Delta Forcing: Trust Region Steering for Interactive Autoregressive Video Generation

2026-05-14

AI Summary

This work addresses the fundamental trade-off in interactive autoregressive video generation between dynamic event responsiveness and long-term temporal consistency, a challenge exacerbated by trajectory drift caused by biased conditioning in existing approaches. To mitigate this, the authors propose Delta Forcing, a novel framework that introduces, for the first time, a trust-region mechanism into this domain. By modeling the trajectory discrepancy between a teacher model and the generator in latent space, Delta Forcing dynamically constructs an adaptive trust region to identify and suppress unreliable external supervisory signals. The method integrates teacher-student distillation, streaming long-sequence fine-tuning, and monotonic continuity constraints, thereby achieving high event responsiveness while significantly enhancing temporal stability and global coherence in generated videos. Experiments demonstrate that Delta Forcing consistently outperforms current state-of-the-art methods across multiple benchmarks.

Abstract

Interactive real-time autoregressive video generation is essential for applications such as content creation and world modeling, where visual content must adapt to dynamically evolving event conditions. A fundamental challenge lies in balancing reactivity and stability: models must respond promptly to new events while maintaining temporal coherence over long horizons. Existing approaches distill bidirectional models into autoregressive generators and further adapt them via streaming long tuning, yet often exhibit persistent drift after condition changes. We identify the cause as conditional bias, where the teacher may provide condition-aligned but trajectory-agnostic guidance, biasing generation toward locally valid yet globally inconsistent modes. Inspired by Trust Region Policy Optimization, we propose Delta Forcing, a simple yet effective framework that constrains unreliable teacher supervision within an adaptive trust region. Specifically, Delta Forcing estimates transition consistency from the latent delta between teacher and generator trajectories, and uses it to balance teacher supervision with a monotonic continuity objective. This suppress unreliable teacher-induced shifts while preserving responsiveness to new events. Extensive experiments demonstrate that Delta Forcing significantly improves consistency while maintaining event reactivity.

Problem

• autoregressive video generation
• temporal coherence
• conditional bias
• interactive generation
• trust region

Innovation

• Delta Forcing
• Trust Region
• Autoregressive Video Generation
• Conditional Bias
• Temporal Coherence