SDXL Turbo generates images while you're typing

Stability AI is releasing a new text-to-image model called SDXL Turbo that can generate images in real-time from text descriptions. SDXL Turbo uses a novel distillation technique to enable single-step high-fidelity image generation, reducing the number of steps required compared to previous models.

What's going on here?

Stability AI has released a new text-to-image model named SDXL Turbo that can generate images from text in real-time.

What does that mean?

SDXL Turbo uses a new technique called Adversarial Diffusion Distillation (ADD) to produce high-quality images in just one processing step, unlike other models that require multiple steps.

It combines the speed of GAN AI models with the detail quality of diffusion models. So SDXL Turbo generates images incredibly fast yet with intricate fidelity—the best of both worlds. Where top existing models need 50 steps to create images, SDXL Turbo only needs 4 and makes images just as good or better. On fast hardware, it generates a 512 x 512 picture in barely over 200 milliseconds!

Why should I care?

In blind testing against leading models, humans consistently picked SDXL Turbo's images as most closely matching text prompts—showing creativity and clarity. This new combo of speed + quality opens possibilities for using AI image generation creatively.

You can try a beta of SDXL Turbo on Stability AI's Clipdrop platform to create visuals on-demand. While not commercially usable yet, SDXL Turbo means exciting progress in text-to-image AI.