Link: Stability AI releases Stable Diffusion 3.5 Large, an 8B-parameter model that makes 1MP images, and 3.5 Large Turbo, and plans to launch 3.5 Medium on October 29 (Kyle Wiggers/TechCrunch)

AI startup Stability AI has launched its new Stable Diffusion 3.5 series, promising more customizable, versatile, and performant image-generation models. This series includes three models, with the most advanced offering resolutions up to 1 megapixel.

Stable Diffusion 3.5 Large Turbo provides quicker image generation at a slightly reduced quality, while Stable Diffusion 3.5 Medium is designed for use on devices like smartphones and laptops.

The new models are designed to produce more diverse images, enhancing representation across different skin tones and features without extensive prompting. Stability’s chief technology officer highlighted their training process, which involves multiple prompt versions to ensure a variety of image concepts.

Despite the advancements, Stability warns that the new models may still encounter issues with prompt accuracy due to engineering compromises. However, they are said to be more robust in generating diverse styles compared to previous models.

The licensing terms remain largely unchanged; the models are free for non-commercial use and for businesses generating less than $1 million annually. Stability has updated its commercial use terms, allowing more freedom following backlash over restrictive fine-tuning policies.

Stability trains its models on a mix of public and synthetic data, while facing growing legal scrutiny over potential copyright claims. They leave legal defenses to their users but have made provisions for data owners to opt out of their datasets. #

--

Yoooo, this is a quick note on a link that made me go, WTF? Find all past links here.