Text-to-image generation creates images from text descriptions using models like DALL-E, Midjourney, and Stable Diffusion. How it works: Text encoder produces embedding, diffusion model conditioned on embedding generates image through iterative denoising. Components: Text encoder (CLIP, T5), diffusion U-Net, VAE for latent space (Stable Diffusion). Training: Pairs of images and captions, learn to denoise images conditioned on text. Inference: Start from random noise → iteratively denoise guided by text conditioning → decode to image (if latent diffusion). Key techniques: Classifier-free guidance (balance quality/diversity), cross-attention between text and image features. Major models: DALL-E 2/3 (OpenAI), Midjourney, Stable Diffusion (open source), Imagen (Google), Firefly (Adobe). Prompting: Detailed descriptions work better, style keywords, artist references, quality modifiers ("highly detailed", "4k"). Applications: Art creation, design prototyping, stock images, advertising, creative tools. Challenges: Text rendering, anatomy issues, copyright concerns, misuse potential. Safety: Content filters, watermarking, provenance tracking. Revolutionary technology for creative industries.