Stability AI's open-source Stable Diffusion 3, still in preview, has made significant advancements in AI image generation, including improved text prompt understanding, image quality, and adherence to complex prompts. Comparisons with DALL-E 3 and Midjourney suggest that Stable Diffusion 3 may be the most capable AI image generator to date. The open-source approach of Stability AI continues to compete with paid-for models, and the AI arms race among big players shows no signs of slowing down, indicating ongoing advancements in AI image generation technology.
Stability AI introduces Stable Diffusion 3, an improved AI image generator that claims to produce higher quality images and handle text prompts more accurately. The company aims to prevent misuse by collaborating with experts and the community before the model's public release, following concerns raised by OpenAI's Sora model. Stable Diffusion 3 is not widely available yet, but interested individuals can join a waitlist for updates.
Stability AI has announced Stable Diffusion 3, a next-generation open-weights image-synthesis model that reportedly generates detailed, multi-subject images with improved quality and accuracy in text generation. The model range in size from 800 million to 8 billion parameters, accommodating different devices, and utilizes a new diffusion transformer architecture and flow matching technique to efficiently scale and produce high-quality images. While not widely available yet, Stability plans to make the model's weights free to download and run locally after testing.
Stable Diffusion 3 (SD3) is the latest image-generating AI model from Stability AI, aiming to compete with recent offerings from OpenAI and Google. SD3 is based on a new architecture and utilizes an updated "diffusion transformer" technique, with a parameter range from 800 million to 8 billion. It promises multimodal understanding and video input/generation capabilities, emphasizing safety measures to prevent misuse. The company is also upgrading its tooling to make it more accessible, and further details on the technical aspects are expected soon.
