Decoder. plain-English AI glossary

Image Generation

● Core

Creating new, realistic images from scratch or conditioned on a description, sketch, or example.

Think of it like

An artist drawing a picture from your verbal description.

Example

"A robot painting in Van Goghs style" generates a novel image that has never existed—no copyright, new synthesis.

How it actually works

Diffusion models (iterative denoising) are now dominant over GANs. Start with noise, iteratively denoise guided by text embedding or other conditioning. Training requires billions of image-text pairs. Inference takes 20-50 steps (seconds on GPU). Quality improves with compute; 4K outputs are now feasible.

For product teams

Opens creative use cases: design mockups, stock imagery, avatar generation, personalization.

For engineers

Text encoder converts prompt to embeddings; UNet denoising network reverses Gaussian noise 50 steps guided by embeddings; outputs image.

Diffusion Model — Based on.
Text-to-Image — Variant.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome