What a per-second rate does and does not include, and which product you are actually calling.
Aspect ratio is not a price lever on a per-second model. What that frees you to do, and the order it changes in a production.
Most of the cost models I have worked with in the last decade punish vertical video. A 9:16 master is a re-frame, a re-grade, sometimes a re-shoot, and always a line item. Agencies quote it as a deliverable. Stock libraries charge for it separately.
MiniMax H3 bills per output second, and ignores the aspect ratio. Six shapes — 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 — all at the same per-second rate, because the upstream meters seconds.
The interesting part is not that vertical is free. It is what stops being a decision.
When vertical costs extra, it gets decided at the top of the project: which platform is primary, and does the budget stretch to a second cut. That decision then propagates into everything. You frame for 16:9 with "protect for vertical", which means you compose loose, which means the 16:9 is slightly worse than it could have been and the 9:16 is a crop that never quite works.
The compromise framing is the real cost, and it never showed up on the invoice.
You stop protecting and start generating twice.
Two renders, each composed natively for its shape — the vertical one framed tight on the subject, the wide one using the space — cost exactly what one render of each length costs. A fifteen-second pair at 768P is 250 credits rather than 125; there is no vertical surcharge on top.
That is a different kind of decision. It is no longer "can we afford the vertical", it is "is this shot worth two renders". Often the answer is no and you take the 16:9 alone. But when the answer is yes, you get two properly composed masters instead of one master and one crop.
Marketplace listings. Amazon and Shopify both want vertical for mobile placements. Generating a native 3:4 or 9:16 rather than cropping a 16:9 means the product occupies the frame the way it should.
Ad testing. If you are testing creative, the shape is a variable like any other. When the shape is free, you can test it.
Social cutdowns. A 1:1 that was composed as 1:1 reads noticeably better than a centre-crop of a wide shot, and the difference is most visible exactly where it matters — the small preview in a feed.
Not the price, but the framing rule: the model composes for the frame you give it. A prompt written for 21:9 — "the figure small in the left third, the horizon low" — does not transfer to 9:16, where the left third is barely a sliver. If you generate both shapes, write the composition line twice.
This is the one place where "free" is misleading. The render costs the same; the brief does not. Reusing the identical prompt across two aspect ratios is how you get one good shot and one that looks like the model was confused, and then blame the model.
References are also free — nine images cost the same as none. Combined with flat-rate aspect ratios, the pricing shape is unusually simple: you pay for seconds and resolution, and nothing else. Five credits for the clip, then eight credits per second at 768P or thirteen at 2K.
Simple pricing is underrated. It means the cost of an idea can be estimated before you start, which is the precondition for deciding an idea is not worth it.
I would not plan a campaign budget on a number in a blog post. The current per-second rates, the flat clip fee and the refund policy for failed generations are all on the pricing page, and the full constraint list — durations, resolutions, the six ratios, the reference caps — is on a dated spec sheet with each row linked to its source.
If you want to see what a native vertical looks like next to a cropped one before you take my word for it, the first clip runs without an account.
Flat-rate aspect ratios do not save you money. They remove a decision that was making your framing worse. Generate each shape natively, write the composition line for each, and stop protecting for a crop you were never happy with.