← Back to the A-Z glossary
AI glossary · P2

Diffusion model

A diffusion model generates an image by starting from noise, then gradually transforming it.

Modality 4 min read Updated 2026-05-22
— Definition

Diffusion model, in plain words

A diffusion model generates an image by starting from noise, then gradually transforming it.

Present diffusion as text-guided denoising, not creative magic.

A concrete example

Midjourney starts from random noise and 'removes the noise' step by step to reveal the image your text describes. That's diffusion.

Why it matters

It's the tech behind Midjourney, DALL-E, Stable Diffusion, and Sora. Understand diffusion and you can read half the image news.

You see it in image and illustration generation tools.

Don't mix it up with

Multimodal: A multimodal model accepts several types of input, like text, images, audio, or video.

Text to image: Text to image generates an image from a written prompt.

Common mistakes

  • Mixing up diffusion (generating) and vision (reading).
  • Thinking diffusion = image only (there's video and audio diffusion too).
  • Forgetting diffusion is slow compared to generating text.

Quick checklist

  • First I check whether the word names a concept, a tool, a risk, or a metric.
  • I tie it to a concrete case: Midjourney starts from random noise and 'removes the noise' step by step to reveal the image your text describes. That's diffusion.
  • I keep the main trap in mind: Mixing up diffusion (generating) and vision (reading).

Quick questions

What is Diffusion model in AI?

A diffusion model generates an image by starting from noise, then gradually transforming it.

Where will I run into Diffusion model?

You see it in image and illustration generation tools.

Which word should I read next?

Start with Multimodal, Text to image, Video generation.

Want to keep going in order?

Head back to the full glossary, search a word, then open only the pages that deserve more than a short definition.

Open the AI glossary