Generative artificial intelligence continues to expand. This is demonstrated by the boom generated by natural language models such as GPT-3, GPT-4 and LaMDA, among others.
However, they are closed-source computational models. That is, their source code is not freely distributed and those who are in charge of their development often charge user licenses for using them. This is not the case with Stable Diffusion, an open source artificial intelligence based on diffusion models.
Diffusion model is a concept within artificial intelligence derived from the diffusion mechanism of thermodynamic systems that explains how particles tend to move from areas of high concentration to those of lower concentration to seek equilibrium.
Under this concept, the input data in a diffusion model are processed by advanced mathematical algorithms that include the addition of Gaussian noise iteratively until a diffuse representation is achieved that is consistent with an equilibrium situation. This new structure, basically composed of noise, however, has essential information for the subsequent operations of the model.
Unlike a thermodynamic system in which any diffusion process is irreversible, in the case of data diffusion based on a diffusion model, the rebuilding process from nothing is reversible but after training and machine learning. which also includes natural language comprehension.
Its most outstanding application is the generation of realistic images from text instructions or prompts that describe the scene to be recreated. And it has many representatives but the most prominent is the aforementioned Stable Diffusion.
Stable Diffusion is an image-generating artificial intelligence that works through a desktop program and additional components that require installation. However, a version of Stable Diffusion to be executed from the browser can be found on the Hugging Face web platform.
Here we leave you the address to your section in Hugging Face:
https://huggingface.co/spaces/stabilityai/stable-diffusion
Image by Dorota Kudyba from Pixabay