InSpatio-World 1.5 Brings Scene Roaming
InSpatio-World 1.5 turns images, panoramas, image sets, or video into explorable scenes with controllable camera trajectories, wider viewpoint changes, and improved spatiotemporal consistency. The open-source 1.3B model also supports bullet-time sequences and dynamic next-view prediction.
This is more compelling as a camera-control and world-model primitive than as another video generator.
- –Multiple input modes make it useful for cinematography, immersive media, and robotics data generation. [Official project page](https://inspatio.github.io/inspatio-world-1.5/)
- –Its key differentiator is maintaining scene structure and subject identity while the camera moves through previously unseen viewpoints.
- –Bullet-time generation could reduce the need for dense synchronized camera rigs in early creative workflows.
- –Local deployment is promising but not turnkey: the repository requires Python 3.10, CUDA 12.6, Depth Anything 3, model checkpoints, and user-supplied target trajectories for video. [GitHub repository](https://github.com/inspatio/inspatio-world-v1.5)
- –The important unanswered question is whether consistency survives long camera paths and heavily occluded geometry outside curated demos.
DISCOVERED
1h ago
2026-10-04
PUBLISHED
1h ago
2026-10-04
RELEVANCE
AUTHOR
AI Search