Stability AI launched their most capable text-to-image generator model, the Stable Diffusion 3. The company says the new model comes with improved performance in multi-subject prompts and image quality. The Stable Diffusion 3 also comes with improved spelling abilities. 

Stable Diffusion 3 is going to give tough competition to DALL-E 3 and Midjourney. It is going to be much better at generating images in great quality with multiple subjects, and the words in the generated images will have much better spellings as compared to previous stable diffusion models.

Technical Details of Stable Diffusion 3

It is based on a diffusion transformer architecture, which is similar to OpenAI’s new text-to-video model, Sora. It also uses a new flow matching technique, which further helps to generate high-quality images.

The model size ranges from 800 million to 8 billion parameters. This helps to run different versions of the model on various hardware.

Stablility AI’s founder and CEO, Emad Mostaque, released more technical details on social media platform X.

According to him, the model can take multimodal inputs too, so users will also be able to provide image and video inputs. Emad says stable diffusion 3 will be released as an open model and will launch with a full ecosystem of tools.

The technical report for Stable Diffusion 3 will be released soon, which will contain all the technical details of the model.

Stability AI on Security with Stable Diffusion 3

Security is a big concern with AI models. With text-to-image models, it gets larger, so Stability AI is doing everything to prevent the misuse of the Stable Diffusion 3 model. The preview of the model will also play an important role in safeguarding the model before its final public release.

Stability AI, mentioning security, wrote in their blog post, “Safety starts when we begin training our model and continues throughout the testing, evaluation, and deployment. In preparation for this early preview, we’ve introduced numerous safeguards. By continually collaborating with researchers, experts, and our community, we expect to innovate further with integrity as we approach the model’s public release. Our commitment to ensuring generative AI is open, safe, and universally accessible remains steadfast.”

How to Access Stable Diffusion 3 

The model is not publicly released, but Stability AI has opened a preview waitlist for users who want to try the model. Users can sign up and join the waitlist from here.

Conclusion

Stable Diffusion 3 is going to give tough competition to competitors like DALL E-3 and Midjourney. With its enhanced performance, it can easily beat its competitors. The most impressive thing is its enhanced spelling abilities, as most image generator AI models struggle with word spellings in the generated images. However, the model is still in preview, and we can only see the example images shared by some users on social media, so it will be wise to judge after the model gets publicly released.

Read More

Humane’s AI Pin Shipping Gets Delayed: Early Adopters Get 3 Months Free Subscription

Google pauses Gemini’s AI Image Generator after it Receives severe Criticism from Users