Stability AI releases SDXL 0.9 to enhance image generation technology.

2023-06-25

Stability AI, headquartered in London, has released SDXL 0.9 (Stable Diffusion XL), the latest addition to its suite of products, including Stable Diffusion. SDXL 0.9 is currently accessible through ClipDrop, with an API version set to be launched in mid-July following its beta release in April. SDXL 0.9 is capable of running on modern consumer-grade GPUs and can generate hyper-realistic creations in the fields of movies, television, music, teaching videos, design, and industrial applications, positioning SDXL at the forefront of AI image realization. The SDXL series encompasses various functionalities, not limited to basic text prompts, but also including image-to-image prompts (obtaining variants of an image), inpainting (reconstructing missing parts in an image), and extrapolation (creating seamless extensions of existing images). SDXL 0.9 boasts one of the largest parameter counts in open-source image models, with a base model consisting of 3.5 billion parameters (rather than 3.1 billion) and a model ensemble pipeline with 6.6 billion parameters. The final output is generated by running the input through two models and aggregating the results. The second-stage model of the pipeline introduces finer details to the output generated by the first stage. It is based on two CLIP models, including one of the largest OpenCLIP models trained to date (OpenCLIP-ViT-G/14), offering increased depth and 1024×1024 resolution images. In March of this year, Stability AI acquired Init ML, the parent company of Clipdrop. The new SDXL 0.9 can run on modern consumer-grade GPUs with Windows 10 or 11 or Linux operating systems and requires 16GB of memory. Additionally, Nvidia GeForce RTX 20 graphics cards (or equivalent higher standards) with at least 8GB VRAM are necessary. Linux users can also utilize AMD cards compatible with 16GB VRAM. A range of new products is being introduced. SDXL 0.9 will be used for research purposes for a limited time to collect feedback and refine the models before the official public release. The code for running it will be publicly available on GitHub. In April, Stability AI launched a software package called StableLM, which includes a set of large-scale language models also named StableLM. These models are currently in the "alpha" stage and can be found on GitHub and Hugging Face. Furthermore, Stability AI has released a new software development kit (SDK) that can be used in conjunction with Stable Diffusion 2.0 and Stable Diffusion XL. This allows users to control the software's output by adjusting various parameters, including style presets, rate, frames per second (FPS), color, 3D depth, and post-processing effects. To address the threat of copyright infringement, Stability AI has faced collective lawsuits from artists since January, accusing the company of illegally using their artwork for training models and creating new images, jeopardizing their professions. Getty Images has also filed a copyright infringement lawsuit against Stability AI. During a public interview at the Bloomberg Technology Summit in San Francisco, Stability AI CEO Emad Mostaque acknowledged the concerns raised by AI-generated realistic deepfake images. Mostaque revealed that the company did indeed possess "realistic models" earlier this year but chose not to release them due to time constraints. He emphasized the need to incorporate features such as watermarks to establish standards that enable tracking and reasonable utilization of AI-generated content. Source: https://analyticsindiamag.com/stability-ai-unveils-sdxl-0-9-to-enhance-image-generation/

RECENT AI NEWS

RECENT AI TOOLS