Stability AIStable Video

Stable Video Diffusion

Stable Video Diffusion turns a still image into a short video. The original image-conditioned model generates 14 frames and uses a decoder adapted for temporal consistency.

Model design and access

The card also provides a frame-wise decoder for comparison. It warns of very limited motion, imperfect faces and unreadable text. SVD is an image-conditioned video model; a still image is the starting condition rather than a complete motion plan.Stable Video Diffusion official model card

The XT checkpoint extends generation to 25 frames and was fine-tuned from the 14-frame model.Stable Video Diffusion XT model card

Stability ended its Stable Video Diffusion API on 24 July 2025. Self-hosted use remains available. This page lists downloadable weights as the available access route.Stability AI API release notes

Practical use

Practical guidance. These weights suit local motion experiments and research that needs a fixed earlier release. A strong starting image helps define appearance, but it does not establish a precise action. Inspect the whole clip for deformation and scene changes, especially when the input contains people or small text.

Specifications & access

Stable Video Diffusion specifications
DetailValue and conditions
AccessDownloadable
Model identifiersSVDSVD XTstable-video-diffusion-img2vid-xt
Base output14 frames at 1024 × 576.Stable Video Diffusion official model card
Current licensingCommunity and Enterprise Core Models include 14- and 25-frame versions.Stability AI Core Models: May 2026 list
Control limitNo direct text-prompt control.Stable Video Diffusion official model card

Related models

  • Stable Video 3D adapts Stable Video Diffusion to generate orbital views of an object from one image.

    Launch date not confirmedDownloadable
  • An image-to-video model that refines the 25-frame SVD XT checkpoint with fixed frame-rate and motion conditioning.

    Launch date not confirmedDownloadable