that reframes the whole argument. To land a 1080-wide vertical from a widescreen master, the widescreen master has to be roughly 3.4K across. That's a 4K-class render, and 4K is a billing dimension — unlike aspect ratio, resolution moves the credit cost on a large share of the video catalog. How much depends heavily on the model: some carry a modest 4K premium over HD, others are several times the HD rate. What 4K video costs instead of 1080p has the scenario written out, and the individual model listing is the authority for any given endpoint.
So "generate wide and crop" is not one generation instead of three. It's one expensive generation instead of three ordinary ones, and whether that trade wins depends entirely on how many ratios you're actually shipping.
What you lose that isn't pixels
Resolution is the measurable loss. The compositional one is bigger and harder to argue about in a review.
A model composes for the frame it was given. Ask for 16:9 and it places the subject, the headroom and the negative space according to widescreen convention — subject off-centre, air to one side, information spread horizontally because there's horizontal room. Crop that to 9:16 and you are not extracting a vertical composition, you are salvaging one. The subject that sat elegantly on a third is now hard against an edge or dead centre with nothing above the head.
Then the safe zones move. Platform UI eats a different part of a 9:16 frame than a 16:9 one, and any text or logo that was clear of the chrome in the wide master may not be after the crop — the safe area moves with the frame, so caption placement has to be re-checked after every reframe, not assumed. The vertical ad safe zone map and designing once for 1.91:1 through 4:5 both cover the discipline for composing something that survives a crop, which is a real skill and a real constraint on what you can put in the frame.
Native generation has none of this. You ask for the aspect ratio you're delivering and the model composes for it. The subject sits where a vertical subject should sit.
The rule
One destination: generate native. There is no upside to a wide master you'll never ship wide. You pay the same credits per generation either way, you get a composition built for the frame, and you skip the crop entirely.
Three or more destinations: generate wide, then reframe — with the resolution caveat above. Budget for a source tall enough that the tightest crop still lands at your delivery resolution, and treat that higher render as the price of the approach rather than a surprise. Then reframe in the editor: edit_video takes an aspect_ratio parameter that reformats the whole output, and changing a video's aspect ratio is the task page. Iterate at 480p first — the preview pass renders free, subject to a short per-user cooldown — and export once per ratio when the framing reads correctly.
Two destinations: it depends on how far apart they are. 9:16 and 4:5 are close enough that one vertical master serves both with a light crop. 9:16 and 16:9 are opposite ends of the space; treat those as two destinations that each deserve a native master, because the crop between them destroys the composition in whichever direction you go.
There's a third path worth knowing about for the two-destination case: generate native to the tighter ratio and outpaint outward to the wider one. That adds picture information rather than discarding it, which inverts the resolution problem — but it's an extra generation per ratio and the extended regions are invented, so it earns its place on static-background shots and struggles on anything with structure running to the edge.
Running it either way
For the native path:
- Decide the destinations before you write the prompt. Not after.
- Generate one master per ratio, composing the prompt for that frame — the subject's position in a vertical is a different instruction from its position in a widescreen.
- Assemble each on its own timeline. The editor is EDL-based, so the final export bills once per timeline no matter how many clips are on it.
For the reframe path:
- Pick the delivery resolution for the tightest crop you'll ship, then work backwards to the source resolution the wide master needs.
- Generate the wide master once, at that resolution.
- Reframe with
edit_video, one call per ratio, previewing at 480p before each export. Making video vertical and cropping to square are the per-platform versions. - Re-check text and logo placement in every output. The safe area moved.
- Watch the crop, not the clip. A reframe fails at the edges, and the edges are where you aren't looking.
FAQ
Does generating 9:16 cost more credits than 16:9?
No. Aspect ratio doesn't appear as a pricing dimension anywhere in the catalog — models price on duration, resolution, audio and flat rates. Same model, same length, same resolution means the same credit cost regardless of shape. The pricing page has the plan and credit structure, and each model listing shows its own cost for the settings you pick.
Can the editor add picture information when I reframe to a wider ratio?
No. Reframing crops or letterboxes existing footage — it can't invent picture outside the original frame. If you need to go wider than you generated, that's an outpainting job, which is a generation rather than an edit, and it invents the new regions rather than recovering them.
Is upscaling the crop a way out of the resolution problem?
Partly, and know what you're buying. Upscaling raises resolution by inventing plausible detail rather than recovering detail that was never recorded, so a 607-pixel-wide crop upscaled to 1080 is smoother and larger, not sharper in any meaningful sense. It's a reasonable rescue for footage you already have and a poor plan to design around.
What if I don't know the destinations yet?
Generate native to your most likely single destination and accept that a second ratio means a second generation. Speculatively rendering a wide 4K master for ratios you may never ship is the most common way this decision goes wrong — you pay the resolution premium up front for optionality that half the time goes unused.