Generative AI: Images & Video
How machines create images and video: VAEs, GANs, diffusion, latent diffusion, and flow matching.
Before this: Deep Learning.
-
20 min
Step 1Variational autoencoders
Learn a latent space you can sample from.
Variational Autoencoders | Generative AI AnimatedDeepia
-
21 min
Step 2GANs
A generator and a critic, trained against each other.
Generative Adversarial Networks (GANs) - ComputerphileComputerphile
-
37 min
Step 3How AI images and videos work
The best visual explanation of diffusion there is.
But how do AI images and videos actually work? | Guest video by Welch Labs3Blue1Brown
-
18 min
Step 4How image generators work
Noise in, picture out, explained simply.
How AI Image Generators Work (Stable Diffusion / Dall-E) - ComputerphileComputerphile
-
32 min
Step 5DDPM, animated
The math of denoising diffusion, made visual.
Diffusion Models: DDPM | Generative AI AnimatedDeepia
huggingface/diffusion-models-class★ 4.4k -
31 min
Step 6Diffusion from scratch
Code and train a small diffusion model in PyTorch.
Diffusion models from scratch in PyTorchDeepFindr
lucidrains/denoising-diffusion-pytorch★ 10.7k -
13 min
Step 7Latent diffusion
Why Stable Diffusion works in a compressed space.
How does Stable Diffusion work? – Latent Diffusion Models EXPLAINEDAI Coffee Break with Letitia
-
17 min
Step 8Stable Diffusion in code
Run the pipeline piece by piece.
Stable Diffusion in Code (AI Image Generation) - ComputerphileComputerphile
huggingface/diffusers★ 35k -
5h 04m
Step 9Code Stable Diffusion from scratch
The full model, every component, in PyTorch.
Coding Stable Diffusion from scratch in PyTorchUmar Jamil
-
16 min
Step 10Flow matching vs. diffusion
The newer training recipe behind today's image models.
Flow-Matching vs Diffusion Models explained side by sideAI Coffee Break with Letitia
-
12 min
Step 11How video generation works
From images to consistent motion.
How do AI video generation models work?Google for Developers
-
39 min
Step 12ComfyUI
Build image-generation workflows visually.
How to use ComfyUI for beginners.Sebastian Kamph
Comfy-Org/ComfyUI★ 147.8k
Build it
Watching isn't learning. Pick one and put it on GitHub.
- Train a VAE on MNIST and render a grid walking through its latent space.
- Train a tiny DDPM on 32×32 images and show samples across denoising steps.
- Train flow matching and diffusion on the same 2D toy distribution and compare sample quality per step.