NVIDIA's WorldTrace: Training-Free Memory Stops Video World Models From Forgetting Scenes

cindy_x_wu · x · 2026-11-21

The Problem

Interactive video world models generate realistic worlds but lack visual persistence: turn the camera away and back, and the scene has changed. The culprit is the KV cache acting as growing visual memory—once rollouts exceed the training horizon, temporal RoPE offsets go out of range and attention fails to retrieve stored content. Naively compressing the cache in RoPE-rotated space also degrades memory by averaging keys with incompatible positional phases.

Method: WorldTrace

Results

On LoopBench, WorldTrace improves average DINOv2 recall by 32.6% over a sliding window at the same attended-cache size.

Authors span NVIDIA, Princeton, Toronto, and Vector Institute (first author Xindi Wu).

Related event: NVIDIA's WorldTrace Fixes Scene Amnesia in Video World Models Without Retraining(6 posts)→

Original post →

More from Research

Research channel →