arXiv:2606. 02521v1 Announce Type: new Abstract: One-step text-to-image generators are attractive for deployment because they generate an image with a single forward pass, but preference finetuning them remains difficult: standard alignment methods often rely on policy likelihoods, denoising trajectories, differentiable reward gradients, or test-time optimization.
One-step text-to-image generators are attractive for deployment because they generate an image with a single forward pass, but preference finetuning them remains difficult: standard alignment methods often rely on policy likelihoods, denoising trajectories, differentiable reward gradients, or test-time optimization. We propose Drifting Preference Optimization (DrPO), an online preference-finetuning method for deterministic one-step generators.
arXiv:2604. 17415v3 Announce Type: replace-cross Abstract: Reward-based fine-tuning steers a pretrained diffusion or flow-based generative model toward higher-reward samples while remaining close to the pretrained model.
arXiv:2607. 02291v1 Announce Type: new Abstract: Conventional reinforcement learning strategies for visual generation typically employ sample-wise reward functions, yet this practice frequently results in reward hacking that degrades image diversity and introduces visual anomalies.
arXiv:2608. 09226v1 Announce Type: cross Abstract: Efficient text-to-image generation requires both reinforcement-learning (RL)-based reward alignment and few-step distillation, yet these procedures are typically performed sequentially, increasing training cost and risking the loss of reward gains during compression.
Efficient text-to-image generation requires both reinforcement-learning (RL)-based reward alignment and few-step distillation, yet these procedures are typically performed sequentially, increasing training cost and risking the loss of reward gains during compression. We instead take an RL-native perspective: diffusion RL already generates reward-scored finite-step trajectories, whose intermediate states provide a natural source of distillation supervision rather than a disposable byproduct of sampling.
arXiv:2607. 11505v2 Announce Type: replace-cross Abstract: Post-training for large language models typically couples policy exploration with model optimization, hindering the reuse of high-reward behaviors from policy exploration.
The paper introduces Wasserstein‑Tilted Flow Maps (WTF), a simulation‑free reinforcement learning method that fine‑tunes pre‑trained flow‑based generative models by adding an optimal transport regularizer derived from the model’s drift. Unlike traditional KL‑reward tilting, WTF transports individual samples toward higher reward, framing the problem as a deterministic optimal control task on the flow map. Experiments on ImageNet‑256 and text‑to‑image demonstrate that WTF achieves higher reward and comparable or better diversity while reducing training compute by up to 280×.
arXiv:2608. 03929v1 Announce Type: new Abstract: Aligning diffusion models with human preferences usually relies on a sparse terminal reward evaluated on the final generated samples, presenting a severe temporal credit-assignment challenge across the multi-step denoising process.
The paper introduces Reward-Aware Trajectory Shaping (RATS), a lightweight framework that aligns teacher and student latent trajectories at key denoising stages while adaptively regulating teacher guidance through a reward-aware gate. By allowing the student to optimize toward reward-preferred generation quality, RATS can surpass the teacher rather than merely imitate it. Experiments show that RATS improves the efficiency–quality trade-off in few-step visual generation, narrowing the gap between few-step students and stronger multi-step generators without extra test-time cost.
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 method...
arXiv:2606. 19162v1 Announce Type: new Abstract: 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.