Clean Forcing: Drift-Resistant Autoregressive Video Diffusion with a Frozen Base

Wenqing Wang; Joonghyuk Shin; Jonathan Tremblay; Chan Hee Song; Yun Fu

cs.CVOct 19, 2026

Clean Forcing: Drift-Resistant Autoregressive Video Diffusion with a Frozen Base

Authors: Wenqing Wang, Joonghyuk Shin, Jonathan Tremblay, Chan Hee Song, Yun Fu

Organizations: Northeastern University · Seoul National University · NVIDIA

Abstract

Autoregressive (AR) video diffusion models generate streaming video of arbitrary length by conditioning each new chunk of frames on those already generated. However, they suffer from exposure bias, where a model trained on clean context must continue from its own imperfect outputs, and errors compound until quality collapses within seconds (drift). Existing methods either require cluster-scale base retraining, apply training-free inference corrections with inconsistent results, or train frozen-base correctors for sampling and cache errors. We instead find that drift is largely deterministic and therefore learnable, as holding the noisy state fixed while varying only the history reveals a counterfactual velocity gap that is ~95% systematic across noise seeds. Based on this observation, we introduce Clean Forcing, which trains a 5.9M-parameter LoRA corrector on the frozen base model to regress this gap, with a closed-loop stage that adds DAgger-style aggregation and a drift-contraction objective to expose the corrector to its own rollouts. Because the clean histories can be the base model's own single-shot generations, training needs no real videos, and the merged corrector adds no inference cost. In 50 s text-to-video generation, Clean Forcing reduces Δ-drift from +11.3 to +1.65 without real video and to -0.16 with 40 real clips, improving on the best published result of +3.57 with over 150x less data. Extensive experiments demonstrate that Clean Forcing outperforms prior methods in aesthetic quality and data efficiency with reduced drift for long-horizon video generation.

Figures & tables

Figure 1: Clean Forcing reduces drift during AR rollout. The first 20 s of two 50 s generations with the same prompt and seed are shown. The adapted AR base (top) becomes oversaturated and loses the glass, while the same model with the merged Clean Forcing corrector (bottom) preserves the scene and pouring motion.

Figure 2: Overview. (1) At the shared noisy state $z_{t}$ , $\Delta v:=v_{\theta}(z_{t},h_{\mathrm{clean}},t)-v_{\theta}(z_{t},h_{\mathrm{gen}},t)$ is the history-induced velocity gap. The corresponding one-step clean-latent predictions differ by $\sigma_{t}\|\Delta v\|$ , contributing to drift over successive chunks. (2) The LoRA student conditions on $h_{\mathrm{gen}}$ and matches the frozen teacher’s prediction under $h_{\mathrm{clean}}$ ( $\mathcal{L}_{\mathrm{CF}}$ ) ( $\phi$ denotes the weights of the LoRA corrector $r_{\phi}$ ). (3a) We generate new training pairs $D_{\phi}$ with the corrector active and continue training on $D_{0}\cup D_{\phi}$ , where $D_{0}$ contains pairs from uncorrected rollouts. (3b) $\mathcal{L}_{\mathrm{con}}$ backpropagates the next chunk’s error through the undetached prediction $\hat{x}_{0}$ .

	Official VBench $\uparrow$					Motion	Long-Horizon Quality
Method	Subj.	Backg.	Aesth.	Imag.	Smooth	Dynamic	MUSIQ $\uparrow$	$\Delta$ -drift $\downarrow$
Matched base
Adapted base	65.7	76.7	44.9	47.9	91.1	29.7	44.8	$+11.27{\scriptstyle\pm 0.81}$
Context noise ( $\sigma=0.2$ )	67.2	77.8	45.2	51.4	85.8	32.0	49.1	$+12.05{\scriptstyle\pm 0.76}$
History guidance ( $w=1.2$ )	64.6	76.1	42.4	47.7	94.5	0.8	42.3	$+20.55{\scriptstyle\pm 0.65}$
Pathwise TTC ( $\{500,250\}$ )	59.7	74.0	40.2	37.4	91.0	64.1	35.5	$+13.96{\scriptstyle\pm 0.91}$

Table 1: Quantitative comparison on 128 MovieGen prompts at 50 s, with

\pm

denoting SEM.

\Delta

-drift is first-20% minus full-video MUSIQ ( Po et al., 2025 ) . Across its variants, Clean Forcing achieves the lowest drift and highest aesthetics. It ranks second in subject and background consistency, imaging quality, and MUSIQ. Among shared-prompt rows, bold with light red marks the best value and underline with light green the second best. Published

\dagger

values use different prompts, and dynamic degree only checks for motion collapse; neither is ranked (Appendix C ).

Figure 3: Per-frame quality over 50 s. Mean MUSIQ

\pm

SEM across 128 prompts. Both Clean Forcing variants remain stable as the base and training-free baselines decline. Ripples mark chunk boundaries.

Panel of Figure 4 — Figure 4: Qualitative comparison over 50 s. Frames are sampled at equal intervals for the spiral staircase (top, p098) and mechanical clock (bottom, p101) prompts. The base and training-free baselines develop saturation or structural artifacts, while Self Forcing shifts color. Both Clean Forcing variants retain the scene throughout the rollout.

Panel of Table 2 — Table 2: Ablation studies. We extend held-out clips by 48 latent frames ( ${\sim}12$ s), using paired clips and seeds across comparable rows. Bold and underlining indicate the best and second-best reported values. (Drift removed is measured relative to the unadapted base; only this metric was recorded for the DAgger-only checkpoint. Factual targets a collapse motion, making drift and MUSIQ uninformative (Appendix C ). GT-anchored targets b were evaluated on a separate dataset and increased drift, so their values are not directly comparable.)

Video Diffusion Models

Video diffusion models generate or edit temporally coherent video by iteratively denoising latent sequences conditioned on text, images, motion, or other signals.

See all 732 papers

Custom Forcing: Training-Free Subject Customization for Autoregressive Video Generation
Oct 2, 2026Yunseung Ok, Hyunsoo Kim, Minseo Kim +1Identity-Preserving Video Generation Streaming Video Generation
Kyung Hee University · The University of Texas at Austin
FreqForcing: Autoregressive Long Video Generation via Spectral Self-Anchoring
Jul 29, 2026Jiatong Li, Leo Liang, Linghe Kong +1Video Diffusion Models Autoregressive Diffusion
Shanghai Jiao Tong University · Tencent HY Team
In-Distribution Forcing for Long Video Generation at Test Time
Oct 2, 2026Jeongwoo Shin, Youngyoon Choi, Sangwoo Jo +5Long Video Generation Video Diffusion Models
Seoul National University · Korea University · Sungkyunkwan University +1

Oct 2, 2026J/K move · Enter open · S save

Custom Forcing: Training-Free Subject Customization for Autoregressive Video Generation

Yunseung Ok, Hyunsoo Kim, Minseo Kim, Suhyun Kim

Kyung Hee University · The University of Texas at Austin

Identity-Preserving Video Generation Streaming Video Generation Video Generation Autoregressive Video Generation Causal Video Generation

Open paper AskPDF arXiv

Autoregressive video models can generate minute-long videos in real time, but they produce generic subjects from text rather than specific subjects from user-provided images. Existing customization methods either require costly per-subject optimization or use pretrained conditioning networks that jointly process all video frames with bidirectional attention. Neither approach is designed for causal streaming. We present Custom Forcing, a training-free method that stores reference-based anchor frames in the persistent KV cache of a frozen autoregressive video model. However, fixed anchors face two limitations: simple conditioning allows identity to drift, and the text prompt continues to favor a generic subject. To address these problems, drift-adaptive value amplification (DVA) scales reference influence with the degree of identity drift, while anchor contrast guidance (ACG) steers generation away from the generic class prior. Over two-minute rollouts, fixed anchors fall from 0.58 to 0.42 in DINO-I, while Custom Forcing keeps it between 0.58 and 0.62 without reducing motion. Custom Forcing also achieves higher subject similarity than bidirectional customization methods and better preserves identity over 30s than causal image-to-video and reference-to-video models, while generating each frame 9.5-28.5 times faster than these long-video baselines.

Abstract

Video Diffusion Models

Explore similar work

Custom Forcing: Training-Free Subject Customization for Autoregressive Video Generation

FreqForcing: Autoregressive Long Video Generation via Spectral Self-Anchoring

In-Distribution Forcing for Long Video Generation at Test Time