Stability AI has launched Stable Diffusion 3.5, a powerful suite of image generation models designed for both commercial and non-commercial use. Featuring enhanced customizability, efficiency, and performance, these models aim to empower creators and researchers across various fields.

On October 29th, Stability AI announced the release of Stable Diffusion 3.5, marking a significant advancement in their suite of image generation tools. This release includes multiple model variants—Stable Diffusion 3.5 Large, Stable Diffusion 3.5 Large Turbo, and the newly introduced Stable Diffusion 3.5 Medium—each designed to cater to a wide range of users from hobbyists to enterprises.

The Stable Diffusion 3.5 models are characterized by their high customizability and efficiency:

• Stable Diffusion 3.5 Large: This model boasts 8.1 billion parameters, offering superior image quality and prompt adherence, making it ideal for professional applications at a resolution of 1 megapixel.
• Stable Diffusion 3.5 Large Turbo: A distilled version that generates high-quality images quickly, completing tasks in just four steps while maintaining exceptional prompt adherence.
• Stable Diffusion 3.5 Medium: This model is optimized for running on consumer hardware, requiring only 9.9 GB of VRAM to unlock its full potential.

The development of Stable Diffusion 3.5 focused on enhancing customizability and performance while ensuring accessibility for users with standard consumer hardware. The integration of Query-Key Normalization within the transformer blocks has stabilized the training process and simplified further fine-tuning efforts.

However, this flexibility comes with trade-offs; users may experience greater variation in outputs from similar prompts due to the intentional design aimed at preserving a diverse knowledge base and style variety.

The new models excel in several key areas:

• Customizability: Users can easily fine-tune the models to suit their specific creative needs or build applications based on customized workflows.
• Diverse Outputs: The models are capable of generating images that represent a wide array of demographics without extensive prompting.
• Versatile Styles: They can produce various visual styles, including photography, painting, line art, and more.
• Efficient Performance: Particularly the Medium and Large Turbo models are optimized for use on consumer-grade hardware without heavy resource demands.

The models are released under a permissive Stability AI Community License that allows free use for both commercial (up to $1 million in annual revenue) and non-commercial purposes. This license structure encourages creativity and innovation while ensuring users retain ownership of their generated media without restrictive licensing implications.

Stability AI emphasizes responsible AI practices throughout the development process of Stable Diffusion 3.5. They have implemented measures to prevent misuse by bad actors, reflecting their commitment to safety in AI deployment.

Looking ahead, Stability AI plans to introduce ControlNets soon, which will provide advanced control features tailored for various professional applications. The company is eager to receive feedback from users as they explore the capabilities of Stable Diffusion 3.5.

The launch of Stable Diffusion 3.5 represents a significant milestone in the evolution of image generation technologies. By combining advanced capabilities with accessibility and user-friendly licensing terms, Stability AI empowers creators across diverse fields to explore new artistic possibilities and enhance their workflows.

To explore ways to integrate this model in you workflow or read full article go here. You can access the models through Hugging Face or through the following platforms:

• Stability AI API
• Replicate
• Fireworks AI
• DeepInfra
• ComfyUI