aiqbench analysis
This research introduces Reward-Based Velocity Matching (RVM), a novel approach to fine-tuning diffusion models that bypasses complex likelihood-based policy gradients by directly optimizing the velocity field. The method achieves competitive or superior performance in reward alignment tasks while significantly reducing computational costs compared to existing trajectory-based techniques. By simplifying the update mechanism, RVM offers a more efficient and generalizable framework for adapting generative models to specific human preferences and dynamic tracking objectives.
Scaling Reinforcement Learning for Diffusion Models via Velocity Matching
arXiv:2608.23664v1 Announce Type: cross Abstract: Reward fine-tuning is becoming an important tool for adapting diffusion models to human preferences and task-specific objectives, but existing methods largely inherit policy-gradient machinery from large language models. Unlike autoregressive models, diffusion models do not provide tractable likelihoods for generated samples. As a result, current approaches either construct trajectory likelihoods from stochastic denoising transitions or approximate endpoint likelihoods with evidence lower bound, introducing additional computation and algorithmic complexity. We demonstrate that this likelihood-based machinery is not necessary for effective diffusion reward fine-tuning. We propose reward-based velocity matching (RVM), a simple trajectory-free update that acts directly on the velocity field. RVM reinforces directions associated with high-reward generations, suppresses those with low reward, and involves an optional anchor term controlling drift from a reference velocity. Notably, it provides a general framework that recovers recent fine-tuning methods, including RAM and DiffusionNFT, as special cases. Across various large-scale diffusion models reward fine-tuning tasks, RVM is competitive with or outperforms trajectory-based policy-gradient methods under substantially reduced training cost. We further find that, once the velocity update is simplified, the particular loss variant matters less than reward and anchor design. For video generation, standard preference rewards can favor visually clean but nearly static outputs; introducing a new dynamic-tracking reward that substantially improve motions while improving overall VBench performance. These results suggest that scalable reward fine-tuning for diffusion models is better posed in the native velocity representation than as likelihood-based policy optimization.
Abstract:Reward fine-tuning is becoming an important tool for adapting diffusion models to human preferences and task-specific objectives, but existing methods largely inherit policy-gradient machinery from large language models. Unlike autoregressive models, diffusion models do not provide tractable likelihoods for generated samples. As a result, current approaches either construct trajectory likelihoods from stochastic denoising transitions or approximate endpoint likelihoods with evidence lower bound, introducing additional computation and algorithmic complexity. We demonstrate that this likelihood-based machinery is not necessary for effective diffusion reward fine-tuning. We propose reward-based velocity matching (RVM), a simple trajectory-free update that acts directly on the velocity field. RVM reinforces directions associated with high-reward generations, suppresses those with low reward, and involves an optional anchor term controlling drift from a reference velocity. Notably, it provides a general framework that recovers recent fine-tuning methods, including RAM and DiffusionNFT, as special cases. Across various large-scale diffusion models reward fine-tuning tasks, RVM is competitive with or outperforms trajectory-based policy-gradient methods under substantially reduced training cost. We further find that, once the velocity update is simplified, the particular loss variant matters less than reward and anchor design. For video generation, standard preference rewards can favor visually clean but nearly static outputs; introducing a new dynamic-tracking reward that substantially improve motions while improving overall VBench performance. These results suggest that scalable reward fine-tuning for diffusion models is better posed in the native velocity representation than as likelihood-based policy optimization.
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2608.23664 [cs.CV]
(or arXiv:2608.23664v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2608.23664 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 335 words
Related coverage
Compositional Average Velocity Modeling for Efficient Diffusion Offline Reinforcement Learning
arXiv · Oct 14, 2026
Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics
Latent Space · Oct 13, 2026
A Survey of Flow Matching in Reinforcement Learning
arXiv · Oct 15, 2026
Build a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod
AWS · Sep 4, 2026