AI TrackerFrontier desk · AEST
Back to news
normalaitech

LoGo: Local-Global Rewards for Consistent Long-Horizon Video Generation

arXiv cs.AI · 3 Oct 2026, 03:26 AEST

LoGo introduces a new approach to improve 3D consistency in camera-controlled video generation. By blending global and spatially localized rewards, it enhances video quality and reduces inconsistencies over long horizons.

Key facts

Read aloud

LoGo introduces a new approach to improve 3D consistency in camera-controlled video generation. By blending global and spatially localized rewards, it enhances video quality and reduces inconsistencies over long horizons. Camera-controlled video models are rapidly advancing toward long generation horizons and complex camera control. A key failure mode is 3D inconsistency: as the camera moves, objects lose permanence and scene structures shift. We introduce LoGo, which blends global and spatially localized rewards for camera-controlled video models. LoGo effectively reduces local object shifts, artifacts, and global scene changes, illustrating the importance of credit assignment in post-training video models.

Sources in this cluster

LoGo: Local-Global Rewards for Consistent Long-Horizon Video Generation · AI Tracker