Stability AI launches Vincentian graph model SDXL0.9, with GPU requirements lowered to consumer-grade levels

IT Home News on June 25, Stability AI recently announced that its text generation image model SDXL 0.9 has been officially launched. This version features significant improvements to image content and compositional detail, and also supports running on consumer-grade GPUs.

It is reported that SDXL 0.9 has the largest number of parameters among all open source image models, with a base model of 3.5 billion parameters and an additional model of 6.6 billion parameters. Around these two models, the working principle of SDXL is to use the basic model to create rough details, and then use additional models to refine the generated pictures. If your friends at IT House have used Stable Diffusion, you should be able to notice this. A progressive work process.

▲ Picture source SDXL team

Stability AI says that two CLIP models are used in SDXL0.9, including OpenCLIP vitg /14, which is the largest OpenCLIP model to date. Using this model, Stable Diffusion is able to generate more realistic images with higher resolution and greater depth.

Stability AI also stated that the SDXL team will publish a research blog detailing the model specification and more parameter details of SDXL 0.9. It is expected that the model will usher in the official version 1.0 in July and will be open source on GitHub.

The above is the detailed content of Stability AI launches Vincentian graph model SDXL0.9, with GPU requirements lowered to consumer-grade levels. For more information, please follow other related articles on the PHP Chinese website!