StabilityAI today released the new Stable Diffusion 3.5 family of AI image models with improved realism, prompt adherence, and text rendering compared to SD3.

Like the first version of SD3, Stable Diffusion 3.5 is available in three sizes: Large (8B), Large Turbo (8B), and Medium (2.6B). All of these are customizable and tuned to work with consumer hardware.

In short, this is a big step toward making it possible for any user to create more realistic AI images, and StabilityAI acknowledged in a press release that the Stable Diffusion 3 Medium model launched in June “does not fully meet our standards and community expectations. not fully meet our standards and community expectations.”

The company added that “after listening to the community's valuable feedback, we took the time to further develop a version that advances our mission to transform visual media rather than a quick fix.”

Our AI editor Ryan Morrison has been testing SD3.5, and says it is a significant upgrade that rivals and potentially exceeds the capabilities of the recently released Flux 1.1 Pro.

According to Stability AI, the models included focus on customizability, efficient performance, and diverse outputs. A spokesperson explains, “Stable Diffusion 3.5 is our most powerful model to date and reflects our commitment to empowering creators with widely available, state-of-the-art tools.”

This means that images can be fine-tuned, models work “out-of-the-box” on consumer hardware, and the generated images feel more unique.

There is also a focus on new style choices, such as photography and painting. Hashtag prompts are incorporated to specify styles such as boho and fashion. It can also use highlighting within the prompts to guide the model in a particular direction.

“Additionally, our analysis shows that Stable Diffusion 3.5 Large leads the market in prompt compliance and rivals much larger models in image quality,” the press release explains.

“The Stable Diffusion 3.5 Turbo offers the fastest inference time for its size, while remaining highly competitive in both image quality and rapid fixation when compared to non-diffusion models of similar size.

“The Stable Diffusion 3.5 Medium outperforms other mid-size models, offering a balance between rapid fixation and image quality, making it the best choice for efficient, high-quality performance.”

The model is free for non-commercial use, including scientific research, and free for small businesses up to $1 million in revenue. Beyond that, an enterprise license is required.