Building generative models that render the world.

The next generation of ads, products, and visual media will not be captured. It will be generated.

We build image models that understand light, material, composition, and taste. The training systems to produce them. The evaluation to measure what no benchmark currently measures. The datasets that do not exist yet. Every layer of the stack, from scratch.

Read the research.

Notes from the lab on what's in progress — open as we go.