Stable Video Diffusion

Stable Video Diffusion turns a still image into a short video clip for artwork, design and video generation research. Developers and researchers can run it on their own hardware, but text prompts can't direct the output.

• Image animation: Generate a short clip that follows the supplied picture, with outputs lasting up to four seconds.
• Model variants: SVD generates 14 frames, while SVD-XT generates 25 frames at 576×1024 resolution.
• Research tools: The downloadable weights work with Stability AI's Python generative-models codebase, which provides inference and training support.

Motion can be absent or limited to a slow camera pan. Results aren't fully photorealistic, faces and people can appear inaccurate, and the model can't render legible text. It isn't intended to create factual representations of people or events.

Generation can use GPU hardware. Optimizations allow use on cards with less VRAM by trading off quality, memory use and speed.

The repository code uses the MIT license. The model weights have separate licensing conditions. SVD-XT 1.1 uses the Stable Video Diffusion Community License, which permits research and noncommercial use and requires registration and eligibility for commercial use.

This listing covers SVD, SVD-XT and the SVD-XT 1.1 checkpoint. Reference code uses MIT, while weights have separate Stable Video Diffusion Community license terms. Other models in the shared generative-models repository are separate releases with their own terms.