"Diffusion" is the movement of things, like atoms, going from an area of high concentration to a region of lower concentration.

Artificial Intelligence (AI) Diffusion models are named after the diffusion process because of the way they work; they initially add random "noise". or random or unpredictable fluctuations in data. This creates less clear information, and then the model tries to recreate the prompt its requested. The noise gets spread in what the model needs to work on, and the model works to reverse that to create a realistic image. Think like a reverse of "diffusion".

This is how many famous programs like Midjourney on the internet create images. Diffusion models are a method to try to make AI create and imagine images.

Diffusion models can create images from text descriptions, fill in missing parts of an image, editing, and even create videos. Diffusion models can create beautiful, detailed, and realistic images with this process.

--

Transformer models are designed to understand the context of language. What makes them special is unlike other models such as CNN and RNN networks, they process data much faster and are more efficient.

The architecture they use tries to generate the next word in a sentence with a contextually relevant output. Every word is analyzed by the architecture and is interpreted through the AI model's overall knowledge found in its database.

Transformer models can be used in text generation, paraphrasing, and translating languages. They are also used in more complex things like, driver less cars! Transformer models, through understanding patterns, can be used to write entire articles in your style. Or even have it learn from thousands of different styles. You can teach it a set of rules and have it be virtual assistant to anyone who needs help understanding one of those rules, like from an FAQ.