Summary:

Synthetic media startup Stability AI has introduced its first generative AI foundation model, called Stable Video Diffusion. This open-source model can generate original videos from text prompts and is available for research preview. Stability AI plans to expand its functionalities similar to its image generator, Stable Diffusion. The company claims its models outperform some of the top proprietary models in external assessments.

Introduction:

Stability AI has released Stable Video Diffusion, a generative AI foundation model that can create videos from text prompts. The model is available as an open-source research preview. Stability AI intends to build an ecosystem of extended functionalities, similar to its image generator, Stable Diffusion. The company asserts that its models have performed better than leading proprietary models in external assessments.

Main Points:

– Stable Video Diffusion comes in two models and can generate short videos ranging from 14 to 25 frames, with adjustable frame rates of 3 to 30 frames per second.
– The model can be fine-tuned for specialized applications, including multi-view 3D model spinning.
– Stability AI plans to expand the functionalities of Stable Video Diffusion, similar to its successful image generator, Stable Diffusion.
– The company emphasizes that the model was trained on publicly available videos for research purposes, addressing concerns about copyright infringement.
– Stability AI compares its model to text-to-video platforms Runway and Pika Labs, stating that it surpasses leading closed models in user preference studies.
– The company will soon provide access to a web interface showcasing text-to-video use cases in various industries, such as advertising, education, and entertainment.
– However, the current version lacks text input, photorealism, and camera motion options beyond panning.
– Stability AI clarifies that the model is not intended for real-world or commercial applications at this stage, with safety and quality refinements planned before full release.

Conclusion:

Stability AI has introduced Stable Video Diffusion, a generative AI foundation model capable of creating videos from text prompts. The company claims that its model outperforms leading proprietary models, based on external assessments. Stability AI intends to expand the functionalities of this model and build an ecosystem of extended features. Although the current version has limitations, such as the absence of text input and photorealism, Stability AI plans to refine the model further before commercial release.