Deep Forcing: Training-Free Long Video Generation with Deep Sink and Participative Compression

TL;DR

This paper introduces Deep Forcing, a training-free method for long video generation that stabilizes and improves quality during extended autoregressive streaming, surpassing existing methods in fidelity and consistency.

Contribution

Deep Forcing presents two novel training-free mechanisms, Deep Sink and Participative Compression, to enhance long video generation without fine-tuning.

Findings

Achieves over 12x extrapolation in video length with quality improvements.

Outperforms existing methods like LongLive and RollingForcing in quality and consistency.

Maintains real-time generation while significantly extending video duration.

Abstract

Recent advances in autoregressive video diffusion have enabled real-time frame streaming, yet existing solutions still suffer from temporal repetition, drift, and motion deceleration. We find that naively applying StreamingLLM-style attention sinks to video diffusion leads to fidelity degradation and motion stagnation. To overcome this, we introduce Deep Forcing, which consists of two training-free mechanisms that address this without any fine-tuning. Specifically, 1) Deep Sink dedicates half of the sliding window to persistent sink tokens and re-aligns their temporal RoPE phase to the current timeline, stabilizing global context during long rollouts. 2) Participative Compression performs importance-aware KV cache pruning that preserves only tokens actively participating in recent attention while safely discarding redundant and degraded history, minimizing error accumulation under out-of-distribution length generation. Together, these components enable over 12x extrapolation (e.g. 5s-trained to 60s+ generation) with better imaging quality than LongLive, better aesthetic quality than RollingForcing, almost maintaining overall consistency, and substantial gains in dynamic degree, all while maintaining real-time generation. Our results demonstrate that training-free KV-cache management can match or exceed training-based approaches for autoregressively streaming long-video generation.