NodeValid

Published video-model weights, licences and hardware notes

SVD calls four seconds a limitation

The card states 25 frames at 576x1024 and separately describes the generated videos as rather short, at four seconds or less. The second sentence is written as a limitation, which is unusual candour for a model card. As of 2026-09-22.

Stable Video Diffusion in the length column, as publishedA length written as an apology rather than as a figure still bounds the output, and it tells a reader how the maintainers saw it.Length as published25 frames, 4sec or lessframes, seconds or bothDescribed as rather short,not as a ceilingOutput size576x1024pixels per frameWhat the model was trainedto generateMemoryAbout 180s on an A100 80GB cardwhat holding them costA timing on a named card; nomemory floor statedOne clip, as this release describes itStated as a shortcoming
Fig. 1 The shortest published output in the register, described as short by its own card.
Stable Video Diffusion in the length column, with the columns that qualify it. Read from huggingface.co/stabilityai on 2026-09-22.
ColumnWhat the pages state
Length, in the vendor's own words25 frames, 4sec or less
The detail published beside itDescribed as rather short, not as a ceiling
Resolution, from the same pages576x1024
Hardware, from the same pagesAbout 180s on an A100 80GB card

Inclusion rule. The value this page is about, the detail the vendor attached to it, and the two columns that change how it should be read. A column the vendor left empty keeps its row and says so. Order. This column first, then its detail, then the two columns that qualify it.

1Candour changes how the value reads

A card that states four seconds as a specification invites a reader to build around it. One that calls it rather short tells them the maintainers expected it to be a problem.

The register keeps the wording. Recording four seconds alone would drop the part that tells a production to plan for assembly rather than for single takes.

2Twenty-five frames, and the rate follows

Twenty-five frames in four seconds or less puts the rate somewhere around six or seven per second, which is low. The card does not state a rate, and the arithmetic here is a reader's rather than the vendor's.

So the register records both published values and leaves the implied rate out of the entry. Two stated numbers are worth more than a third that nobody wrote.

3The shortest output in the register

Against sixty seconds on one release and a minute at 30fps on another, four seconds is the low end by a wide margin. It is also one of the oldest releases here.

That is worth holding onto when comparing published lengths across the register: the column mixes releases from different years, and length is where the field moved fastest.

4Sources

Lengths above are in the unit the vendor chose, from huggingface.co/stabilityai on 2026-09-22, with no conversion applied. The full entry is at this family's page; frames against seconds is on length. Nearby: A minute, at a price, Ten seconds, or five.