Text to Image

Definition

Text-to-image is a generative AI task where models create visual images based on natural language prompts.

Purpose

The purpose is to enable creative design, art generation, and visualization from text.

Importance

• Expands human creativity and productivity.
• Raises copyright and misinformation concerns.
• Requires safeguards for harmful prompts.
• Related to diffusion models and GANs.

How It Works

1. Train model on paired text-image datasets.
2. Encode text into embeddings.
3. Map text embeddings to image representations.
4. Generate images using diffusion or GAN techniques.
5. Refine with user prompts or constraints.

Examples (Real World)

• DALL·E (OpenAI): generates creative images from text.
• Stable Diffusion: open-source image generation model.
• MidJourney: AI-powered art generation.

References / Further Reading

• Ramesh et al. “Zero-Shot Text-to-Image Generation.” OpenAI.
• Stable Diffusion Model Card — Stability AI.
• IEEE Computer Graphics and Applications: Generative AI in Imaging.