DiReCT: Disentangled Regularization of Contrastive Trajectories for Physics-Refined Video Generation

• URL: http://arxiv.org/abs/2603.25931v1
• Date: Thu, 26 Mar 2026 21:53:47 GMT
• Title: DiReCT: Disentangled Regularization of Contrastive Trajectories for Physics-Refined Video Generation
• Abstract summary: Flow-matching video generators produce temporally coherent, high-fidelity outputs yet routinely violate elementary physics.<n>We identify a fundamental obstacle in the text-conditioned video setting: semantic-physics entanglement.<n>We formalize this gradient conflict, deriving a precise alignment condition that reveals when contrastive learning helps versus harms training.
• Score: 40.41107421160271
• License: http://creativecommons.org/publicdomain/zero/1.0/
• Abstract: Flow-matching video generators produce temporally coherent, high-fidelity outputs yet routinely violate elementary physics because their reconstruction objectives penalize per-frame deviations without distinguishing physically consistent dynamics from impossible ones. Contrastive flow matching offers a principled remedy by pushing apart velocity-field trajectories of differing conditions, but we identify a fundamental obstacle in the text-conditioned video setting: semantic-physics entanglement. Because natural-language prompts couple scene content with physical behavior, naive negative sampling draws conditions whose velocity fields largely overlap with the positive sample's, causing the contrastive gradient to directly oppose the flow-matching objective. We formalize this gradient conflict, deriving a precise alignment condition that reveals when contrastive learning helps versus harms training. Guided by this analysis, we introduce DiReCT (Disentangled Regularization of Contrastive Trajectories), a lightweight post-training framework that decomposes the contrastive signal into two complementary scales: a macro-contrastive term that draws partition-exclusive negatives from semantically distant regions for interference-free global trajectory separation, and a micro-contrastive term that constructs hard negatives sharing full scene semantics with the positive sample but differing along a single, LLM-perturbed axis of physical behavior; spanning kinematics, forces, materials, interactions, and magnitudes. A velocity-space distributional regularizer helps to prevent catastrophic forgetting of pretrained visual quality. When applied to Wan 2.1-1.3B, our method improves the physical commonsense score on VideoPhy by 16.7% and 11.3% compared to the baseline and SFT, respectively, without increasing training time.

Related papers

• Balancing Emotional Alignment and Semantic Consistency in Image Generation via Reinforcement Learning with Valence-Arousal Anchoring [55.57242007016976]
  Continuous emotion control in text-to-image generation requires a model to improve affective alignment without changing the objects, layout, or scene described by the prompt.<n>Existing supervised emotion-injection methods often optimize feature-space proxies and may therefore exhibit emotion-semantic drift.<n>We address this problem with a flow-matching image-generation framework that combines continuous valence-arousal conditioning, Group Relative Policy Optimization, and a neutral semantic anchor.
  arXiv  Detail & Related papers  (2026-09-11T13:24:35Z)
• Contrast encodes inductive bias: separating slow noise from dynamics in predictive representation learning [0.0]
  Self-supervised methods have been shown to confuse slowly varying noise with the dynamical signals they aim to capture.<n>We show that contrastive predictive objectives preferentially encode these features instead of the true latent variables governing the system.<n>Our results point toward principles for designing contrastive predictive objectives in dynamical representation learning.
  arXiv  Detail & Related papers  (2026-06-05T18:28:43Z)
• Flow Mismatching: Unsupervised Anomaly Detection via Velocity Discrepancies in Flow Matching Models [4.861045498353029]
  unsupervised anomaly detection method that deliberately avoids reconstruction-based paradigms.<n>We treat flow matching as geometric dynamics and leverage a key insight: anomalies occur at places where the learned normal flow disagrees with the geometric path toward a test image.
  arXiv  Detail & Related papers  (2026-05-21T22:01:31Z)
• Physics-Informed Generative Solver: Bridging Data-Driven Priors and Conservation Laws for Stable Spatiotemporal Field Reconstruction [19.98127585683643]
  We introduce a physics-informed generative solver that separates stable prior learning from inference-time enforcement of conservation laws.<n>In acoustics, the method co-generates pressure and particle velocity from sparse sensors, enabling dense virtual arrays that suppress spatial aliasing.<n>The same framework generalizes to real-world ERA5 meteorological fields under extreme sparsity.
  arXiv  Detail & Related papers  (2026-05-21T11:24:48Z)
• MotionCFG: Boosting Motion Dynamics via Stochastic Concept Perturbation [60.48914865049489]
  MotionCFG is a framework that enhances motion dynamics by contrasting a target concept with its noise-blured counterparts.<n>Unlike explicit negations, this approach facilitates implicit hard negative guidance without shifting the global semantic identity.<n> MotionCFG consistently improves motion dynamics across state-of-the-art T2V frameworks.
  arXiv  Detail & Related papers  (2026-03-14T18:39:37Z)
• On Exact Editing of Flow-Based Diffusion Models [97.0633397035926]
  We propose Conditioned Velocity Correction (CVC) to reformulate flow-based editing as a distribution transformation problem driven by a known source prior.<n>CVC rethinks the role of velocity in inter-distribution transformation by introducing a dual-perspective velocity conversion mechanism.<n>We show that CVC consistently achieves superior fidelity, better semantic alignment, and more reliable editing behavior across diverse tasks.
  arXiv  Detail & Related papers  (2025-12-30T06:29:20Z)
• Enhancing Physical Plausibility in Video Generation by Reasoning the Implausibility [37.011366226968]
  Diffusion models can generate realistic videos, but existing methods rely on implicitly learning physical reasoning from large-scale text-video datasets.<n>We introduce a training-free framework that improves physical plausibility at inference time by explicitly reasoning about implausibility and guiding the generation away from it.
  arXiv  Detail & Related papers  (2025-09-29T12:32:54Z)
• The Butterfly Effect: Neural Network Training Trajectories Are Highly Sensitive to Initial Conditions [51.68215326304272]
  We show that even small perturbations reliably cause otherwise identical training trajectories to diverge-an effect that diminishes rapidly over training time.<n>Our findings provide insights into neural network training stability, with practical implications for fine-tuning, model merging, and diversity of model ensembles.
  arXiv  Detail & Related papers  (2025-06-16T08:35:16Z)
• Curricular Contrastive Regularization for Physics-aware Single Image Dehazing [56.392696439577165]
  We propose a novel curricular contrastive regularization targeted at a consensual contrastive space as opposed to a non-consensual one. Our negatives, which provide better lower-bound constraints, can be assembled from 1) the hazy image, and 2) corresponding restorations by other existing methods. With the unit, as well as curricular contrastive regularization, we establish our dehazing network, named C2PNet.
  arXiv  Detail & Related papers  (2023-03-24T18:18:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.