Hi MAGI-1 team,
Autoregressive video generation is exactly where our DreamX-World work intersects with the broader direction of this repo.
We recently open-sourced AMAP-ML/DreamX-World, an interactive video world model focused on chunked rollouts, camera-conditioned generation, memory over revisits, and promptable scene events.
A few cases we have been studying closely are repeated camera loops, return-to-origin views, event-triggered scene changes, and control accuracy after many generated chunks.
If these long-rollout AR video cases are useful to the community here, we would be happy to share examples or discuss how to make them easier to reproduce.
Thanks for the great work and for pushing autoregressive video generation forward.
Hi MAGI-1 team,
Autoregressive video generation is exactly where our DreamX-World work intersects with the broader direction of this repo.
We recently open-sourced AMAP-ML/DreamX-World, an interactive video world model focused on chunked rollouts, camera-conditioned generation, memory over revisits, and promptable scene events.
A few cases we have been studying closely are repeated camera loops, return-to-origin views, event-triggered scene changes, and control accuracy after many generated chunks.
If these long-rollout AR video cases are useful to the community here, we would be happy to share examples or discuss how to make them easier to reproduce.
Thanks for the great work and for pushing autoregressive video generation forward.