Diffusion Transformer

State-of-the-art single-image 3D reconstruction methods often rely on complex hybrid architectures and loss functions, or compress geometry into latent spaces in order to leverage pre-trained latent diffusion models. In this work, we show that such architectural overhead and intricate loss formulations are unnecessary. We introduce a minimalist pixel-space Diffusion Transformer, built on a plain ViT, that operates directly on raw 3D point map patches and is conditioned on image tokens from a pre-trained DINOv3. Unlike existing latent diffusion approaches, we train our diffusion backbone entirely from scratch, eliminating the need for point map tokenizers. Despite its simplicity, our approach surpasses complex latent-based diffusion models while remaining significantly simpler than hybrid alternatives. Notably, it produces sharper geometric structure and is more robust in highly ambiguous regions, such as transparent objects.

Representation alignment has become an effective way to accelerate diffusion transformer training and improve generation quality. Recent self-alignment methods, such as SRA and Self-Flow, further remove the dependency on external pretrained encoders by constructing alignment within the diffusion model itself. However, the mechanism behind the improvement from SRA to Self-Flow, dual-time scheduling, remains under-examined: Self-Flow attributes its gain to interactions between tokens at different noise levels, where cleaner tokens help infer noisier ones. In this work, we revisit this explanation and ask whether the gain instead comes from data augmentation along the noise dimension. To disentangle these factors, we introduce Attention Separation, which preserves the same dual-timestep input as Self-Flow while blocking attention between tokens assigned to different noise levels. Surprisingly, removing such interaction does not degrade performance and can even improve it, suggesting that the improvement from SRA to Self-Flow mainly comes from data augmentation. Furthermore,We show that Attention Separation itself provides an augmentation effect by splitting a single image into multiple effective training parts to expand the training data. Based on these observations, we combine self-representation alignment with dual-timestep and attention-separation augmentation, and demonstrate the effectiveness of this design on ImageNet.

Diffusion transformers (DiTs) achieve state-of-the-art image and video generation, but their multi-step sampling and growing parameter count make inference expensive. Post-training quantization (PTQ) is the natural remedy, yet DiT activations shift across timesteps, prompts, and guidance branches, forcing prior methods to re-fit calibration data for every new checkpoint or modality. We present OrbitQuant, a data-agnostic weight-activation quantizer that bypasses range estimation by quantizing in a normalized, rotated basis. In this basis, a randomized permuted block-Hadamard (RPBH) rotation concentrates each coordinate around one fixed, known marginal regardless of the input, so a single Lloyd-Max codebook serves all timesteps, prompts, and layers of a given input dimension. We extend the same quantizer to weight rows offline, absorbing the rotation into the weights so that it cancels inside each linear layer and only a forward rotation on the activations remains at runtime. The same recipe transfers from image to video with no per-modality tuning. Across FLUX.1, Z-Image-Turbo, Wan 2.1, and CogVideoX, it sets the state of the art for PTQ at several low-bit settings. It also pushes PTQ of image diffusion transformers to W2A4 with usable generation quality.

Video diffusion transformers (DiTs) generate high-fidelity and temporally coherent videos, yet motion control remains implicit, primarily relying on text prompts. As a result, achieving desired motion often requires extensive prompt engineering and repeated resampling. While fine-tuning models with additional spatial prompts (e.g., bounding boxes or point trajectories) enables explicit control, it demands substantial data curation and computation, and may compromise the generative capabilities of pretrained models. Consequently, training-free motion control using such spatial prompts has been explored in U-Net-based video diffusion models, but remains largely unexplored for DiTs. We introduce QWERTY, a training-free framework that enables flexible motion control in pretrained image-to-video DiTs via user-defined object warping and optical flow. We carefully manipulate the 3D full attention of DiTs by warping the frame-invariant semantic subspace of queries. We find that the noise predicted by the query-warped DiT naturally guides the diffusion trajectory toward the desired motion, and further show that leveraging this noise as self-guidance for latent optimization improves control stability and visual quality. Experiments show that QWERTY achieves the most effective motion control among existing training-free approaches on a recent image-to-video DiT, with performance comparable to fine-tuning-based methods.

Text-conditioned human interaction generation must capture both long-range temporal causality within each individual and tightly coupled coordination between partners. Existing interaction diffusion models typically denoise full sequences using bidirectional attention, which obscures causality and hinders streaming and long-horizon generation. Autoregressive alternatives enforce causality but often suffer from temporal drift, leading to coordination degradation and unstable interaction dynamics over time. We propose InterCMDM, a block-causal latent diffusion framework for autoregressive two-person interaction generation. InterCMDM introduces a Dual-Stream Causal Diffusion Transformer that maintains separate causal streams for each person while modeling inter-person dependencies via unified dual-stream attention with multi-task attention masks. These masks unify interaction modeling within a single attention mechanism and support diverse coordination behaviors, including simultaneous actions, reactive responses, leader-follower dynamics, and independent motion. By training a single model across these mask configurations as a form of data augmentation, InterCMDM enables controllable interaction generation by simply selecting the desired attention mask at inference time. Finally, a block-wise diffusion objective enables stable latent rollout over long sequences without repeated decode-encode cycles. InterCMDM achieves state-of-the-art performance on InterHuman and Inter-X, improving text-motion alignment, realism, and long-horizon continuity.

Diffusion Transformers (DiTs) have demonstrated impressive performance in image generation but suffer from substantial computational overhead and resource consumption. Post-training pruning offers a promising solution; however, due to DiTs' unique architectural design and parameter distribution, traditional pruning methods are inapplicable, leading to significant performance degradation. Specifically, prior methods developed for LLMs, which derive metrics through a series of approximations, amplify the relative contribution of weights in the saliency metric. In addition, weights in DiTs exhibit significantly larger magnitudes than those in LLMs. Moreover, existing pruning granularity overlooks variations in model structures. In this paper, we propose DiT-Pruning, which improves pruning performance by introducing customized saliency criteria and pruning granularity. We design a novel metric that balances the contributions of weights and activations from an energy-based perspective, enabling more effective identification of important elements. Furthermore, we observe distinct clustering patterns in the two-dimensional weight space. Accordingly, we adopt a clustering-aware pruning granularity, enabling effective sparse allocation. Extensive evaluations on various DiTs show that our method consistently preserves image quality, especially under high sparsity. For FLUX.1-dev at 512x512 resolution on MJHQ, DiT-Pruning achieves only a 0.001 loss in CLIP score at 50% sparsity, dramatically outperforming recent pruning methods.

We propose the first compression approach for image-to-shape Diffusion Transformers (DiTs) that substantially reduces model size while preserving geometric fidelity. Despite remarkable progress in 3D shape generation, large DiT-based models remain computationally prohibitive in resource-constrained settings. Furthermore, it is difficult to directly transfer existing diffusion model compression strategies developed for different domains to 3D generation, and prior 3D efficiency approaches focus primarily on inference speed rather than backbone compression. To address this limitation, we build a geometry-aware compression framework tailored to image-to-shape DiTs. Guided by the observation that 3D DiT layers exhibit non-uniform importance for geometry synthesis, we introduce a vitality-guided framework integrating structured pruning, adaptive quantization, and targeted fine-tuning. Our method achieves up to 66% model-size reduction across state-of-the-art image-to-3D models while maintaining synthesis fidelity comparable to full-sized counterparts. This highlights the potential of our framework as a plug-and-play solution for efficient 3D shape generation across diverse models.

We present a zero-shot, training-free and optimization-free framework for generating 360 panoramic images and videos by directly injecting spherical priors into pre-trained diffusion transformers. Existing methods either rely on costly fine-tuning on scarce panoramic data that limits generalization, or leverage multi-step optimization that incurs prohibitive inference latency. We observe that contemporary generative models natively exhibit some panoramic priors from large-scale training. However, these emergent capabilities are insufficient, as the models fundamentally fail to satisfy the rigorous topological constraints imposed by equirectangular projection (ERP). We introduce a zero-shot and optimization-free approach that resolves these constraints at inference time. Spherical RoPE replaces standard rotary position embeddings: low-frequency channels are re-parameterized as 3D Cartesian coordinates to natively encode the spherical manifold, while high-frequency channels are harmonically quantized to enforce exact periodicity. Coupled with complementary Semantic Distortion classifier-free guidance (CFG) that explicitly steers geometry, we avoid retraining and inherit the full creative breadth of state-of-the-art models. Our approach generalizes across diverse backbones and 360 generation modalities. We demonstrate this across text-to-panorama using Flux.1, Flux.2, and LTX-Video backbones, achieving competitive performance against baselines, all while remaining training-free. Project page: https://orhir.github.io/SpheRoPE

Modern text-to-image diffusion models, such as diffusion transformers (DiT), rely on timestep or prompt embeddings to modulate the strength of the denoising process in each timestep. While this modulation communicates the current noise level, it does not provide any quality-aware information, which can lead to generated images that are unaligned, visually inconsistent, and lacking in fidelity. In this paper, we propose the Quality Representation Module (QRM), a lightweight transformer module that learns a quality-aware representation based on existing model inputs, and produces a set of vectors $M_{qrm}$. These vectors adjust the adaptive LayerNorm modulation within the DiT transformer blocks, thereby injecting a quality-sensitive signal into the denoising parameters. The QRM introduces no significant changes to the sampling schedule or diffusion backbone. Experiments include ablations on QRM training losses and architectures, as well as empirical results demonstrating consistent image quality improvements over baseline DiT-based models.

Diffusion Transformers (DiTs) have significantly advanced audio-driven portrait animation, but their high computational cost leads to substantial inference latency. Although training-free diffusion caching accelerates inference significant, existing methods are primarily developed for text-conditioned generation and overlook the spatial and modality imbalances inherent in audio-driven portrait animation. In this paper, we propose SyncCache, a training-free caching acceleration method tailored for DiT-based portrait animation that explicitly exploits asymmetric dynamics. Specifically, high-frequency dynamics driven by audio conditions and concentrated in human regions are more challenging and critical to cache and reuse than the low-frequency visual background in portrait animation. First, we introduce Spatially-Asymmetric Probing to prioritize error sensitivity in dynamic human region. Second, through Modality-Decoupled Caching, we bypass heavy DiT block by reusing stable inter-block residuals, while continuously recomputing lightweight audio blocks to preserve precise lip synchronization. Furthermore, we introduce a cache ratio to control cache capacity and formulate memory-adaptive cache selection as an offline dynamic programming problem without online overhead. Extensive experiments demonstrate that SyncCache achieves superior speed-quality trade-offs, delivering up to 4.12x acceleration on HunyuanVideo-Avatar and 3.75x on Wan-S2V with near-lossless visual fidelity and precise audio alignment.

Recent advances in diffusion models have shown impressive performance in controllable image generation and dense prediction tasks. However, existing approaches typically treat diffusion-based controllable generation and dense prediction as separate tasks, overlooking the potential benefits of jointly modeling the heterogeneous distributions. In this work, we introduce UniGP, a framework built upon MMDiT, which unifies controllable generation and dense prediction through simple joint training, without the need for complex task-specific designs or losses, while preserving the backbone's versatile priors. By learning controllable generation and prediction under different conditions, our model effectively captures the joint distribution of image-geometry pairs. UniGP is capable of versatile controllable generation, dense prediction, and joint generation. Specifically, the proposed UniGP consists of DUGP and a unified dataset training strategy. The former, following the principle of Occam's razor, uses only a copied image branch of MMDiT to model dense distributions beyond RGB, while the latter integrates heterogeneous datasets into a unified training framework to jointly model generation and perception tasks. Extensive experiments demonstrate that our unified model surpasses prior unified approaches and performs on par with specialized methods. Furthermore, we demonstrate that multi-task joint training provides complementary benefits: generative priors enrich perceptual details, while perceptual learning improves structural alignment in generation.

High-resolution Diffusion Transformer (DiT) inference contains substantial spatial redundancy, but many spatially adaptive implementations encode regional computation as attention masks, which can inadvertently move scaled dot-product attention (SDPA) away from FlashAttention fast paths. We identify this avoidable systems bottleneck as Mask-Induced Dispatch Tax (MIDT) and show that it grows with latent sequence length. We introduce SAFE-DiT, a training-free Semantics-Aware Fast-path Execution framework that separates exact mask elision from approximation-based spatial scheduling. SAFE-DiT removes only provenance-certified image self-attention masks that induce a row-wise constant shift in attention logits, preserves semantics-bearing masks such as text-padding masks, and realizes spatial adaptation through prompt-conditioned token partitioning, selective state updates with global context, and periodic context refresh. We call this acceleration-only configuration SAFE-Core and report sensitivity-weighted classifier-free guidance separately as SAFE-DiT+SW. On the evaluated PyTorch SDPA stack, redundant masks make long-sequence attention $4.1\times$ to $5.8\times$ slower than the mask-free path. On Lumina-Next, SAFE-DiT achieves $2.69\times$ end-to-end acceleration at $1024^2$ resolution and $5.09\times$ at $2560^2$, reduces peak memory at $2560^2$ from 94.1 to 27.9 GB, and enables $3072^2$ generation when dense inference runs out of memory. Paired metrics, component ablations, and a blinded human study support visual non-inferiority of SAFE-Core to the dense fast-path baseline, while SAFE-DiT+SW provides a separate prompt-alignment operating point without reintroducing spatial self-attention masks. Code is available at https://github.com/xuanhuayin/SAFE-DiT.

As the most perceptually powerful Face Video Super-Resolution (FVSR) method, existing works in Generative FVSR (GFVSR) mainly exploit the generative prior of pretrained diffusion models. However, viewed as full generation, they suffer from fixed sampling and expensive inference costs if without large-scale auxiliary training. Furthermore, an excessive pursuit of generic perceptual metrics often results in low fidelity. To address these issues, we present Dynamic Trajectory Initialization (DTI) paradigm for GFVSR, which reformulates GFVSR as an input-driven directional restoration. With a novel enhancement-and-injection conditioning mechanism for pretrained DiT backbone, fidelity of our model has been significantly improved without compromising perceptual quality. To dynamically set the starting sampling point, we propose a Discriminative Guide (DG) trained via objective Signal-to-Noise Ratio (SNR) alignment. With only minor model adaptation and fine-tuning, our method achieves a SOTA overall performance across diverse metrics and benchmarks. An analysis of relationship between actual comprehensive quality and common metrics is also conducted, which demonstrates the perception-distortion trade-off and that the LPIPS is the most convincing metric in our case.

End-to-end pixel-space diffusion models bypass the lossy compression of Latent Diffusion Models (LDMs) but struggle to jointly model low-frequency semantics and high-frequency signals in high-dimensional space. Existing works heavily rely on complex pixel decoders to alleviate this issue. In this paper, we challenge this trend by revealing that these decoders primarily compensate for the optimization difficulties inherent to velocity prediction ($v$-prediction). Under the clean data paradigm ($x$-prediction), they are redundant. Motivated by this insight, we advocate for simplicity over complexity and introduce PixelU, a minimalist, single-stage U-shaped Diffusion Transformer tailored for pixel space. PixelU abandons auxiliary decoders in favor of zero-cost skip connections, which provide an "information highway" that directly routes uncorrupted high-frequency spatial details from shallow to deep layers. To further enable the backbone to focus exclusively on modeling low-frequency semantics, we introduce a constant-channel spatial down-sampling mechanism as a natural low-pass filter, which compresses deep features into a compact, low-frequency semantic manifold. Extensive experiments demonstrate that this decoupling of frequencies could outperform the strong baseline (JiT-G) with only about 1/3 of its computation cost. On ImageNet 256$\times$256 and 512$\times$512, PixelU achieves FID of 1.63 and 1.92 respectively, surpassing recent pixel-space methods and establishing a simple yet powerful new paradigm for end-to-end diffusion models.

We introduce ABC, a fully open-source stack for manipulation with behavior cloning. At its core is ABC-130K: the largest open-source teleoperation dataset to date, featuring 3,500 hours of data spanning over 130K episodes across 195 diverse tasks. Furthermore, we open-source our accessible hardware setup, training infrastructure, and simulation pipeline. We also release 400 hours of sim-teleop data and provide a co-training recipe that produces correlated simulation and real-world evaluation, offering a reliable proxy for ablating model-design and training decisions before costly real-world evaluation. We explore various training recipes and compare common architectural choices for Diffusion Transformers (DiT) and Vision-Language-Action (VLA) models, grounding our findings in real-world evaluations. The resulting policies successfully execute dexterous tasks such as box folding and extracting credit cards from wallets. By providing a reproducible toolkit, we aim to place researchers on an equal footing, establishing the necessary foundation to learn the ABCs of Behavior Cloning together as a community.

Modern video diffusion transformers position their tokens through RoPE on the (u,v,t) axes -- a description of the camera's sampling grid that says nothing about the 3D structure of the scene. We observe that the geometric relation between two camera rays is captured by the Plucker reciprocal product, which is bilinear in the two rays -- the same algebraic form as the dot product in Transformer attention. Building on this analogy, we propose RayPE, a positional-encoding extension that injects per-token 6D Plucker coordinates additively into the queries and keys of self-attention, with a query/key flip arrangement under which the symmetric identity configuration coincides exactly with the reciprocal product. The injection is additive, the resulting attention score decomposes into a content term, a geometry term, and two content and geometry cross-terms -- all of which our experiments find individually necessary. To make the encoding stable across video data with heterogeneous camera-translation scales (SfM, deep SLAM, metric), we further decouple ray direction from moment magnitude, gate the encoding by a learned function of the log-magnitude, and apply RMSNorm to align it with the QKNorm-normalized content branch. The full module adds less than 0.1% parameters to a pretrained video DiT, is zero-initialized to start from the pretrained weights, and improves camera controllability, cross-frame 3D consistency, and overall video quality on a four-dataset training mixture.

Moving an object in a single image requires geometry-consistent spatial rearrangement, including handling occlusions, revealing previously unseen regions, and maintaining coherent shadows and reflections. Existing approaches are not well suited to this setting and often fail to preserve such scene-level consistency. We address this problem by introducing a geometry-aware object motion method that operates directly on the positional representations of diffusion transformers. Our key insight is that rotary positional embeddings (RoPE) define a structured spatial field that can be explicitly manipulated to induce controlled motion. We extend 2D RoPE into a depth-aware formulation that encodes 3D spatial structure, enabling consistent object displacement and scene-aware updates. Our model is trained using synthetic data combined with a small set of real images via parameter-efficient fine-tuning. Despite minimal real supervision, it preserves object identity under large spatial displacements, generates plausible content in newly revealed regions, and consistently updates scene-dependent effects such as shadows and illumination. Experimental results on standard object motion benchmarks demonstrate state-of-the-art performance across all evaluation metrics.

Diffusion Transformers (DiTs) have driven substantial progress in image and video generation but suffer from prohibitive computational costs. Feature caching accelerates inference by reusing intermediate representations. Existing methods rely on historical features for implementation simplicity, yet suffer from severe error accumulation at high acceleration ratios. To address this limitation, we investigate the nature of the requisite feature correction. We demonstrate that the optimal calibration update is characterized by a shared low-rank subspace across diverse prompts. Guided by this structural insight, we propose LearniBridge, a learnable calibration mechanism for feature caching that bridges multiple timesteps through lightweight LoRA updates. This mechanism enables effective calibration requiring only 3-5 training samples. Extensive experiments on image and video generation show that LearniBridge achieves up to $5.87\times$, $5.75\times$, and $4.10\times$ acceleration on FLUX, HunyuanVideo, and WAN2.1, respectively. On WAN2.1, it improves VBench by 1.28% over the previous SOTA at $4.10\times$ acceleration. Our code is available at https://github.com/Iiiiiiirene/LearniBridge.

The adoption of powerful diffusion models is hindered by their significant inference latency. Recent ``cache-then-forecast'' schemes alleviate this issue by accelerating DiTs using derivative-based polynomials, but they suffer from severe quality degradation at high acceleration ratios. Our analysis reveals its root cause: the discrete extrapolation performed on representations that are misaligned with the continuous diffusion trajectory and are numerically unstable. Thus, accelerated DiTs suffer from accumulated spatial errors, noisy derivative amplification, and high-order instability. We therefore reformulate accelerated inference as stable macro-trajectory extrapolation in ordinary differential equation (ODE) space. Instead of predicting intermediate features, we align forecasting with the model's Global Drift (GD), i.e., the end-to-end state evolution, thereby eliminating feature inconsistency and memory overhead. However, even this smooth macro-trajectory remains vulnerable to the derivative fallacy: its higher-order temporal derivatives are intrinsically noisy. Thus, we introduce a derivative-free barycentric Lagrange extrapolator to effectively bypass derivative instability and approximation error. We further propose a bounded Phase Mapping that regularizes the extrapolation domain, suppressing oscillatory error growth. These elements collectively constitute ResilPhase, a noise-resilient acceleration framework. Experiments on FLUX.1-dev and HunyuanVideo demonstrate state-of-the-art fidelity under aggressive acceleration ratios.

Diffusion transformer (DiT) research on image generation has converged to a single evaluation setup: class-conditional generation on ImageNet. While methods improve the FID and related metrics, it is increasingly unclear whether they reflect real progress in generative modeling. The natural alternative, i.e., text-to-image (T2I) generation, is perceived as too costly or inconvenient to train and evaluate and is often skipped. We argue that this perception no longer holds. We introduce NanoGen, a unified DiT training and evaluation framework. NanoGen matches state-of-the-art DiT baselines on ImageNet and, with 12 lines of configuration change, also trains competitive text-to-image models. It currently supports RAE, VAE, pixel-space, and MeanFlow diffusion methods under both ImageNet and T2I setups. Under NanoGen, training T2I requires comparable compute to ImageNet. After training 21 latent diffusion models with NanoGen, we observe that method ranking shows no strong correlation between ImageNet and T2I generation: Pearson correlation is between -0.377 and -0.580 across three metrics. This suggests that a method which improves class-conditional ImageNet FID may show no corresponding improvement on T2I, clearly indicating the necessity of evaluating DiTs on both tasks. To this end, we summarize ImageNet and text-to-image results, which yields DiffusionBench, a holistic benchmark for DiT research. We recommend reporting DiffusionBench in place of ImageNet alone: methods that improve DiffusionBench are more likely to reflect broader progress.

Can representations learned for image generation also support the evaluation of generated images? We study text-to-image reward prediction as a downstream task of generative representation learning. To this end, we introduce DiT-Reward, which converts a pretrained text-to-image Diffusion Transformer into a reward model by processing near-clean image latents and aggregating text-conditioned image representations across transformer layers. Under the same training data mixture as HPSv3, DiT-Reward outperforms HPSv3 on all four evaluated preference benchmarks, reaching 85.6% on HPDv2 and 77.6% on HPDv3. When the generative backbone is frozen, a lightweight learned head can still extract meaningful preference predictions from its representations. Probing across depth further reveals that downstream reward performance is strongest in the middle-to-late layers and benefits from combining representations across different stages. We also observe consistent positive scaling with generative backbone capacity. Finally, when used to optimize Stable Diffusion 3.5 Large with Flow-GRPO, DiT-Reward outperforms HPSv3 along the matched training trajectory, with particularly clear gains in realism. Direct latent scoring also achieves a 1.65x inference speedup over HPSv3 with comparable peak memory. These results show that pretrained generative DiTs provide transferable representations for reward modeling and policy optimization.

Video diffusion models have enabled remarkable progress in video generation and editing. However, content preservation remains a core challenge: existing methods regenerate every pixel and often alter elements that should remain unchanged, such as characters or background scenes. We introduce Vera, a layered diffusion framework for content-preserving video editing. Instead of regenerating the entire video, Vera generates an edit layer along with an alpha matte for compositing with the source video, separating creative editing from content preservation by design. To encourage coherent composition with the source video, we extend the text-to-video DiT into a Mixture-of-Transformers (MoT) architecture, with separate DiTs for each layer that interact through joint self-attention. To support the training of Vera, we further construct a high-quality layered dataset with accurate alpha mattes, diverse scenes and dynamics, and visual effects. Across our quantitative benchmark and human preference study, Vera outperforms leading open-source video editing models in content preservation while remaining competitive in edit quality, using 486K frames of layered training data.

Meshes are among the most common 3D scene representations, but directly generating meshes is challenging because the representation contains important symmetries, including permutation invariance of faces and vertices. MeshFlow learns to generate triangle meshes directly as triangle soups, avoiding the need to serialize meshes into long autoregressive sequences. We adopt equivariant optimal-transport flow matching models that respect the key symmetries of triangle soups: arbitrary permutations of faces and permutations of the vertices within each face. Toward this goal, we propose a simple yet effective modification to the Diffusion Transformer architecture, resulting in a scalable network capable of modeling a velocity field while maintaining the desired equivariance. We further introduce an optimal-transport-based training objective that improves convergence by eliminating supervision signals that violate these symmetries. MeshFlow achieves mesh quality comparable to state-of-the-art autoregressive mesh generators while providing about an 18$\times$ speedup during inference. Project page is at https://qiisun.github.io/MeshFlow/.

While Diffusion Transformers (DiTs) have revolutionized high-fidelity video generation, their reliance on 3D full attention creates a quadratic computational bottleneck. Existing sparse methods face a dilemma: dynamic pruning suffers from prohibitive runtime overhead and memory fragmentation, while static heuristics fail to capture fine-grained dependencies. In this work, we propose ScalingAttention, a training-free framework grounded in a key inductive bias: while individual activations are input-dependent, the high-mass attention regions for each head rapidly converge to a stable, prompt-agnostic Intrinsic Sparse Topology. This topology is weight-encoded, scale-invariant, and efficient to extract. ScalingAttention decouples topology discovery from sparsity control via: (1) WEST (Weight-Encoded Sparse Topology), which extracts a robust block-sparse prior mask offline to eliminate runtime search; (2) FAST (Fidelity-Aware Sensitivity Tuning), which adaptively tunes head-wise sparsity based on diffusion fidelity requirements. To ensure practical acceleration, we co-design a hardware-aligned bit-wise block-sparse kernel. Experiments on Wan2.1 show up to 1.90X end-to-end speedup with superior fidelity, establishing a new Pareto frontier over state-of-the-art baselines.

Realistic integration of user-specified textures into scene images is a fundamental task in computer graphics and image editing. While existing material transfer and reference-guided inpainting methods can edit surface appearances, they often fail to address the specific requirements of texture tiling. This task necessitates precisely repeating a reference pattern according to user-defined parameters such as frequency, orientation, and scale. Furthermore, current generative approaches often struggle to maintain the structural fidelity of the reference texture, limited by either destructive pixel-level resampling or the lack of fine-grained spatial information in semantic image encoders, and they frequently fail to preserve the coherent lighting and geometry of the original scene. In this paper, we propose a novel framework for controllable and high-fidelity texture tiling based on Diffusion Transformers. Our approach introduces two key technical innovations to decouple spatial manipulation from content generation. First, we propose a Coordinate-Transformed Rotary Embedding mechanism. By applying 2D affine transformations directly to the relative positional embeddings between the target latent and the image condition, we achieve precise control over tiling patterns without explicit pixel warping, thereby utilizing the full information of the reference condition without degradation. Second, a Disjoint Attention Mask is employed to shield reference features from semantic leakage. This preserves structural integrity while seamlessly blending the synthesized texture with the scene's original lighting and geometry. Extensive experiments demonstrate that our method outperforms state-of-the-art baselines in both control accuracy and texture fidelity.

Choreographic motion generation poses unique challenges for AI, demanding precise semantic control over complex, temporally structured, and expressive full-body dynamics. While existing models can synthesize motion from music, they remain largely black boxes. Conversely, attempting to condition generation on both text and music frequently leads to modality collapse, where dense acoustic rhythms overwhelm sparse semantic text prompts, destroying user controllability. To resolve this spatial-temporal conflict, we propose STREAM (Structural-Temporal Rhythmic Energy-based Attention for Motion), a modality-decoupled diffusion transformer. STREAM strictly separates conditioning pathways: global text semantics dictate the kinematic structure via Adaptive Layer Normalization (AdaLN), while a novel Bimodal Energy-Based Attention Module (BEAM) routes these features to the musical beat without overwriting the semantics. We further introduce Motorica++, a newly curated dataset enriched with domain-specific dance vocabulary and frame-level semantic annotations from existing Motorica dataset. Additionally, to rigorously quantify zero-shot editability, we propose the Exchange Evaluation Protocol and Editable Dance Score (EDS). Through extensive experiments, STREAM achieves state-of-the-art alignment between motion and music while perfectly preserving choreographic semantics, positioning AI not merely as a reactive synthesizer, but as a controllable, collaborative partner for artistic direction. The source code and datasets are available at https://github.com/SeongJong-Yoo/STREAM.

Diffusion models rely heavily on explicit timestep embeddings to modulate the denoising process across various noise scales. In this work, we challenge the necessity of these temporal signals by analyzing their impact on U-Net and Diffusion Transformer architectures. Beyond empirical evidence, we provide a theoretical framework demonstrating that, under certain conditions, the global minimizer of the diffusion training objective can be achieved without explicit timestep conditioning. Our findings reveal a surprising robustness when timestep embeddings are completely removed. Extensive ablation studies on the CelebA and CIFAR-10 datasets show that these time-agnostic models can maintain high structural fidelity and even surpass their conditioned counterparts in competitive metrics, including FID, precision, and recall. Our analysis suggests these architectures can implicitly infer noise scales from the corrupted input under specific assumptions, rendering explicit temporal conditioning redundant. This study challenges long-standing temporal conditioning paradigms and paves the way for more efficient and structurally focused generative architectures.

Generating high-resolution 3D CT volumes with fine details remains challenging due to substantial computational demands and optimization difficulties inherent to existing generative models. In this paper, we propose the Pixel-Level Residual Diffusion Transformer (PRDiT), a scalable generative framework that synthesizes high-quality 3D medical volumes directly at voxel-level. PRDiT introduces a two-stage training architecture comprising 1) a local denoiser in the form of an MLP-based blind estimator operating on overlapping 3D patches to separate low-frequency structures efficiently, and 2) a global residual diffusion transformer employing memory-efficient attention to model and refine high-frequency residuals across entire volumes. This coarse-to-fine modeling strategy simplifies optimization, enhances training stability, and effectively preserves subtle structures without the limitations of an autoencoder bottleneck. Extensive experiments conducted on the LIDC-IDRI and RAD-ChestCT datasets demonstrate that PRDiT consistently outperforms state-of-the-art models, such as HA-GAN, 3D LDM and WDM-3D, achieving significantly lower 3D FID, MMD and Wasserstein distance scores.

Audio editing aims to modify specific content in an existing audio clip according to a text instruction or description while preserving the remaining acoustic content. Despite the remarkable progress of diffusion models, existing training-based editing methods mainly rely on the local inductive biases and cross-attention interaction in convolutional U-Net backbones, which often hinder long-range semantic alignment and precise understanding and localization of instructions. In contrast, diffusion transformers provide stronger global modeling and multimodal fusion, but existing editing architectures usually adopt a simple stack of diffusion transformer blocks. Applying joint attention over concatenated audio and text tokens in all blocks results in quadratic complexity with respect to token length. To balance editing performance and efficiency, we propose a novel instruction-guided audio editing framework based on rectified flow matching (RFM), named RFM-Editing 2, built on a hybrid two-stage diffusion transformer. The proposed model performs joint attention over audio and text tokens to establish coarse semantic alignment at the low-resolution stage, then switches to alternating joint-attention and cross-attention blocks to refine editing details at the high-resolution stage. This coarse-to-fine strategy enables efficient and accurate instruction-guided audio editing. Experiments show that the proposed framework achieves notable performance gains on challenging editing tasks involving overlapping audio events and complex instructions, while substantially improving editing efficiency.

Handwritten Mathematical Expression Generation (HMEG) is challenging due to the complex two-dimensional layouts and long-range structural dependencies of mathematical expressions. Existing methods typically rely on explicit spatial supervision, such as symbol-level bounding boxes, which incurs high annotation costs and limits scalability. In this work, we propose DiffMath, a symbol- and graph-aware latent diffusion framework that leverages the hierarchical structure inherent in LaTeX as a structural prior, eliminating the need for positional supervision. First, we design a Relational Abstract Syntax Tree (RelAST), a generation-oriented representation that distills MathML trees into compact triplet sequences [S, R, D], where each token directly encodes a symbol identity, spatial relation, or nesting depth. Second, we introduce MathVAE, which learns structure-preserving latent representations through symbol-aware and relation-aware perceptual regularization, ensuring that the latent space captures both character semantics and spatial topology. Third, MathDiT performs conditional denoising in this structured latent space, further guided by a global symbol-count prior via Adaptive Layer Normalization (AdaLN) to improve structural coherence. Experiments show that DiffMath produces structurally consistent handwritten expressions, achieves superior performance over existing methods, and improves the accuracy of downstream OCR models through synthetic data augmentation.