NodeValid

Published video-model weights, licences and hardware notes

A 2B release that states no memory floor

The card states 2B params. It does not state a memory floor: the hardware line is a timing, about 180s on an A100 80GB card, which tells a reader about speed on a large GPU and nothing about the smallest one that would work. As of 2026-09-22.

Stable Video Diffusion in the size column, as publishedThe smallest published count in the register sits beside the least usable hardware statement, which is the reverse of what a reader would expect.Fixed by the published countLeft to the runThe count2B paramsWhat it belongs toThe figure shown on the card itselfLicence on that releasestable-video-diffusion-communityMemory for a runAbout 180s on an A100 80GB cardOutput size576x1024A timing is not a floor
Fig. 1 Two billion parameters and a stopwatch, where a memory figure would have helped more.
Stable Video Diffusion in the size column, with the columns that qualify it. Read from huggingface.co/stabilityai on 2026-09-22.
ColumnWhat the pages state
Size, in the vendor's own words2B params
The detail published beside itThe figure shown on the card itself
Licence, from the same pagesstable-video-diffusion-community
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.

1Small models are where a floor matters most

A 2B release is the one a reader is most likely to try on hardware they already own. That is exactly the case a stated minimum would serve, and it is the case this card leaves open.

Another 2B release in the register publishes 32.54 GB to two decimal places. Same order of size, opposite approach to documenting the run.

2What the timing does tell you

That the reference run happened on an 80GB card, and roughly how long it took. Both are useful for capacity planning if the plan involves that class of GPU, and neither transfers down.

A timing also ages differently from a memory figure. Inference code gets faster; the weights do not shrink. The register keeps the card the number belongs to.

3Why the register leaves the gap visible

Filling it with a community estimate would put a number in a register whose whole claim is that its numbers are published. The absence is recorded as an absence.

It is also a finding in its own right. A reader scanning the hardware column for small models will see that one of the two smallest releases here declines to answer.

4Sources

The count comes from huggingface.co/stabilityai, read 2026-09-22, exactly as the vendor wrote it. Everything else this release publishes is on its family entry; why the count sits in its own column is on size. Nearby: 13B against 6GB, Three sizes, one table.