---
title: "Diffusion: make images and video"
lesson_id: "03"
---

# Diffusion: make images and video

- **Lesson ID:** 03
- **Goal:** Explain the official `diffusion` presets by the media they make.
- **Human lesson:** [03-diffusion.html](03-diffusion.html)

## Prerequisites

- Lesson 01 and the difference between making new media and describing existing media.
- The image-model lesson in [the architectures course](../model-architectures-plain/03-image-models.html).

## Inputs, outputs, and artifacts

- **Inputs:** A prompt and a desired still-image or video output.
- **Outputs:** A chooser for Stable Diffusion, FLUX.2 Klein, and Wan.
- **Artifacts:** Progressive text-to-media `.flow` diagrams and prior-course links.

## Agent build steps

1. Define `diffusion` as the path for making new media.
2. Name `sd-1.5`, `sd-2.1`, and `sd-3.5-medium` as Stable Diffusion text-to-image presets.
3. Name `flux2-klein-4b` as the FLUX.2 Klein text-to-image preset.
4. Name `wan-t2v-1.3b` as the Wan text-to-video preset.
5. Contrast making new media with ClothingCNN classification and connect the path to runtime work.

## Constraints

Use only the listed diffusion names. No invented image capabilities, math, CUDA, quizzes, Mermaid, or em dashes.

## Key concepts

- `diffusion` turns a prompt into new media, not a label for an existing image.
- Still-image presets include `sd-1.5`, `sd-2.1`, `sd-3.5-medium`, and `flux2-klein-4b`.
- `wan-t2v-1.3b` is the Wan text-to-video preset.
- Classification describes what is already in a picture. Diffusion draws new pixels or frames.

## Takeaways

- You can pick `diffusion` when the app must create media from a prompt.
- You can choose a still-image preset or `wan-t2v-1.3b` for moving frames.
- You can say when utility vision fits better than diffusion for an existing photo.

## Acceptance checks

- Stable Diffusion, FLUX.2 Klein, and Wan jobs are clearly separated.
- Still-image versus video output is explicit.
- Classification is contrasted with making new media.
- Progressive `.flow` diagrams, prior-course links, and the Agent brief link are present.