Step by Step
G
GANs — two networks competing adversarially
A Generator network produces fake data, while a Discriminator network tries to classify inputs as real or fake; the two are trained adversarially, each pushing the other to improve.
Example: the Generator tries to create increasingly convincing fake images, while the Discriminator tries to get increasingly better at telling real images from the Generator's fakes.
A
GAN challenges — mode collapse and instability
GANs are notoriously prone to mode collapse (the generator producing only a limited variety of outputs) and general training instability, both significant practical challenges.
Example: a GAN trained to generate faces might get stuck producing only a narrow range of similar-looking faces, rather than the full diversity present in the training data.
N
Diffusion — denoising from pure noise
Diffusion models work differently: a forward process gradually adds noise to an image over many steps, and the model learns a reverse process to remove that noise step by step — generation happens by starting from pure random noise and progressively denoising it.
Example: Stable Diffusion generating an image by starting with pure random static and gradually refining it, step by step, into a coherent picture.
S
Diffusion's advantages — stability, diversity, easier conditioning
Diffusion models (used by Stable Diffusion, DALL-E 3, and Midjourney) are generally more stable to train than GANs, produce better output diversity, and make text-conditioning (generating images from text prompts) easier to implement effectively.
Example: modern text-to-image tools overwhelmingly use diffusion rather than GANs, specifically because of these stability and diversity advantages.
Applied Walkthrough
1
A team building an image generation tool from scratch needs to choose between a GAN-based approach and a diffusion-based approach.
2
They consider GANs' well-known challenges: mode collapse (limited output diversity) and general training instability, both of which can make GANs difficult to train reliably.
3
Choosing diffusion instead, they get a more stable training process, better output diversity, and easier text-conditioning for generating images from text prompts.
4
This is exactly why modern leading text-to-image tools like Stable Diffusion, DALL-E 3, and Midjourney all use diffusion rather than the GAN approach that once dominated image generation.
Exam Application
Exams test whether you can describe both architectures' core mechanisms (GAN: generator vs. discriminator adversarial competition; diffusion: forward noise addition, reverse denoising) and whether you know diffusion has become the dominant approach for modern image generation specifically due to greater stability, diversity, and easier text conditioning.
⚠ Common Trap
The most common trap is assuming GANs are still the dominant image generation approach, since they were historically the most famous generative technique. In practice, diffusion models have now overtaken GANs as the dominant approach for image generation specifically because of their training stability and output diversity advantages.
✓ Quick Self-Check
1. What are the two competing networks in a GAN?
The Generator (produces fake data) and the Discriminator (classifies real vs. fake).
Tap to reveal / hide
2. What is mode collapse, and why is it a problem for GANs?
When the generator produces only a limited variety of outputs, failing to capture the full diversity of the training data.
Tap to reveal / hide
3. How does a diffusion model generate an image?
By starting from pure random noise and progressively denoising it, using a learned reverse process.
Tap to reveal / hide
4. Name three well-known tools that use diffusion models.
Stable Diffusion, DALL-E 3, and Midjourney.
Tap to reveal / hide
5. What are diffusion models' main advantages over GANs?
More stable training, better output diversity, and easier text conditioning.
Tap to reveal / hide