The Reward Was in Your Data All Along: Correcting Flow Matching with Discriminator-Guided RL
For practitioners of generative modeling, DRL offers a method to improve sample fidelity without relying on expensive human preferences, addressing a key bottleneck in aligning generative models with data realism.
Flow matching models suffer from a structural mismatch between their training loss (ℓ2 regression on velocity/score) and the visual/semantic properties that determine sample quality. Discriminator-Guided RL (DRL) uses a discriminator trained in a pretrained representation space to provide a reward signal that aligns with data realism, reducing guidance-free FID from 9.38 to 2.62 on SiT and improving semantic FD from 88.2 to 19.3 on DINOv3, with consistent gains across multiple backbones.
Score- and flow-matching models often rely on preference-based reinforcement learning for two purposes: aligning with subjective preferences and, surprisingly, recovering properties such as visual realism and coherent object structure that matching-based training is intended to learn from the data itself. We argue that this reflects a structural mismatch. Matching losses measure $\ell_2$ regression error on the velocity or score field under training-time marginals, a proxy poorly aligned with the visual and semantic properties that determine sample quality at inference. Given a reward aligned with these properties, RL sidesteps the mismatch by evaluating the model on its own samples and following the reward landscape directly. The challenge is to obtain such a reward without relying on human preferences, which are expensive and conflate data realism with annotator inclinations. We propose Discriminator-Guided RL (DRL). DRL trains a discriminator to separate data from base-model samples in a pretrained representation space and uses its logit as the reward in KL-regularized RL. The pretrained space restricts the discriminator to perceptually meaningful directions, and the logit estimates the log-likelihood ratio between data and model, which is the optimal reward for targeting the data distribution. Across SiT, JiT, REPA, and RAE, DRL reduces guidance-free FID (e.g., $9.38 \to 2.62$ on SiT) and semantic-space FD (e.g., $88.2 \to 19.3$ on DINOv3 for SiT), with consistent gains across all backbones, and improves human-preference rewards without training on them. It also yields a better Pareto frontier between preference reward and image fidelity under subsequent preference-based post-training, increasing alignment while reducing low-level artifacts such as oversaturation and excessive brightness.