Stable Video 4D
Stable Video 4D generates novel views of a moving object from a single-view video. It builds on SVD and SV3D to coordinate camera viewpoints with time.
Model design and access
The documented workflow first uses SV3D to obtain orbital views from the initial frame. SV4D then combines these reference views with video frames. Longer sequences use generated anchors and interpolation. Its output is a time-and-view image matrix, not a rigged animated mesh.Stable Video 4D official model card
Practical use
Practical guidance. This release is useful for studying how an object changes across views and time. Review motion consistency separately from view consistency. A convincing single angle can hide errors elsewhere in the matrix, so inspect each view before using the sequence for reconstruction.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | SV4D |
| Output block | Five video frames across eight camera views.Stable Video 4D official model card |
| Resolution | 576 × 576 for each generated image.Stable Video 4D official model card |
| License | Stability AI Community License.Stable Video 4D official model card |
Related models
- Stable Video 3D
Stability AI · Stable Video
Stable Video 3D adapts Stable Video Diffusion to generate orbital views of an object from one image.
Launch date not confirmedDownloadable - Stable Video 4D 2.0
Stability AI · Stable Video
A video model that generates new views of moving objects with four-view and eight-view variants.
Launch date not confirmedDownloadable