An open-source curriculum that takes you from a single array element to building, training, and reasoning about modern generative models, across fifteen progressive modules, each producing a visual artifact.
Three paths, each a few modules deep, building lesson by lesson. Linear by design, but experienced coders can jump to a later path. Open any module to see inside.
No lessons match that search.
Pixels, geometry, recursion, simulation. NumPy and PIL only, no networks yet. Visual intuition first.
Procedural generation, classical ML, animation, first neural networks, real-time. Arrays become inference.
GANs, VAEs, diffusion, language models, integration, capstone. The generative stack, end to end.
A recurring one-day workshop: the whole curriculum in a single sitting. First cohort below; the next is in planning.
Six modules, one sitting, in Berlin. The first cohort produced the data reported in /research.
The next one: a day with Academis, same shape as 01. Register interest now; date and venue follow.
A group show with Kristian Rother and Academis: six artists, printed and code-driven pieces, several built from curriculum modules. Archive below; next show unscheduled.
Over two days, 6 artists turned the IT studio Academis office into a gallery: prints on the walls, code-driven work on screen and projection. Many pieces grew straight from curriculum modules: equations, algorithms, and models arranged as objects you could walk between.
The curriculum is the artifact of a design-based research thesis. Three cycles, the same five questions; this is the third.
Which teaching patterns best scaffold the climb from arrays to generative AI?
How should hard concepts be split and sequenced so novices avoid overload?
When and how should real-time systems join the learning path?
How do you measure technical skill, creativity, and understanding together?
Do learners carry core concepts into new creative AI work, and what helps?