Current research in generative models is increasingly focused on enhancing sample quality and diversity while addressing inherent biases and inefficiencies. Recent advancements have introduced innovative frameworks that refine generative outputs without the need for extensive noise injection or complex resampling processes, thereby improving fidelity and coverage in high-dimensional data. Techniques such as Condition-Degradation Guidance and instance-aware discretization are being developed to optimize the generative process, allowing for more precise control over output semantics and better adaptation to input complexities. Additionally, the integration of reinforcement learning paradigms is enabling generative models to leverage non-differentiable rewards, which are crucial for real-world applications. These developments have significant implications for industries ranging from urban planning to robotics, where the ability to generate high-quality, diverse synthetic data can enhance model training and decision-making processes. As the field matures, the emphasis is shifting toward creating generative systems that are not only efficient but also robust and adaptable to various practical challenges.
Decomposing complex data into factorized representations can reveal reusable components and enable synthesizing new samples via component recombination. We investigate this in the context of diffusion...
Classifier-Free Guidance (CFG) is a cornerstone of modern text-to-image models, yet its reliance on a semantically vacuous null prompt ($\varnothing$) generates a guidance signal prone to geometric en...
Generative models, including diffusion and flow-based models, often exhibit systematic biases that degrade sample quality, particularly in high-dimensional settings. We revisit refinement methods and ...
Recent advancements in Generative Reward Models (GRMs) have demonstrated that scaling the length of Chain-of-Thought (CoT) reasoning considerably enhances the reliability of evaluation. However, curre...
While few-step generative models have enabled powerful image and video generation at significantly lower cost, generic reinforcement learning (RL) paradigms for few-step models remain an unsolved prob...
We establish a theoretical link between the recently proposed "drifting" generative dynamics and gradient flows induced by the Sinkhorn divergence. In a particle discretization, the drift field admits...
Rectified Flow (RF) models achieve state-of-the-art generation quality, yet controlling them for precise tasks -- such as semantic editing or blind image recovery -- remains a challenge. Current appro...
Latent diffusion models (LDMs) enable high-fidelity synthesis by operating in learned latent spaces. However, training state-of-the-art LDMs requires complex staging: a tokenizer must be trained first...
We propose Ambient Dataloops, an iterative framework for refining datasets that makes it easier for diffusion models to learn the underlying data distribution. Modern datasets contain samples of highl...
Flow Matching enables simulation-free training of generative models on Riemannian manifolds, yet sampling typically still relies on numerically integrating a probability-flow ODE. We propose Riemannia...