Identifying Stable Diffusion XL 1.0 images from VAE artifacts

AI generated images have become so realistic that they are often mistaken for real photographs. The new SDXL 1.0 text-to-image generation model has introduced colorful artifacts around the edges of the generated images. These artifacts can be easily spotted by visually examining the images or by training a simple neural network. The PSNR values show that the 0.9 VAE in the SDXL model has higher quality compared to the 1.0 VAE. Detecting and marking AI generated content is an important topic in the industry.