The approach behind most image generators. It is trained by taking real images and progressively adding random noise until nothing recognisable is left, then learning to reverse that process. To generate a new image, it starts from pure noise and repeatedly removes a little of it, guided by your text description, until a picture emerges.
Technical view
The model learns to predict the noise component at each step of a denoising schedule. Text conditioning is usually supplied through cross-attention to an embedding of the prompt. Latent diffusion performs the process in a compressed representation rather than pixel space, cutting compute cost substantially, which is what made these models practical to run widely. Sampling requires many sequential denoising steps, though distilled variants reduce this considerably.
Related
Part of AI Terminology Index, section: Other Kinds of Neural Network.