Stable Diffusion just got better and cheaper

An AI-generated image that shows a sunrise in a fantastical landscape.
An AI-generated image made with Stable Diffusion. Andrew Tarantola / Stable Diffusion

Stability AI, the startup behind popular image-generation engine Stable Diffusion, has announced an updated licensing agreement that expands the free tier and removes limits on the number of generated images. The company even announced some improvements to the quality of its current model.

Let’s tackle the license first, which came in a blog post on Monday where the company admitted that the “commercial license originally associated with SD3 caused some confusion and concern in the community.”

Now only Enterprise customers making more than $1 million in annual revenue and folks using the output images and videos in commercial applications will be required to pay for the Stable Diffusion software.

The “Stability AI Community License” drastically broadens the free uses of the company’s generative AI models, including the recently released (and poorly received) Stable Diffusion 3 Medium (SD3M) model.

“People and organizations that install and run Stability AI models on their own devices (directly or through installing free open-source packages that include them) for non-commercial uses can continue doing so for free,” the company wrote.

This includes hobbyists, researchers, developers, students and teachers, as well as “fine-tuners” who help improve and distribute the models. The new license freedoms apply to small businesses as well. Stability AI is fine with them leveraging either an SD model directly or a derivative (say, their own specialized and fine-tuned version of SD3), just so long as the company’s total annual revenue falls under $1 million (or their local currency equivalent). Companies that exceed that financial limits will need to contact Stability directly and obtain an Enterprise license.

What’s more, the new license scheme has no set limits on the number of generated pieces a user or organization can create, whether it’s images, video, or audio content.

“As long as you don’t use it for activities that are illegal, or clearly violate our license or AUP,” the company wrote, “Stability AI will never ask you to delete resulting images, fine-tunes, or other derived products — even if you never pay Stability AI.”

Stability also announced that it is tweaking its SD3M model to improve its performance after hearing boisterous criticism from the developer community. The company notes that the new model outperformed its predecessor, the SDXL model (now known as AuraDiffusion), during testing and development.

However, critical flaws were found when users asked the newer AI to create novel avatar poses or used words not commonly found in the model’s training data. “SD3 Medium is still a work in progress,” the company wrote. “We aim to release a much-improved version in the coming weeks.”