Physics in 2-Steps: Locking Motion Priors Before Visual Refinement Erases Them

Hugging Face Daily Papers Papers

Summary

PhaseLock is a training-free framework that preserves motion priors from early-step inference to improve physical consistency in image-to-video diffusion models, achieving 6.2 point improvement with minimal overhead.

Image-to-Video diffusion models leverage input images to generate visually stunning content, yet frequently produce motion that violates physical laws. We reveal a surprising finding: a 2-step generation often exhibits better physical consistency than a 50-step output from the same model. Through spectral analysis, we trace this to phase erosion during denoising; the phase degrades significantly (dropping by approx 18% from step 2 to step 50), whereas the magnitude remains relatively stable. Building on this insight, we propose PhaseLock, a training-free framework that preserves the valid motion priors from few-step inference throughout the denoising trajectory. Rather than relying on full-step inference for physical consistency, PhaseLock extracts a motion prior from just 2 steps and enforces it onto high-fidelity generation via Latent Delta Guidance. Our approach effectively mitigates phase degradation, improving physical consistency by an average of 6.2 points across diverse models while largely maintaining visual fidelity, with negligible overhead (1.06times time, 1.02times memory) and reduced reliance on expensive external guidance methods (sim5times time).
Original Article
View Cached Full Text

Cached at: 06/08/26, 11:15 AM

Paper page - Physics in 2-Steps: Locking Motion Priors Before Visual Refinement Erases Them

Source: https://huggingface.co/papers/2606.06361

Abstract

PhaseLock is a training-free framework that improves physical consistency in image-to-video diffusion models by preserving motion priors from early-step inference throughout the denoising process.

Image-to-Video diffusion modelsleverage input images to generate visually stunning content, yet frequently produce motion that violates physical laws. We reveal a surprising finding: a 2-step generation often exhibits betterphysical consistencythan a 50-step output from the same model. Throughspectral analysis, we trace this tophase erosionduringdenoising; the phase degrades significantly (dropping by approx 18% from step 2 to step 50), whereas the magnitude remains relatively stable. Building on this insight, we propose PhaseLock, a training-free framework that preserves the validmotion priorsfrom few-step inference throughout thedenoisingtrajectory. Rather than relying on full-step inference forphysical consistency, PhaseLock extracts a motion prior from just 2 steps and enforces it onto high-fidelity generation viaLatent Delta Guidance. Our approach effectively mitigates phase degradation, improvingphysical consistencyby an average of 6.2 points across diverse models while largely maintainingvisual fidelity, with negligible overhead (1.06times time, 1.02times memory) and reduced reliance on expensive external guidance methods (sim5times time).

View arXiv pageView PDFProject pageGitHub2Add to collection

Get this paper in your agent:

hf papers read 2606\.06361

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.06361 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.06361 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.06361 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

PhysisForcing: Physics Reinforced World Simulator for Robotic Manipulation

Hugging Face Daily Papers

PhysisForcing is a training framework that enhances embodied video generation for robotic manipulation by enforcing physical consistency through pixel-level trajectory alignment and semantic-level relational alignment losses in a DiT-based architecture, achieving notable improvements on benchmarks.