MCLR: Improving Conditional Modeling via Inter-Class Likelihood-Ratio Maximization and Unifying Classifier-Free Guidance with Alignment Objectives

Diffusion models achieve strong performance in generative modeling, but their success relies heavily on classifier-free guidance (CFG), an inference-time heuristic that modifies the sampling trajectory. In theory, diffusion models trained with standard denoising score matching (DSM) should recover the target data distribution, raising the question of why inference-time guidance is necessary in practice. In this work, we argue that a key limitation of standard DSM is insufficient inter-class separation. Motivated by this view, we propose MCLR, an alignment objective that explicitly maximizes inter-class likelihood-ratios during training. Fine-tuning diffusion models with MCLR leads to CFG-like improvements under standard sampling, substantially improving conditional generation quality without requiring inference-time guidance. Beyond empirical benefits, we show theoretically that the CFG-guided score is exactly the optimal solution to a weighted MCLR objective. This result connects CFG to alignment-based objectives, providing a mechanistic interpretation of CFG.