Direct Forcing：面向少步自回归视频生成的单阶段表征空间监督训练

View PDF HTML (experimental)

Abstract:Few-step autoregressive video generation enables low-latency streaming, yet prevailing training pipelines separate causal adaptation from distribution matching. Recent representation-based text-to-video methods retain this separation by starting with an ODE initialization stage, even when the target distribution is defined directly by reference videos. We introduce \method, the first single-stage framework that uses reference videos both as paired examples of how a history continues and as a target population for self-generated rollouts. Specifically, Direct Forcing is trained with two complementary signals: (1) Causal Continuation Alignment learns paired spatiotemporal correspondence under reference histories; and (2) Rollout Distribution Matching employs maximum mean discrepancy to align the representation distributions of self-generated and reference videos. To coordinate these signals, we introduce Dynamic Gradient Balancing, which adapts their relative gradient contributions during training. Experiments with Wan2.1-1.3B show that Direct Forcing learns a four-step autoregressive generator in 250 updates without a separate ODE initialization stage or online diffusion-score supervision, achieving 84.63 VBench Total and 86.01 Quality.

Subjects:	Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Cite as:	arXiv:2610.30861 [cs.CV]
	(or arXiv:2610.30861v2 [cs.CV] for this version)
	https://doi.org/10.48550/arXiv.2610.30861 arXiv-issued DOI via DataCite

Submission history

From: Haojun Xu [view email]
[v1] Thu, 22 Oct 2026 07:41:41 UTC (20,786 KB)
[v2] Sat, 24 Oct 2026 18:24:41 UTC (20,833 KB)