Lab · 4 min read
Latent navigation: steering a model through its own imagination
Treating the latent space as a map you can walk, not a black box you query.
Noetic Lab
The space between two images
Every generative model holds a continuous latent space where each point is a possible image. Latent navigation treats that space as terrain to be walked rather than a black box to be queried — interpolating, nudging and steering between concepts to find frames that no single prompt would land on.
By moving deliberately along learned directions — more light, older, wider lens — we turn one-shot generation into something closer to exploration, where the interesting results live in the gaps between obvious prompts.