NodeValid

Published video-model weights, licences and hardware notes

Aspect ratio, and why it is a production fact

Aspect ratio is the shape of the frame. It decides whether generated output drops into a vertical or a horizontal timeline without reframing, and it is usually readable only from the dimensions rather than from a tier label. As of 2026-09-22.

Three frame shapes in this registerPublished frames here include portrait, landscape and square, and each implies a different production pipeline around the same model.Shapes that appear in published dimensions herePortraitTaller than wide, as several releases publish their dimensions.LandscapeWider than tall, the shape most tier labels assume.SquareOne release sets a square default explicitly, for speed.Shape first, size second
Fig. 1 The shape a release was documented at, which is a decision rather than a limit.
Where a reader can find the frame shape. Recorded 2026-09-22.
SourceReliability
Written-out dimensionsReliable
A tier labelNot indicated
A sample clipSuggestive, not stated
A memory table configurationReliable, but it is a setting

Inclusion rule. Places the frame shape can be read from. A source that only suggests it is listed, because it is the one people use. Order. From the most reliable source to the least.

1Square defaults are a speed decision

One release in this register sets 720x720 as its default and says the choice was made to accelerate inference, while describing arbitrary resolutions as supported. The default is a configuration rather than a capability.

So the published hardware figures for that release describe a square frame, and a production needing vertical output has no figure that covers its actual case.

2Reframing is not free

Cropping a landscape generation to vertical throws away pixels that were generated at cost, and it moves the composition the model chose. Generating at the target shape avoids both.

That is why the register writes dimensions in the order the vendor gave them rather than normalising to width by height. The order is part of what was published.

3Which shapes appear in the register

Published dimensions here include portrait frames, landscape frames and one square default chosen for speed. The shape is readable from written-out dimensions and not from a tier name.

Three families publish no output size at all, so their frame shape is unknown from their own pages. For anyone planning vertical output that is the first missing value rather than the last.

A reading note, not an entry: no vendor value appears on this page. Where the register records this term for a particular release, it is on the resolution column and on that family's own page. Nearby terms: Pixel count, Context frame.