Between Midjourney and Dall-E, image generation models continue to improve and Stable Diffusion is no exception. On February 22, 2024, Stability.ai announced version 3 of its artificial intelligence which promises “improved performance†.
Stable Diffusion 3 improves the quality of generated images
The start-up announces that with Stable Diffusion 3, “multi-topic prompts, image quality and writing capabilities†have been significantly improved. Artificial intelligence uses a different architecture from previous versions. Stability.ai says its artificial intelligence models range from 800 million to 8 billion parameters, to cover all user preferences. Stable Diffusion therefore aims to meet a multitude of needs.
On X, Stability.ai shared several images generated via Stable Diffusion 3 which show improved generation quality, particularly in terms of details. Artificial intelligence can even write sentences, a task usually difficult for image generation models.
For the moment, Stable Diffusion 3 is being tested. Simply register on the waiting list then wait for an email allowing you to join the dedicated Discord server. The start-up will collect initial comments and feedback from users to refine its image generation models.
The start-up emphasizes security
The trial launch of Stable Diffusion 3 is accompanied by a company update on security. The start-up promises to avoid deepfakes and specifies that efforts to prevent any misuse of its artificial intelligence are being deployed –from training and throughout testing, evaluation and launch†.
For several months, image-generating artificial intelligence has been used for malicious purposes, to spread false information or harm people with the use of deepfakes. Companies specializing in AI like Stability.ai are committed to fighting against these misuses while governments take the necessary measures.