SVD states 576x1024 as a property of training
The card states that the model was trained to generate 25 frames at resolution 576x1024, given a context frame of the same size. Output and input are the same shape, and the size is described as a training property. As of 2026-09-22.
| Column | What the pages state |
|---|---|
| Resolution, in the vendor's own words | 576x1024 |
| The detail published beside it | What the model was trained to generate |
| Hardware, from the same pages | About 180s on an A100 80GB card |
| Length, from the same pages | 25 frames, 4sec or less |
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.
1The input side is the part that catches people
A context frame of the same size means source images have to be prepared at 576x1024 before anything else happens. That is a pipeline requirement, not just an output description.
Releases that generate from text alone have no such constraint, so this row is a different kind of value from most of the column.
2Trained to generate is narrower than supports
The phrasing describes what the training did, not what the code accepts. Other sizes may run and were not what the weights learned, which is a real distinction for output quality.
Another release in the register uses the word supports for its sizes. The register keeps both wordings rather than normalising them into one column heading.
3Twenty-five frames, and a candid note about it
The same card describes the generated videos as rather short, at four seconds or less. That is written as a limitation rather than as a specification, and it is unusually frank for a model card.
So the output is fully described in both dimensions and one of them comes with an apology. The register records the size, the frame count and the wording.
- Resolution and framesTrained to generate 25 frames at resolution 576x1024, given a context frame of the same sizeas a property of the training
- LengthThe generated videos are rather short, at 4sec or lesswritten as a limitation rather than a ceiling
4Sources
Output sizes above appear on huggingface.co/stabilityai, checked 2026-09-22. The full entry is at this family's page; what counts as a stated size is on resolution. Nearby: No frame size given, One size per checkpoint.