A prompt is an instruction, question, outline, or other text input that guides a language or image model toward a desired response. Think of it as giving the model a creative direction: the clearer the direction, the more useful the starting result can be. section 12.3 section 11.44
Prompts help turn generative models into practical assistants. An image model uses a text prompt to guide reverse diffusion from random noise toward a clean image, while a language model uses a prompt to produce a first draft that a person can refine. section 12.23 section 11.44
From direction to draft
A user might provide an outline and ask for an article, marketing slogan, business report, product description, practice question, or lesson summary. For visual work, a complex text prompt can describe a scene; the image model uses that description while iteratively removing predicted noise until an image appears. section 11.44 section 12.3
Figure 12.7 A workflow for training a diffusion model
Follow the diffusion workflow to see where a user’s text prompt guides image generation. section 12.3
