ReAlign: Text-to-Motion Generation via Step-Aware Reward-Guided Alignment

Wanjiang Weng^1,2*, Xiaofeng Tan^1,2*, Junbo Wang³, Guo-Sen Xie⁴, Pan Zhou⁵, Hongsong Wang^1,2†

¹Southeast University ²PALM Lab ³Northwestern Polytechnical University
⁴Nanjing University of Science and Technology ⁵Singapore Management University

* Equal Contribution. † Corresponding Author.

TL;DR: We propose ReAlign, a plug-and-play reward-guided alignment strategy for text-to-motion generation and retrieval, which explicitly enhances both semantic consistency and motion realism throughout the denoising process.

Paper arXiv AAAI 2026 Code Poster

Abstract

Text-to-motion generation, which synthesizes 3D human motions from text inputs, holds immense potential for applications in gaming, film, and robotics. Recently, diffusion-based methods have been shown to generate more diversity and realistic motion. However, there exists a misalignment between text and motion distributions in diffusion models, which leads to semantically inconsistent or low-quality motions. To address this limitation, we propose Reward-guided sampling Alignment (ReAlign), comprising a step-aware reward model to assess alignment quality during the denoising sampling and a reward-guided strategy that directs the diffusion process toward an optimally aligned distribution. This reward model integrates step-aware tokens and combines a text-aligned module for semantic consistency and a motion-aligned module for realism, refining noisy motions at each timestep to balance probability density and alignment. Extensive experiments of both motion generation and retrieval tasks demonstrate that our approach significantly improves text-motion alignment and motion quality compared to existing state-of-the-art methods.

Poster

Click the poster to view the PDF version

Toy Example

Reward-guided sampling in diffusion-based motion generation. The blue region represents the sampling distribution \( p_t(\cdot) \) learned by the diffusion model, while the green region depicts the ideal sampling distribution \( p_t^I(\cdot) \) achieved by incorporating our proposed reward-guided sampling strategy.

Main Results

Visual comparison of text-to-motion generation results. This figure presents motions generated by existing methods such as Mo.Diffuse, MDM, MLD, and MotionLCM. Observations reveal persistent mismatch with input texts and missing motion details. ReAlign enhances these models to generate motions more consistent with the text inputs.

Comparison of text-to-motion generation performance on the HumanML3D dataset. Percentages in subscripts indicate improvements over respective baselines.

Comparison on the KIT-ML dataset. Since the models MLD and MLD++ for KIT-ML have not been released, we use the widely used MDM as the baseline.

Plug-and-play functionality of ReAlign. Performance enhancement of motion generation methods with step-aware reward guidance on HumanML3D. MLCM¹/MLCM⁴ denote the 1-step/4-step model in MotionLCM. MDiff denotes MotionDiffuse.

Video Visualizations

More Visual Comparisons

More visual comparison on HumanML3D. ReAlign can be integrated into MLD to improve text-motion alignment and enhance motion quality. The red text denotes descriptions inconsistent with the generated motions.