Back to Gap Map Model components / Image, video, 3D & music generation

Stable Video Diffusion

Stability AI
open weights / Overall score: 1.8

Stable Video Diffusion is Stability AI's latent diffusion model that turns a single still image into a short video clip. The original SVD generates 14 frames and SVD-XT 25 frames at 576x1024; SVD-XT 1.1, released in February 2024, is a fine-tune of SVD-XT for more consistent output.

Stable Video 3D (SV3D) and Stable Video 4D (SV4D, SV4D 2.0 of May 2025) are built on SVD but generate orbital and multi-view video of an object rather than a clip from an image, so they are not part of this entry; the category already treats Stability AI's 3D models as separate products. The generative-models repository is shared with Stable Diffusion and publishes SVD inference only. Dormant: no SVD release since February 2024, but the two 2023 checkpoints still draw about 330K downloads a month.

Openness

3 medium confidence
3.0
weights
open (SVD and SVD-XT download without a gate; SVD-XT 1.1 behind an automatically granted click-through)
data
described (the SVD paper and cards describe the curated training video set and its filtering; no dataset is released)
code
partial (Stability-AI/generative-models publishes SVD inference configs and sampling scripts under MIT; its example training configs cover ImageNet, CIFAR-10, MNIST and a CLIP-conditioned text-to-image model, not SVD)
license
Stability-AI-Community-License (the LICENSE.md in every SVD repository: free below USD 1M annual revenue, after which an enterprise license is required; the SVD-XT 1.1 click-through still asks for non-commercial or research use only)

Stable Video Diffusion is free to use, commercially included, for organizations under USD 1M in annual revenue, with a paid license above that; the newest checkpoint's download form still asks for non-commercial use. Stability publishes inference code only, and the training videos are described but not released. Stability-AI-Community-License allows commercial use only within a bound, so the license tier is use_bounded.

Adoption

3 high confidence
3.0

Adoption is measured as Hugging Face downloads across the SVD checkpoints, almost all of them the original SVD-XT and SVD rather than the newer 1.1.

Capability

1 medium confidence
1.0

Stable Video Diffusion generates only from an image and has no entry on either public video arena. It sits below Mochi 1, which places on the text-to-video board.

Verified 2026-09-27