A large, visual course on generative AI — from VAEs and GANs to the diffusion models behind Stable Diffusion, text-to-image systems, and generation beyond images. Every idea is a figure you can drive, not an equation you skim.
The forward (noising) process, live.
A diffusion model is trained on a strangely simple task: take a noisy image and make it slightly less noisy.
Repeat that thousands of times, starting from pure static, and a picture appears. This course builds that idea from the ground up — through the VAEs and GANs that came before, the math that makes diffusion work, and the systems that turn a text prompt into an image.
Autoencoders and VAEs (with a latent space you can explore), then GANs — the adversarial game, mode collapse, and StyleGAN's control.
The forward and reverse processes, the training objective, samplers (DDPM, DDIM), classifier-free guidance, and the score-based view.
Latent diffusion and Stable Diffusion, text conditioning and ControlNet, text-to-image craft, and generation of video, audio, and 3D.
Seven sections are live now (through Latent & Conditional Diffusion); the rest are being written and appear in your dashboard as they ship.
You'll get the most from this if you're comfortable with neural networks and the idea of training by gradient descent. New to that? Start with Finguard ML, then come here for the generative half of the field.
Who it's for
No account, no install. Progress saves automatically in your browser, separate from your other courses.