I've been experimenting with Waypoint, the newly-released world model from Overworld. I've been thinking about world models as a potential new medium for visual artists. When I'm painting, I often find myself inhabiting a scene or a world, and offering viewers an immersive way to interact with that world is a compelling concept.

To test this out, I fed one of my own paintings into the model. The results show some coherence challenges (possibly due to the soft lines of the painting, the 9:16 output vs. 16:9 input, or because the model is trained primarily on video game footage that probably doesn't include many snowy forest paths).

I'd love to hear ideas for adjusting model behavior. I'm also sharing the source painting here for anyone who wants to experiment. My only asks: share the result with me, and credit me in any public post.