# Diffusion Model

> A diffusion model generates an image by starting from random noise and removing it step by step, guided by a prompt, until a coherent picture remains.

A diffusion model is the technique behind most image, video, and audio generators: it learns to reverse a noising process, so it can start from static and denoise it into an image that matches your prompt. Step count, guidance, and seed are the knobs designers actually touch.

**A diffusion model generates an image by reversing a noising process: it starts from random static and removes noise step by step, guided by your prompt, until a coherent picture remains.** Training teaches it the reverse of a simple procedure — take real images, add noise until they are unrecognizable — so at generation time it can walk that path backwards from pure noise.

Most systems run this in a compressed latent space rather than on full-resolution pixels, which is why an image can be produced in seconds on one GPU. Three settings surface to users: step count, which trades time for refinement; guidance strength, which controls how literally the model follows the prompt at the cost of variety; and the seed, which fixes the starting noise so a result is reproducible. Those three explain most of the difference between a tool that feels controllable and one that feels like a slot machine.

Diffusion is the technique; [text-to-image](/glossary/text-to-image) is the task it is best known for, though the same approach drives video, audio, and inpainting. Designers meet these models through a canvas app, and developers meet them through an inference cloud: [fal](/tools/fal) specializes in generative media billed per output or per compute, and [Replicate](/tools/replicate) runs a broad open-model catalog billed per second or per output. Which model families are worth reaching for is covered in [the best AI image generators in 2026](/guides/comparisons/best-ai-image-generators-2026), and the surrounding design stack in the [Claude Design guide](/guides/design/claude-design-guide).

---

_Source: https://agentscamp.com/glossary/diffusion-model — Term on AgentsCamp._