From the text to the image: This is how Stable diffusion works

The underlying process, diffusion, is the current gold standard to create high -resolution images using artificial intelligence. It generates raiding pixel values raided sharp images. The content of this also happens by the image elements that were included in the model’s training material. In combination with an additional model that correlates text and image content, systems such as stable diffusion are created that create images with definable content instead of random.

The image quality that can be reached with diffusion requires a lot of computing effort, but is significantly higher than that of other procedures. So there are the early image generators from Deepmind and Nvidia, Biggan and Gaugangenerative neural networks (GAN) with an equally high quality with comparatively low computing effort. However, these are associated with a risk for the manufacturers, because they can collapse suddenly when they are training the underlying models.

That was the reading sample of our Heise Plus article “From the text to the picture: This is how Stable diffusion works”. With a Heise Plus subscription you can read the whole article.