Stability AI announces Stable Diffusion 3, a new generation of generative AI for images and video
Stability AI has announced the early access of Stable Diffusion 3, the latest version of its legendary image generation model, and, as expected, video. This model is based on an updated architecture and can run on a variety of hardware, although it still requires a powerful GPU.
Stable Diffusion 3 uses an advanced “diffusion transformer” technique that was developed in 2022 and updated in 2023. It also uses a “flow matching” technique to help improve image quality.
The company has released different versions of the model, from 800 million to 8 billion parameters, to ensure compatibility with different equipment. Unlike OpenAI and Google, which have also recently announced their new language models, Stable Diffusion is not limited to APIs, making it more accessible for use. Specifically, AI models can be installed on your computer and do not depend on subscriptions, networks, and other restrictions. The only limitation is the power of your PC.
One of the key features of SD3 is its ability to provide multimodal video understanding and generation. Although these capabilities are still theoretical, there seems to be no technical obstacles to implementing them in future updates.
Stable Diffusion aims to become a universal tool in the field of generative AI by lowering the entry threshold for users. The company also announced that it has taken security measures to prevent the model from being misused, including safeguards at the training, testing, and implementation stages of the model.
The details of these security measures have not yet been disclosed, but they will be clarified before the public release of the model. The community is looking forward to further news and innovations related to this development.
