Stability AI has announced the launch of their latest model, Stable Diffusion 3, which is the newest and most powerful version of their AI-powered image generation model.
The new model is being introduced in response to competing models from Google and OpenAI.
Stability AI has released several models over the past year, and these models showcase significant advancements and quality.
The release of SDXL in July led to a significant improvement in the base Stable Diffusion model, and the company is now looking to move forward.
Stable Diffusion 3 is based on a new architectural design and is compatible with a variety of devices. It has not been released by Stability AI yet, but you can join the waitlist.
The new model relies on the diffusion transformation technology, which was pioneered in 2022, improved in 2023, and is now scalable.
Additionally, the Flow Matching technology is used to enhance quality without the need for additional necessary resources for a specific task.
Imad Moshtaq, the founder and CEO of Stability AI, says that the new model can understand multimedia and add a video generation feature.
These features remain theoretical for now, but there is no technical barrier preventing their inclusion in future releases.
The new model, Stable Diffusion 3, aims to improve image quality and enhance image generation performance for various requirements.
Additionally, Stable Diffusion 3 provides a better printing method compared to previous versions of Stable Diffusion models, allowing for precise and consistent image generation.
Printing was previously a weak point for stable diffusion models, a point that competitors also worked on with recent versions like DALL-E 3, Ideogram, and Midjourney.
Stability AI is developing their new model, Stable Diffusion 3, in various sizes ranging from 800 million to 8 billion parameters, aiming to run it on a wide range of devices.
The established capabilities of AI in 3D image generation and video creation have evolved in recent months.
The company stated: “We produce open models that can be used anywhere and modified to suit any need, including a range of models of different sizes that support the development of our visual models for future generations.”
