NodeValid

Published video-model weights, licences and hardware notes

CogVideoX gives 1360x768 and 720x480

CogVideoX1.5-5B states 1360x768 and the 5B states 720x480. Both are output sizes rather than test settings, and each is attached to the release it belongs to. As of 2026-09-12.

CogVideoX in the resolution column, as publishedTwo output sizes in one family with the release named beside each is the form that lets a reader pick a model from a requirement rather than the other way round.Output size1360x768 on 1.5-5Bsizes the memoryfigureMemory it sizesFrom 4GB in FP16 on the2Bheld over thismany framesClip lengthUp to 161 frames on1.5-5BEach size names its release720x480 on the 5B
Fig. 1 Choosing by output instead of by parameter count, which this family makes possible.
CogVideoX in the resolution column, with the columns that qualify it. Read from github.com/THUDM on 2026-09-12.
ColumnWhat the pages state
Resolution, in the vendor's own words1360x768 on 1.5-5B
The detail published beside it720x480 on the 5B
Hardware, from the same pagesFrom 4GB in FP16 on the 2B
Length, from the same pagesUp to 161 frames on 1.5-5B

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.

1Working backwards from the output

Most releases here force a reader to pick a model and then discover the frame size. This family lets them start from the output: 1360x768 means the 1.5-5B, which means a 10GB floor.

That chain only closes because all three values name the same release. A family that stated one size and three memory figures would not support it.

2The newer release is wider, not just larger

1360x768 against 720x480 is not a uniform scale-up; the aspect ratio changes as well. A pipeline built around one shape will need reframing for the other.

The length figures move with them: up to 161 frames on the 1.5-5B, described as five or ten seconds, which is the other half of what changed between the releases.

3What neither size settles

Whether either can be pushed higher. The repository states these as what the models do, not as ceilings, and no memory figure is published for any size above them.

The register keeps the two sizes with their releases and adds nothing. A reader wanting a third size has no published figure to plan against.

  • Length and resolution
    CogVideoX1.5-5B states 1360x768 and up to 161 frames, described as 5 or 10 secondsper modelTHUDM, CogVideo repository / recorded 2026-09-12
  • Hardware
    From 4GB in FP16 for the 2B model, from 5GB in BF16 for the 5B, and from 10GB for CogVideoX1.5-5Bthe lowest figures the repository statesTHUDM, CogVideo repository / recorded 2026-09-12

4Sources

Output sizes above appear on github.com/THUDM, checked 2026-09-12. The full entry is at this family's page; what counts as a stated size is on resolution. Nearby: 480p, for now, 720 x 1280, in full.