cs.CVOct 19, 2026

Risk Decomposition for Hallucination-Aware Video Sampling

Authors: Chenyan Zheng, Fan Wu, Ruibo Li, Jiaxing Li, Minghao Yang, Bang Zhang, Guosheng Lin

Organizations: Zhejiang University · Nanyang Technological University · Tsinghua University · Tongyi Lab, Alibaba Group

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Despite achieving high visual fidelity, video models still suffer from physical hallucinations, limiting their reliability as world simulators. Existing methods leverage model-internal or external signals to improve physical plausibility, but their roles and limitations require a unified explanation. We propose a two-dimensional physical-hallucination risk framework that defines low-density sampling risk and high-density drifting risk. The first risk captures generations assigned low likelihood by the video model, for example, a ball suddenly teleporting despite the model favoring continuous motion, while the second risk captures physical inconsistencies in high-likelihood generations, for example, a ball bouncing progressively higher without additional energy. Our analysis explains how model-internal signals steer sampling toward higher-density regions of the learned distribution to address low-density sampling risk. However, higher model likelihood does not guarantee physical plausibility, motivating external guidance to address high-density drifting risk. Based on this framework, we develop Hallucination-Aware Sampler (HA-Sampler), which applies external physical guidance within latent refinement to generate physically plausible videos, and HA-Scorer, which derives a likelihood-based signal from the generator itself for evaluating generated videos using continuous normalizing flow (CNF), with coarse ODE integration and finite-difference divergence estimates. Experiments on Physics-IQ and VideoPhy benchmarks support our theoretical analysis and demonstrate that our methods outperform the compared sampling and scoring SOTA baselines.

Figures & tables

Figure 1: Overview of our two-dimensional physics hallucination risk framework. Left: Low-density sampling risk captures generations assigned low likelihood by the video model, while high-density drifting risk captures physical errors in high-likelihood generations. Middle: Model-internal physics-aware signals steer sampling toward higher-density regions of the learned distribution to reduce low-density sampling risk. Right: External physics-aware guidance helps correct the drifting between model-preferred generation and physical plausibility to address high-density drifting risk.
Figure 2: Overview of our methods. (a) HA-Sampler uses an external loss on the predicted clean video to guide the perturbation noise. The guided noise is combined with the original endpoint prediction before denoising continues. Grey dashed lines denote the flow of gradients. (b) HA-Scorer integrates the clean latent z0\mathbf{z}_{0} to the noise endpoint z1\mathbf{z}_{1} under the fixed base velocity field using a coarse Heun solver. Finite-difference Hutchinson estimator provides the divergence integral. In step 3, ztz_{t} and zt+1z_{t+1} fall on the same velocity line.
Figure 3: Performance comparison across the two-dimensional risk. Each axis reports the fraction of 100 cases with lower corresponding risk than the Euler sampler.
Figure 4: Sampling results on 2D toy distributions. The stacked layers compare Euler sampler, P&P (risk-1 baseline), guidance based on the negative log-likelihood (NLL) under the true data distribution (risk-2 baseline), and HA-Sampler. The Euler sampler uses a large number of sampling steps to reduce discretization error. Purple reference curves indicate the training-data geometry.
Figure 5: Qualitative comparison of samplers on Physics-IQ. We compare sampling methods in a dynamic collision scene (left) and a static scene (right). HA-Sampler produces more physically plausible results than the base sampler and P&P in both cases.
Figure 6: Qualitative comparison of candidates selection on Physics-IQ. For each prompt, all scorer methods select from the same candidate pool generated by the base model. Each row shows three frames capturing key stages of motion in the video.
Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

SamplerBoNRandomWMRewardGeoPhysQwen3-VLHA-Scorer (ours)
Base sampler227.7428.8426.5727.7328.99
427.4530.7526.5028.6430.77
826.3830.5125.6728.1830.88
P&P sampler227.2828.7927.0828.3329.38
427.0929.8826.3428.4230.48
827.0230.6426.9728.7431.40
Appendix
Table 6: Best-of- NN results with different candidate pool sizes. We report Physics-IQ Verified scores for selection methods using N∈{2,4,8}N\in\{2,4,8\} candidates from each sampler. Random denotes uniform random selection. Higher scores indicate better performance. The best and second-best scores in each row are bold and underlined, respectively.
SamplerRisk DecompositionΔ\Delta Risk/dim
Prior termVolume termRisk/dim ↓\downarrow
Euler1.10503-2.54120-1.436170.00000
C-TAG ( λ=1.25\lambda=1.25 )1.22623-2.76701-1.54079-0.10462
P&P1.21343-2.68283-1.46940-0.03323
CFG1.21803-2.70666-1.48863-0.05246
WMReward ( ∇\nabla )1.22506-2.21974-0.99468+0.44149
Appendix
Table 7: Low-density sampling risk on 100 cases. The prior and volume terms sum to the low-density sampling risk. All quantities are reported per latent dimension. Δ\Delta Risk is measured relative to the paired Euler baseline without classifier-free guidance. We use H=8H=8 Heun steps and P=2P=2 shared Hutchinson probes. ∇\nabla denotes gradient guidance.
SamplerRisk ↓\downarrowΔ\Delta Risk
Base (CFG =5=5 )0.4149410.000000
P&P0.415273+0.000332
WMReward ( ∇\nabla )0.414746-0.000195
HA-Sampler (ours)0.414180-0.000761
Appendix
Table 8: High-density drifting risk on 100 cases. We report mean WMReward surprise as a proxy for high-density drifting risk. Δ\Delta Risk is measured relative to the base sampler with CFG =5=5 . ∇\nabla denotes gradient guidance.
Table 9: Sensitivity of HA-Scorer to integration steps and probe count. We evaluate the full 100-case set and a fixed stratified 25-case subset. We fix p=2p=2 in (a), H=8H=8 in (b), and δrel=0.0125\delta_{\mathrm{rel}}=0.0125 throughout. ρ\rho denotes the Spearman rank correlation coefficient. NFE counts model forward evaluations per case.
Figure 7: Qualitative comparison of T2V generation on VideoPhy. For each prompt, we compare different samplers using three representative frames from each generated video. HA-Sampler improves physical plausibility in both cases and helps correct inconsistencies in the generated interactions.
Figure 8: Qualitative comparisons on Physics-IQ. Each example shows GT, Base, and Ours from top to bottom, with five frames arranged chronologically. Ours denotes videos generated with HA-Sampler and selected using HA-Scorer.
Figure 9: Additional qualitative comparisons on Physics-IQ. The layout follows Figure 8 .
Figure 10: Qualitative comparisons on VideoPhy. For each prompt, we show five frames from videos produced by the base model and HA-Sampler (Ours).
More in this topic

Physics-Guided Video Generation

See all 16 papers

Explore similar work

CardsList

  1. Off-Manifold Refinement: Guiding Video Generators with a Frozen World Model

    Aug 30, 2026Hai Nguyen-Truong, Tuan-Anh Vu, Dang HuynhPhysics-Informed Generative ModelingPhysical Consistency in Video Generation

    Fulbright University Vietnam · University of California, Los Angeles

  2. Proprio: Latent Self-Scoring and Inference-Time Refinement for Physically Plausible Video Generation

    May 27, 2026Mariam Hassan, Kaouther Messaoud, Wuyang Li +1Physical Consistency in Video GenerationImage-to-Video Generation

    1 École Polytechnique Fédérale de Lausanne (EPFL) · Télécom Paris, IP Paris

  3. PhysVid: Physics Aware Local Conditioning for Generative Video Models

    Mar 27, 2026Saurabh Pathak, Elahe Arani, Mykola Pechenizkiy +1Physics-Guided Video GenerationPhysical Consistency in Video Generation

    Eindhoven University of Technology