Atlas user describes generating camera-controlled video from a few scene views
The user says Atlas combines input images, previously generated views and camera positions and orientations into a “spatial context” that guides future frames.
TLDR
A user describes feeding Atlas a few views of the “gardenvase” scene, turning the camera to face the other side and using text to prompt a fourth view containing an Atlas ad. They say they can then generate a smooth, camera-controlled video sequence through the scene. According to the same user, Atlas grounds future frames in a “spatial context” combining input images, its own generated views and their camera positions and orientations.
Combined views
14.3K
4 Sources, first seen ago
