2023-06-25 19:23:03
AI models can currently only be watched while learning, but in fast motion. Now Stability AI, which famously released Stable Diffusion with Runway and the CompVis Group of Ludwig Maximilian University in Munich, has released a new version of the text-to-image generator. “SDXL 0.9” is initially available on the market for research purposes and, according to the creators, is intended to “generate hyper-realistic creations for films, television, music and instructional videos” and is therefore one of the pioneers of AI images.
SDXL 0.9 should not only understand text prompts, but also image-to-image prompting (entering an image to get variations of that image), inpainting (reconstructing missing parts of an image), and outpainting (constructing a seamless extension of an existing one). picture) allow. Also noteworthy: The model has 6.6 billion parameters, making it one of the largest in its class, but it should still be able to run on a modern consumer GPU.
For the AI image model, it only needs Windows 10 or 11 or a Linux operating system with 16 GB RAM and an Nvidia GeForce RTX 20 graphics card (equivalent or higher standard) with at least 8 GB VRAM. Linux users can also use a compatible AMD card with 16 GB VRAM. The full version SDXL 1.0 should come in mid-July.
AI spending will double to $300 billion between 2023 and 2026
Controversial AI models
A special feature that Stability AI emphasizes: The AI model can finally draw realistic human hands. Up until now, many image generators have failed at precisely this, and in the past have often created hands with too many or too few fingers. This should now work better, even if some of the examples shown by Stability AI do not necessarily deserve the title “hyper-realistic”. But you can see that the AI is already producing very acceptable results.
As mentioned, SDXL 0.9 can initially only be used for research purposes and has limited access. Users of the Stability AI Clickdrop app as well as Stability AI API and DreamStudio customers will have access, and starting Monday, June 26th, there will also be the opportunity to access the code via Github and Hugging Face. However, the prerequisite is that it is not used for commercial purposes.
Stability AI around CEO Emad Mostaque is headquartered in London and has received around 100 million dollars from well-known investors such as Coatue, Lightspeed Venture Partners and O’Shaughnessy Ventures. It is also behind the quite controversial stable-diffusion model, which brought the startup lawsuits from artists and Getty Images. They accuse the British startup of having trained the AI models with copy-protected content and thus having committed copyright infringements. With StableLM, Stability AI also develops a large language model.
Midjourney, Dall-E & Imagen: The Unstoppable Rise of AI Images
1687733478
#image #generator #finally #draw #realistic #hands
