Ψ-talk
Notes, papers, and ideas from the PSI project.
-
Ψ-0.6: Adaptive Streams of Thought in a Unified Vision-Language World Model
A unified vision-language world model that reasons through streams of thought: sequences interleaving images and text with its own predictions of depth, optical flow, and camera motion.
Read release →
-
Physical Object Understanding with Physically Controllable World Modeling
A single probabilistic world model discovers objects from their physics, segments articulated parts, manipulates objects in 3D, and reasons about support relationships—achieving SOTA on SpelkeBench, DragAMove, and 3DEditBench.
Read article →
-
Releasing PSIv0.5
Introducing PSIv0.5, an 8B RGBCFD model for richly controllable world modeling across RGB, camera, flow, and depth tokens.
Read release →
-
PSI: Richly Controllable Physical World Modeling
A working gallery of PSI rollouts and control sketches across multi-turn physical prompts, unconditional generation, 3D pushes and pulls, rotations, hand manipulation, and object interactions.
Read gallery →