Masked generative models (MGMs) are widely used to capture complex data and enable faster generation than autoregressive models (AR) through parallel decoding. However, MGMs typically operate on fixed-length inputs, which can be inefficient: early in sampling, most tokens are masked and carry no information, leading to wasted computation. In contrast, AR models process only tokens generated previously, making early iterations faster. In this work, we introduce the Partition Generative Model (PGM), a novel approach that combines the strengths of AR and MGMs. Rather than masking, PGM partitions tokens into two groups and employs sparse attention to block information flow between them. Since there is no information flow between partitions, the model can process the previously-generated tokens only during sampling, while retaining the ability to generate tokens in parallel and in any order. On OpenWebText, PGMs offer at least $5\times$ improvements in sampling latency and throughput, while producing samples with superior Generative Perplexity, compared to Masked Diffusion Language Models. On ImageNet, PGMs achieve a $7.5\times$ higher throughput than MaskGIT, with only a slight increase in FID (5.54 vs. 5.35). With twice as many sampling steps, the FID reduces to 4.56 while while being $3.9\times$ faster than MaskGIT. Finally, PGMs integrate seamlessly with MGM distillation, providing further inference speedups.

Generative models have witnessed remarkable progress in recent years, with diffusion models (Sohl-Dickstein et al., 2015) emerging as a dominant paradigm across various domains including images (Ho et al., 2020; Dhariwal & Nichol, 2021), audio (Kong et al., 2021) and video (Ho et al., 2022). While continuous diffusion models have garnered significant attention, discrete diffusion models (Austin et al., 2021; Campbell et al., 2022; Gu et al., 2022; Lou et al., 2024) offer compelling advantages for inherently discrete data, such as tokenized representations in all the above domains. Among discrete diffusion approaches, masked diffusion models (Sahoo et al., 2024; Shi et al., 2024; Ou et al., 2025) have demonstrated exceptional performance, particularly in generating high-quality samples in language domains. However, their sampling process remains computationally intensive, requiring hundreds of function evaluations. Recent efforts to accelerate sampling in discrete/masked diffusion models have explored various directions, including optimized scheduling (Park et al., 2025), distillation (Deschenaux & Gulcehre, 2025; Hayakawa et al., 2025; Zhu et al., 2025), and adaptive token selection (Kim et al., 2025; Ben-Hamu et al., 2025).

Compositional generalization, the ability to generate novel combinations of known concepts, is a key ingredient for visual generative models. Yet, not all mechanisms that enable or inhibit it are fully understood. In this work, we conduct a systematic study of how various design choices influence compositional generalization in image and video generation in a positive or negative way. Through controlled experiments, we identify two key factors: (i) whether the training objective operates on a discrete or continuous distribution, and (ii) to what extent conditioning provides information about the constituent concepts during training. Building on these insights, we show that relaxing the MaskGIT discrete loss with an auxiliary continuous JEPA-based objective can improve compositional performance in discrete models like MaskGIT.

Discrete Flow-based Models (DFMs) are powerful generative models for high-quality discrete data but typically suffer from slow sampling speeds due to their reliance on iterative decoding processes. This reliance on a multi-step process originates from the factorization approximation of DFMs, which is necessary for handling high-dimensional data. In this paper, we rigorously characterize the approximation error from factorization using Conditional Total Correlation (TC), which depends on the coupling. To reduce the Conditional TC and enable efficient few-step generation, we propose Rectified Discrete Flow (ReDi), a novel iterative method that reduces factorization error by rectifying the coupling between source and target distributions. We theoretically prove that each ReDi step guarantees a monotonic decreasing Conditional TC, ensuring its convergence. Empirically, ReDi significantly reduces Conditional TC and enables few-step generation. Moreover, we demonstrate that the rectified couplings are well-suited for training efficient one-step models on image generation. ReDi offers a simple and theoretically grounded approach for tackling the few-step challenge, providing a new perspective on efficient discrete data synthesis. Code is available at https://github.com/Ugness/ReDi_discrete

Although masked image generation models and masked diffusion models are designed with different motivations and objectives, we observe that they can be unified within a single framework. Building upon this insight, we carefully explore the design space of training and sampling, identifying key factors that contribute to both performance and efficiency. Based on the improvements observed during this exploration, we develop our model, referred to as eMIGM. Empirically, eMIGM demonstrates strong performance on ImageNet generation, as measured by Fr\'echet Inception Distance (FID). In particular, on ImageNet 256x256, with similar number of function evaluations (NFEs) and model parameters, eMIGM outperforms the seminal VAR. Moreover, as NFE and model parameters increase, eMIGM achieves performance comparable to the state-of-the-art continuous diffusion models while requiring less than 40% of the NFE. Additionally, on ImageNet 512x512, with only about 60% of the NFE, eMIGM outperforms the state-of-the-art continuous diffusion models.

Learning world models can teach an agent how the world works in an unsupervised manner. Even though it can be viewed as a special case of sequence modeling, progress for scaling world models on robotic applications such as autonomous driving has been somewhat less rapid than scaling language models with Generative Pre-trained Transformers (GPT). We identify two reasons as major bottlenecks: dealing with complex and unstructured observation space, and having a scalable generative model. Consequently, we propose a novel world modeling approach that first tokenizes sensor observations with VQVAE, then predicts the future via discrete diffusion. When applied to learning world models on point cloud observations, our model reduces prior SOTA Chamfer distance by more than 65% for 1s prediction, and more than 50% for 3s prediction, across NuScenes, KITTI Odometry, and Argoverse2 datasets. Our results demonstrate that discrete diffusion on tokenized agent experience can unlock the power of GPT-like unsupervised learning for robotic agents. Figure 1: Our unsupervised world model can produce accurate near-term 1s predictions and diverse multi-future 3s predictions directly on the level of point cloud observations. World models explicitly represent the knowledge of an autonomous agent about its environment. They are defined as a generative model that predicts the next observation in an environment given past observations and the current action. Such a generative model can learn from any unlabeled agent experience, and can be used for both learning and planning in the model-based reinforcement learning framework (Sutton, 1991). This approach has excelled in domains such as Atari (Kaiser et al., 2019), robotic manipulation (Nagabandi et al., 2020), and Minecraft (Hafner et al., 2023). Learning world models can be viewed as a special case of sequence modeling on agent experience. While Generative Pre-trained Transformers (GPT) (Brown et al., 2020) have enabled rapid progress Prediction systems in autonomous driving still require supervised learning, either on the level of bounding boxes (Luo et al., 2018), semantic segmentation (Sadat et al., 2020), or instance segmentation (Hu et al., 2021). However, just as GPT learns to understand language via next token prediction, if a world model can predict unlabeled future observations really well, it must have developed a general understanding of the scene including geometry and dynamics.

We propose to replace vector quantization (VQ) in the latent representation of VQ-VAEs with a simple scheme termed finite scalar quantization (FSQ), where we project the VAE representation down to a few dimensions (typically less than 10). Each dimension is quantized to a small set of fixed values, leading to an (implicit) codebook given by the product of these sets. By appropriately choosing the number of dimensions and values each dimension can take, we obtain the same codebook size as in VQ. On top of such discrete representations, we can train the same models that have been trained on VQ-VAE representations. For example, autoregressive and masked transformer models for image generation, multimodal generation, and dense prediction computer vision tasks. Concretely, we employ FSQ with MaskGIT for image generation, and with UViM for depth estimation, colorization, and panoptic segmentation. Despite the much simpler design of FSQ, we obtain competitive performance in all these tasks. We emphasize that FSQ does not suffer from codebook collapse and does not need the complex machinery employed in VQ (commitment losses, codebook reseeding, code splitting, entropy penalties, etc.) to learn expressive discrete representations. Vector quantization (VQ), initially introduced by Gray (1984), has recently seen a renaissance in the context of learning discrete representations with neural networks. Spurred by the success of VQ-VAE (Van Den Oord et al., 2017), Esser et al. (2020) and Villegas et al. (2022) showed that training an autoregressive transformer on the representations of a VQ-VAE trained with a GAN loss enables powerful image and video generation models, respectively.

In recent years, generative modeling has made significant advancements. The mainstream frameworks for generative image modeling have evolved through various phases, initially utilizing Variational AutoEncoder (VAE) [1], then progressing to Generative Adversarial Network (GAN) [2], and eventually Vector Quantized-Variational AutoEncoder (VQ-VAE) [3] and diffusion models [4]. Masked generative modeling on VQ-tokens and diffusion models have demonstrated state-of-the-art (SOTA) performance in the last couple of years on image modeling, audio modeling, and time series modeling [5, 6, 7, 8, 9, 10]. Both masked generative modeling and diffusion models utilize sampling schemes to iteratively unmask/predict tokens or denoise noisy inputs. In recent years, diffusion models' sampling schemes have garnered considerable attention [11, 12]. While the sampling scheme for masked generative modeling has not received the same level of attention, notable progress has been made in [13, 14, 10]. This paper identifies certain limitations of the masked generative modeling sampling schemes in [13, 14, 10], and introduces a novel and improved sampling scheme which we call Enhanced Sampling Scheme (ESS) to address these limitations. ESS comprises three stages: 1) Iterative non-autoregressive decoding, as proposed in [13], 2) Critical Reverse Sampling, and 3) Critical Resampling using self-Token-Critic. Figure 1 provides an overview of the method, and Sect. 3 presents a detailed explanation of ESS.

Recent conditional image generation methods produce images of remarkable diversity, fidelity and realism. However, the majority of these methods allow conditioning only on labels or text prompts, which limits their level of control over the generation result. In this paper, we introduce MaskSketch, an image generation method that allows spatial conditioning of the generation result using a guiding sketch as an extra conditioning signal during sampling. MaskSketch utilizes a pre-trained masked generative transformer, requiring no model training or paired supervision, and works with input sketches of different levels of abstraction. We show that intermediate self-attention maps of a masked generative transformer encode important structural information of the input image, such as scene layout and object shape, and we propose a novel sampling method based on this observation to enable structure-guided generation. Our results show that MaskSketch achieves high image realism and fidelity to the guiding structure. Evaluated on standard benchmark datasets, MaskSketch outperforms state-of-the-art methods for sketch-to-image translation, as well as unpaired image-to-image translation approaches.

Converting an image of an object to a sketch. Reproduced under a CC BY-NC-SA 4.0 license. Here are the most tweeted papers that were uploaded onto arXiv during February 2022. Results are powered by Arxiv Sanity Preserver. Abstract: We present Block-NeRF, a variant of Neural Radiance Fields that can represent large-scale environments.