Stable Diffusion 3.5 Released: Generating More Realistic Images than Ever

Stability AI, a developer of artificial intelligence technology, has recently launched the Stable Diffusion 3.5 series of AI image generation models. Compared to the earlier SD 3, the latest SD 3.5 offers improved realism, faster adherence to prompts, and enhanced text rendering capabilities.

SD 3.5 also comes in different parameter versions, including Large-8B, Large Turbo-8B, and Medium-2.6B. These models are customizable and adjustable, for example, targeted adjustments can be made for local use on consumer-grade hardware.

Upon a closer look, the improvements in SD 3.5 might seem modest. Stability AI has devoted much of its recent efforts to optimizing the functionality of SD 3, as it did not meet the expectations of Stability AI and the community upon its release.

In an official blog post, Stability AI stated:

"All models introduced offer customizability, high performance, and diverse outputs. SD 3.5 is our most powerful model to date, reflecting our commitment to providing creators with widely accessible and cutting-edge tools."

"Our analysis indicates that SD 3.5 Large leads the market in prompt adherence and competes with larger models in image quality."

"The SD 3.5 Turbo offers the fastest inference times at its parameter size while remaining highly competitive in image quality and prompt adherence; SD 3.5 Medium outperforms other medium-sized models, balancing prompt adherence and image quality, making it the preferred choice for efficiency and quality."

Licensing for the Models:

The SD 3.5 series models are available for non-commercial use, such as personal use, scientific research, and other non-commercial purposes. For businesses, the licensing is based on annual revenue: companies with annual revenues exceeding $below$1 million can use the models for free.

For questions regarding the Stable Diffusion license, please refer to the detailed explanation by Stability AI: https://stability.ai/license