IMPACT: Image-Conditioned Motion with Physics-Aware Collision and Trajectory Guidance for Video Generation
Authors: Shiwei Yang, Yunlong Yu, Zhao Wang, Jungong Han
Organizations: the College of Information Science and Electronic Engineering, Zhejiang University, Hangzhou, 310027 China · Ningbo Innovation Center, Zhejiang University, Ningbo, China · the Department of Automation, Tsinghua University, Beijing 100084
Visually realistic video generation does not guarantee physically consistent motion. Generated objects may exhibit implausible collision responses or rebounds without energy dissipation, while even physically valid motion predictions may not be faithfully realized during synthesis. We propose IMPACT, a physics-state-driven image-to-video framework that addresses these challenges by coupling explicit dynamics prediction with complementary sparse--dense physical guidance. A learnably calibrated Physics State Generator unifies continuous free-flight dynamics with discrete collision responses, linking impact timing, post-impact velocity, rebound attenuation, and eventual settling within a coherent physical-state sequence. To translate this evolution into effective generative constraints, IMPACT derives sparse trajectories and dense region-level motion priors from the same physical rollout. Sparse trajectories constrain global object displacement, while dense priors guide diffusion sampling to encourage coherent object-region motion and suppress unwanted background changes. Their shared physical basis coordinates global and local motion guidance to reduce prediction--generation mismatch. Experiments on 100 controlled sphere--plane collision scenes spanning different material combinations and visual environments demonstrate improved physical plausibility over representative general-purpose, physics-grounded, and motion-controllable baselines. The project repository is available at https://weiweiwiewei.github.io/impact
Figures & tables
Fig. 1: Illustration of two challenges in physically consistent video generation. (a) Inconsistent motion evolution: a sphere rebounds above its initial release height despite being released from rest, indicating an implausible gain in energy. (b) Prediction-generation mismatch: even when object positions follow a physically valid prediction (cyan outlines), the generated video can exhibit object deformation, unstable contact, and background drift.
Fig. 2: Overview of IMPACT. The Image-to-Physics (I2P) module estimates a compact physical configuration from an initial image and a text prompt. The Physics State Generator (PSG) combines continuous free-flight dynamics with discrete collision events to model free fall, rebound attenuation, and settling. Sparse trajectories and dense region-level motion priors are derived from the same physical rollout and provide complementary guidance for video synthesis. Both the video backbone and the pretrained trajectory-control module remain frozen.
Fig. 3: Visualization of the Image-to-Physics (I2P). (a) Object localization and segmentation. (b) Monocular depth estimation. (c) Camera calibration and gravity estimation. (d) Sphere–plane geometry recovery from back-projected observations. (e) Material conditioning from the text prompt. (f) Construction of the plane-aligned canonical physical state and camera parameters.
Fig. 4: Complementary Sparse-Dense Physical Guidance. The physics-derived sparse trajectory Ctraj conditions the frozen LTX-Video backbone through pretrained Motion Track IC-LoRA. Dense physical priors Pdense define object-motion and background-stability constraints on the predicted clean latent. Their gradients update the noisy latent during selected denoising steps without modifying the pretrained model parameters.
Fig. 5: Pairwise human rating win-rate heatmap based on the Overall Preference (OPP) metric. Each cell reports the percentage of non-tied comparisons in which the row method receives a higher rating than the column method. Ratings are paired by test video and evaluator, with ties excluded from the calculation. Warmer colors indicate higher winning fractions.
Fig. 6: Qualitative comparison of the evaluated free-fall and bouncing sequences. Representative frames illustrate free fall, contact, rebound attenuation, and settling.
Method
Resolution
Frames
FPS
Additional Input
CogVideoX-I2V
1024×512
81
25
None
HunyuanVideo-I2V
1344×672
81
25
None
Wan-I2V
1024×512
81
25
None
LTX-Video
1024×512
81
25
None
PhysGen
512×512
81
25
Physical parameters
MotionCraft
512×512
81
25
Simulated motion
TABLE I: Inference configurations and additional input requirements of the compared methods.
Method
Human Evaluation
VLM Evaluation
PEC ↑
FDR ↑
CCR ↑
REC ↑
OPP ↑
CFR (%) ↓
PEC ↑
FDR ↑
CCR ↑
REC ↑
OPP ↑
CFR (%) ↓
General-Purpose Video Generators
CogVideoX-I2V
1.60±0.39
1.58±0.51
1.47±0.36
1.28±0.23
1.36±0.25
37.4
2.49
1.79
2.27
1.27
1.92
14.0
HunyuanVideo-I2V
2.54±1.04
2.18±1.05
1.99±0.79
1.73±0.53
1.85±0.67
16.2
2.88
2.00
2.71
1.57
2.16
10.0
Wan-I2V
2.19±0.58
1.96±0.63
1.85±0.50
1.70±0.47
1.75±0.47
30.4
3.33
2.69
3.43
2.11
2.90
0.0
LTX-Video
2.52±0.84
1.95±0.67
2.11±0.74
1.83±0.58
1.95±0.59
10.5
3.32
2.73
3.43
2.12
2.87
0.0
TABLE II: Quantitative comparison with general-purpose, physics-grounded, and motion-controllable video generation methods under both human and VLM evaluations. Human evaluation scores are reported as the mean ± standard deviation of evaluator-level mean ratings on a 1-5 scale.
Fig. 7: Qualitative comparison with physics-grounded methods at matched timestamps. PhysGen exhibits delayed ground contact and cutout-like motion, while MotionCraft suffers from object duplication and scene artifacts. In contrast, our method produces coherent ground contact and rebound dynamics while maintaining visual consistency.
Fig. 8: Qualitative comparison with motion-controllable methods at matched timestamps. MotionPro exhibits ground penetration and object disappearance, while WanMove suffers from visual artifacts and missing rebound behavior. In contrast, our method produces coherent ground contact, damped rebounds, and eventual rest.
Method
PE ↓
VE ↓
F1 ↑
TE ↓
AE ↓
Analytic Physics
0.5249
1.4353
0.372
0.0333
0.7746
PSG (Full)
0.2658
1.1685
0.530
0.0261
0.1751
TABLE III: Physical-state prediction accuracy on the Kubric validation set under identical I2P initialization.
Variant
PE ↓
VE ↓
F1 ↑
TE ↓
AE ↓
w/o Impact Loss
0.2677
1.6037
0.477
0.0626
0.1769
Full (Oracle)
0.2422
1.2131
0.537
0.0529
0.1573
Full (I2P)
0.2658
1.1685
0.530
0.0261
0.1751
TABLE IV: Impact-supervision ablation and initial-state comparison on the Kubric validation set.
Method
Human Evaluation
VLM Evaluation
PEC ↑
FDR ↑
CCR ↑
REC ↑
OPP ↑
CFR (%) ↓
PEC ↑
FDR ↑
CCR ↑
REC ↑
OPP ↑
CFR (%) ↓
Trajectory Variants
PSG-KF Linear
3.46±0.61
3.05±0.40
3.22±0.40
2.79±0.19
2.95±0.23
0.3
4.18
3.29
4.02
3.12
3.25
0.0
PSG-KF Smoothstep
3.40±0.58
3.06±0.41
3.26±0.41
2.74±0.44
2.94±0.35
0.3
4.24
3.62
4.13
3.34
3.51
0.0
Naive Physics
3.33±0.76
2.91±0.53
3.28±0.53
2.62±0.49
2.86±0.38
0.3
4.22
3.47
4.10
3.27
3.42
0.0
Sparse–Dense Guidance
TABLE V: Ablation of motion-prior construction and sparse–dense physics guidance under human and VLM evaluations. Human evaluation scores are reported as the mean ± standard deviation of evaluator-level mean ratings on a 1-5 scale.
Fig. 9: Qualitative comparison of LTX-Video, PSG + Sparse, and Ours (Full) at matched timestamps. LTX-Video exhibits spatial deviations and physically inconsistent motion despite producing ground collisions. PSG + Sparse follows the predicted trajectory but suffers from local object enlargement, revealing a prediction–generation mismatch. In contrast, our full method preserves more consistent object scale and coherent collision dynamics.
Physical consistency in video generation studies how to make and evaluate videos whose geometry, motion, interactions, and state changes obey real-world laws.
Max Planck Institute for Informatics, Saarland Informatics Campus · Saarbrücken Research Center for Visual Computing, Interaction and Artificial Intelligence · Singapore University of Technology and Design +2