NodeValid

Published video-model weights, licences and hardware notes

Step-Video-T2V stays above 72GB in every setting

The repository states 78.55 GB peak at 768px768px204f, 77.64 GB at 544px992px204f and 72.48 GB at 544px992px136f, all at batch size one. Every published configuration sits above seventy gigabytes. As of 2026-09-22.

Step-Video-T2V in the hardware column, as publishedThree settings and a six gigabyte spread means the configuration barely moves the requirement, which is unusual and useful to know before planning.The memory statement as published78.55 GB peak at768px768px204f77.64 GB and 72.48 GB at two smaller settings768px768px and 544px992pxStated as the memory table's configurationsUp to 204 framesFrames, with no frame rate attachedCutting the setting barely helps
Fig. 1 Three configurations, a narrow spread, and a floor that stays above seventy gigabytes.
Step-Video-T2V in the hardware column, with the columns that qualify it. Read from github.com/stepfun-ai on 2026-09-22.
ColumnWhat the pages state
Hardware, in the vendor's own words78.55 GB peak at 768px768px204f
The detail published beside it77.64 GB and 72.48 GB at two smaller settings
Resolution, from the same pages768px768px and 544px992px
Length, from the same pagesUp to 204 frames

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.

1A narrow spread is a planning fact

Dropping from 204 frames at 768px768px to 136 frames at 544px992px saves about six gigabytes out of seventy-eight. For a reader hoping to fit the model on a smaller card by shortening the clip, that is the answer.

Elsewhere in the register the same kind of change moves a figure by a factor of three. Here the weights dominate so completely that the settings hardly matter.

2Batch size one is stated, which is rare

Most memory figures in the register leave batch size implicit. Stating it removes the commonest hidden variable, and it also means the figures are the lowest the vendor could honestly publish.

Anyone planning throughput rather than feasibility should read the numbers as a per-stream cost, with no batching headroom included.

3Frames without a frame rate

The ceiling is given as up to 204 frames and no frame rate is attached anywhere, so the clip length in seconds is not something this row settles.

That is why the register keeps length in the unit the vendor chose. Converting 204 frames into seconds would require a frame rate the repository never published.

4Sources

Every figure above is the vendor's own, taken from github.com/stepfun-ai on 2026-09-22 with nothing measured here. The rest of the release is on its family entry, the whole column is on hardware, and all memory figures together are on requirements. Nearby: One card or four, A stopwatch, not a floor.