Stability AI unveils SDXL Turbo, a new AI-powered text-to-image mode with real-time output

Stability AI has launched Stable Diffusion XL Turbo, a novel text-to-image mode that leverages Adversarial Diffusion Distillation (ADD), a unique distillation technique. This approach enables the model to generate real-time text-to-image outputs maintaining high sampling fidelity, synthesizing image outputs in just a single step.

SDXL Turbo, by integrating ADD, inherits many benefits common to Generative Adversarial Networks (GANs), such as single-step image outputs. It also averts the artifacts and blurriness typically seen in other distillation techniques, which is a significant advantage.

In tests against other top-tier diffusion models, SDXL Turbo consistently outperformed in terms of image output quality, as judged by human evaluators. Furthermore, it required significantly fewer inference steps to achieve these results.

Stability AI has made SDXL Turbo available for trial on its image editing platform, Clipdrop. A beta demonstration of the real-time text-to-image generation capabilities of this new mode is also accessible for users to experience.