Rethinking Classifier-Free Guidance in On-Policy Diffusion Distillation
2026-07-27 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial IntelligenceMachine Learning
AI summaryⓘ
The authors study how a method called on-policy distillation (OPD), used to improve diffusion models, works when combined with a technique called classifier-free guidance (CFG). They find that simply matching the outputs of teacher and student models can lead to problems because errors in different parts can cancel each other out incorrectly, a problem they name Negative Branch Asymmetry (NBA). To fix this, the authors propose a new approach called Positive-Direction Matching (PDM) that treats parts of the model separately, leading to better and more stable learning. They demonstrate that this new method works well in video applications where old methods struggled.
on-policy distillationdiffusion modelsclassifier-free guidancevelocity matchingpositive branchnegative branchNegative Branch AsymmetryPositive-Direction Matchingknowledge transfervideo control
Authors
Bingnan Li, Haozhe Wang, Haozhong Xiong, Fangtai Wu, Jinpeng Yu, Yang Shi, Jiaming Liu, Ruihua Huang
Abstract
On-policy distillation (OPD) adapts diffusion models by querying a teacher along trajectories generated by the current student, but how it should behave under classifier-free guidance (CFG), a default component of modern diffusion systems, remains poorly understood. Existing OPD methods naturally extend velocity matching to the CFG-composed prediction, directly matching teacher and student guided velocities. We show that this objective is under-identified at the branch level: positive- and negative-branch errors can compensate in the guided prediction. Through two contrasting cases, we find that naive matching remains effective under shared negative conditioning, where both branch errors decrease jointly. When the model's native CFG schema retains privileged information in the teacher's negative branch that is unavailable to the student, however, this joint reduction breaks down and the composed objective induces antagonistic branch-error dynamics, reducing the positive-branch error while increasing the negative-branch error. We term this failure mode Negative Branch Asymmetry (NBA). To address NBA, we introduce Positive--Direction Matching (PDM), a branch-aware OPD objective that separately constrains the positive prediction and the CFG conditional direction. We apply PDM to dense-to-sparse video control, where naive guided matching is highly sensitive to inference guidance scales, while branch-aware supervision enables more robust and effective knowledge transfer.