The paper introduces DiffusionOPSD, an on‑policy self‑distillation framework that transforms image‑level reinforcement learning rewards into explicit targets for intermediate denoising predictions in diffusion models. By generating trajectories with a frozen behavior policy and constructing bounded positive and negative targets around query states, the method trains a policy to fit these targets before updating the behavior policy via an exponential moving average. Experiments on SD 3.5‑M and Z‑Image‑Turbo show that DiffusionOPSD achieves the best held‑out scores in 19 of 20 reward‑matched settings, outperforms the strongest competitor by up to 44 % and cuts GPU‑hour usage by 40–63 % compared to DiffusionNFT.
By Wei Zhou, Xiongwei Zhu, Lingdong Kong, Bo Chen, Lei Zhang, Yongyuan Liang, Xiaoxia Hou, Ye Tian, Xian Sun, Yingshuo Wang, Linfeng Li, Shengqiong Wu, Leigang Qu, Feng Li, Wei Liu, Julian McAuley, Tat-Seng Chua
Reinforcement learning can align diffusion models with human preferences and task-specific objectives, but endpoint rewards do not specify how an intermediate denoising prediction should change. We in...
arXiv:2604. 26985v2 Announce Type: replace-cross Abstract: Masked diffusion models (MDMs) generate discrete sequences by iterative denoising under an absorbing masking process.
By Michael Cardei, Huu Binh Ta, Ferdinando Fioretto
arXiv:2607. 05319v1 Announce Type: cross Abstract: We study why diffusion autoencoders can achieve similar image quality while learning substantially different latent structures.
By Rajat Rasal, Avinash Kori, Tian Xia, Ben Glocker
arXiv:2609.37974v1 Announce Type: cross
Abstract: Masked diffusion models (MDMs) generate text by unmasking several tokens per step, but they are trained and sampled under different conditions. The m...
By Manuel Madeira, Amitis Shidani, Alice Bizeul, Victor Turrisi, Louis B\'ethune, Bhavika Devnani, Dan Busbridge, Pierre Ablin, Jo\~ao Monteiro
CAST introduces a reinforcement‑learning fine‑tuning framework for diffusion models that addresses three key limitations: it automatically selects the denoising window based on each model’s trajectory, decomposes prompts into verifiable semantic atoms via Causal Scene Graphs, and applies atom‑level rewards spatially weighted in the policy objective. The method is applied to FLUX.2‑dev and Qwen‑Image‑2512, yielding up to 3.07× improvement on the hardest GenEval 2 prompts compared with Flow‑GRPO while also enhancing overall generation quality.
By Shu Yu, Chaochao Lu
The paper introduces Learned End-to-End Guidance Schedules (LEEGS) for diffusion models, which train a time‑dependent guidance schedule to balance data quality and requirement satisfaction while reducing sampling steps. LEEGS minimizes the guidance function over a small set of examples using stochastic gradient descent and employs a gradient approximation to cut training time by a factor of four. Experiments on tasks such as image inpainting, noisy image inverse problems, face‑ID‑guided generation, and PDE problems show that LEEGS outperforms baselines at the same computational budget or matches constant guidance with only 10% of the steps.
By Aneesh Barthakur, Mathias Niepert, Luiz F. O. Chamon
arXiv:2606. 28094v1 Announce Type: cross Abstract: Real-world object removal is challenging due to two key difficulties: the target object's non-local effects, such as shadows and reflections, which are difficult to model, and the fact that user-provided masks are often inaccurate or incomplete.
By Qinming Zhou, Chenxi Sun, Deyang Kong, Junhao He, Xiangheng Tang, Peike Yu, Haotian Wu, Leilei Cao, Linfeng Zhang
The paper introduces the first controlled benchmark for optimizers in discrete diffusion models, evaluating seven optimizers (AdamW, Lion, Muon, SOAP, MARS, MARS‑M, Schedule‑Free) across four diffusion formulations: masked diffusion on text8, uniform diffusion on QM9 and LM1B, and Gaussian diffusion on CelebA‑64. Each optimizer undergoes the same search protocol and is retrained with full budget and multiple seeds, revealing that AdamW, while strong, is not universally optimal and that optimizers validated on autoregressive language models (Muon, MARS‑M, SOAP) can outperform tuned AdamW on certain tasks.
By Arman Bolatov, Egor Shulgin, David Li, Abduragim Shtanchaev, Sebastian U. Stich, Maxim Panov, Eric Moulines, Peter Richt\'arik, Martin Tak\'a\v{c}
arXiv:2606. 23898v1 Announce Type: cross Abstract: Distilling conditional diffusion models aims to transfer the behavior of a large teacher to a smaller student while preserving alignment across conditioning inputs.
By Loay Mualem, Vinh Tong, Samir Darouich, Mathias Niepert
The paper presents a preference-driven unlearning framework for few-step distilled text-to-image diffusion models. It identifies that standard Direct Preference Optimization (DPO) fails on these models due to altered generation dynamics and proposes a modified formulation aligned with few-step properties. The method enables direct concept removal while preserving desirable capabilities, and is evaluated on identity, NSFW, and object-level unlearning tasks with strong forgetting and retention results.
By Gaurav Patel, Jun Fang, Greg Ver Steeg, Qiang Qiu, Sravan Sripada
The paper introduces TRACK, a training‑free trajectory routing method that accelerates video diffusion by selectively switching between large and small models during denoising steps. A calibration process generates a disagreement score map, guiding the selection of the appropriate model at each step to maintain quality while reducing computational cost. Experiments on Wan 2.1, Cosmos 3, TurboDiffusion, and FastVideo show speedups ranging from 1.95× to 2.73× with comparable quality and diversity.
By Mustafa Munir, Huy Vu, Shreyas Misra, Rohit Jena, Sajad Norouzi, Ali Taghibakhshi, Anis Ahmad, Anjul Patney, Pavlo Molchanov, Nima Tajbakhsh