Xin Tao

Tianxiong Zhong, Xingye Tian, Xuebo Wang et al.

Latent diffusion models leverage visual tokenizers to compress images into latent spaces for efficient generative modeling. However, better reconstruction quality of a tokenizer does not necessarily translate into better generation quality, suggesting that latent representations should be evaluated not only by fidelity but also by their diffusability. Recent studies have proposed diverse explanations for diffusion-friendly latent spaces, including semantic separability, affine equivariance, distribution uniformity, spatial structure, spectral smoothness, and manifold continuity. Yet these properties are often validated on a limited set of tokenizers, leaving it unclear which factors are most predictive of downstream generation quality and whether such conclusions hold beyond the specific settings in which they are introduced. In this work, we conduct a systematic study of latent diffusability by training a large collection of tokenizers with diverse regularization strategies, architectures, and latent configurations, and evaluating them with multiple downstream diffusion backbones. Our analysis identifies several latent properties that consistently correlate with generation quality and exhibit strong generalization across experimental settings. Beyond existing metrics, we introduce Velocity Irreducible Variance (VIV), a measure of velocity ambiguity induced by trajectory crossings. Extensive experiments show that VIV is one of the most stable predictors of generation quality.

Tao Zhang, Xingye Tian, Haoran Wei et al.

Video panoptic segmentation is a challenging task that serves as the cornerstone of numerous downstream applications, including video editing and autonomous driving. We believe that the decoupling strategy proposed by DVIS enables more effective utilization of temporal information for both "thing" and "stuff" objects. In this report, we successfully validated the effectiveness of the decoupling strategy in video panoptic segmentation. Finally, our method achieved a VPQ score of 51.4 and 53.7 in the development and test phases, respectively, and ultimately ranked 1st in the VPS track of the 2nd PVUW Challenge. The code is available at https://github.com/zhang-tao-whu/DVIS

Tao Zhang, Xingye Tian, Yikang Zhou et al.

Video instance segmentation is a challenging task that serves as the cornerstone of numerous downstream applications, including video editing and autonomous driving. In this report, we present further improvements to the SOTA VIS method, DVIS. First, we introduce a denoising training strategy for the trainable tracker, allowing it to achieve more stable and accurate object tracking in complex and long videos. Additionally, we explore the role of visual foundation models in video instance segmentation. By utilizing a frozen VIT-L model pre-trained by DINO v2, DVIS demonstrates remarkable performance improvements. With these enhancements, our method achieves 57.9 AP and 56.0 AP in the development and test phases, respectively, and ultimately ranked 1st in the VIS track of the 5th LSVOS Challenge. The code will be available at https://github.com/zhang-tao-whu/DVIS.

Takashi Isobe, Xu Jia, Xin Tao et al.

Temporal modeling is crucial for video super-resolution. Most of the video super-resolution methods adopt the optical flow or deformable convolution for explicitly motion compensation. However, such temporal modeling techniques increase the model complexity and might fail in case of occlusion or complex motion, resulting in serious distortion and artifacts. In this paper, we propose to explore the role of explicit temporal difference modeling in both LR and HR space. Instead of directly feeding consecutive frames into a VSR model, we propose to compute the temporal difference between frames and divide those pixels into two subsets according to the level of difference. They are separately processed with two branches of different receptive fields in order to better extract complementary information. To further enhance the super-resolution result, not only spatial residual features are extracted, but the difference between consecutive frames in high-frequency domain is also computed. It allows the model to exploit intermediate SR results in both future and past to refine the current SR output. The difference at different time steps could be cached such that information from further distance in time could be propagated to the current frame for refinement. Experiments on several video super-resolution benchmark datasets demonstrate the effectiveness of the proposed method and its favorable performance against state-of-the-art methods.

Qi Fan, Xin Tao, Lei Ke et al.

The Segment Anything Model (SAM) achieves remarkable promptable segmentation given high-quality prompts which, however, often require good skills to specify. To make SAM robust to casual prompts, this paper presents the first comprehensive analysis on SAM's segmentation stability across a diverse spectrum of prompt qualities, notably imprecise bounding boxes and insufficient points. Our key finding reveals that given such low-quality prompts, SAM's mask decoder tends to activate image features that are biased towards the background or confined to specific object parts. To mitigate this issue, our key idea consists of calibrating solely SAM's mask attention by adjusting the sampling locations and amplitudes of image features, while the original SAM model architecture and weights remain unchanged. Consequently, our deformable sampling plugin (DSP) enables SAM to adaptively shift attention to the prompted target regions in a data-driven manner, facilitated by our effective robust training strategy (RTS). During inference, dynamic routing plugin (DRP) is proposed that toggles SAM between the deformable and regular grid sampling modes, conditioned on the input prompt quality. Thus, our solution, termed Stable-SAM, offers several advantages: 1) improved SAM's segmentation stability across a wide range of prompt qualities, while 2) retaining SAM's powerful promptable segmentation efficiency and generality, with 3) minimal learnable parameters (0.08 M) and fast adaptation (by 1 training epoch). Extensive experiments across multiple datasets validate the effectiveness and advantages of our approach, underscoring Stable-SAM as a more robust solution for segmenting anything. Codes will be released upon acceptance. https://github.com/fanq15/Stable-SAM

Junhao Cheng, Liang Hou, Tianxiong Zhong et al.

The recent "Reasoning with Video" paradigm utilizes Video Generation Models (VGMs) to generate temporally coherent visual trajectories to complete reasoning tasks. Although state-of-the-art VGMs excel at visual quality, they often struggle to understand and follow task-specific rules, leading to logical failures across diverse reasoning scenarios. Existing efforts try to utilize Vision-Language Models (VLMs) as problem pre-solvers to produce or refine textual guidance for the VGM. However, textual descriptions fail to capture intricate spatiotemporal details, and VGMs often struggle to faithfully execute fine-grained or long-tail instructions even with a valid plan. While VLMs struggle as solvers, they possess strong perception capabilities to evaluate process-constraint satisfaction and final-goal achievement. Leveraging this strength, we introduce a paradigm shift that transitions the role of VLMs to "teachers". Specifically, a VLM teacher extracts task-specific rules to formulate differentiable rewards, guiding a VGM Reasoner via test-time online optimization of a lightweight LoRA module. This strategy enables adaptive test-time optimization and extends the reasoning capabilities beyond the VGM's intrinsic boundaries. Evaluations on symbolic (VBVR-Bench) and general-purpose (RULER-Bench) video reasoning benchmarks show that the proposed method yields a 16.7-point average performance gain, outperforming the VLM-as-Solver paradigm (+0.4 points) and Best-of-N scaling (+2.2 points) by a large margin at comparable test-time cost. These findings reveal that integrating VLMs as test-time teachers offers a promising paradigm for achieving generalizable video reasoning. Project Page: https://VLM-as-Teacher.github.io/

Jiayin Cai, Changlin Li, Xin Tao et al.

This paper proposes a novel video inpainting method. We make three main contributions: First, we extended previous Transformers with patch alignment by introducing Deformed Patch-based Homography (DePtH), which improves patch-level feature alignments without additional supervision and benefits challenging scenes with various deformation. Second, we introduce Mask Pruning-based Patch Attention (MPPA) to improve patch-wised feature matching by pruning out less essential features and using saliency map. MPPA enhances matching accuracy between warped tokens with invalid pixels. Third, we introduce a Spatial-Temporal weighting Adaptor (STA) module to obtain accurate attention to spatial-temporal tokens under the guidance of the Deformation Factor learned from DePtH, especially for videos with agile motions. Experimental results demonstrate that our method outperforms recent methods qualitatively and quantitatively and achieves a new state-of-the-art.

Songlin Tang, Wenjie Pei, Xin Tao et al.

Existing methods for interactive segmentation in radiance fields entail scene-specific optimization and thus cannot generalize across different scenes, which greatly limits their applicability. In this work we make the first attempt at Scene-Generalizable Interactive Segmentation in Radiance Fields (SGISRF) and propose a novel SGISRF method, which can perform 3D object segmentation for novel (unseen) scenes represented by radiance fields, guided by only a few interactive user clicks in a given set of multi-view 2D images. In particular, the proposed SGISRF focuses on addressing three crucial challenges with three specially designed techniques. First, we devise the Cross-Dimension Guidance Propagation to encode the scarce 2D user clicks into informative 3D guidance representations. Second, the Uncertainty-Eliminated 3D Segmentation module is designed to achieve efficient yet effective 3D segmentation. Third, Concealment-Revealed Supervised Learning scheme is proposed to reveal and correct the concealed 3D segmentation errors resulted from the supervision in 2D space with only 2D mask annotations. Extensive experiments on two real-world challenging benchmarks covering diverse scenes demonstrate 1) effectiveness and scene-generalizability of the proposed method, 2) favorable performance compared to classical method requiring scene-specific optimization.

Huimin Zeng, Weinong Wang, Xin Tao et al.

Recent interactive segmentation methods iteratively take source image, user guidance and previously predicted mask as the input without considering the invariant nature of the source image. As a result, extracting features from the source image is repeated in each interaction, resulting in substantial computational redundancy. In this work, we propose the Feature Decoupling-Recycling Network (FDRN), which decouples the modeling components based on their intrinsic discrepancies and then recycles components for each user interaction. Thus, the efficiency of the whole interactive process can be significantly improved. To be specific, we apply the Decoupling-Recycling strategy from three perspectives to address three types of discrepancies, respectively. First, our model decouples the learning of source image semantics from the encoding of user guidance to process two types of input domains separately. Second, FDRN decouples high-level and low-level features from stratified semantic representations to enhance feature learning. Third, during the encoding of user guidance, current user guidance is decoupled from historical guidance to highlight the effect of current user guidance. We conduct extensive experiments on 6 datasets from different domains and modalities, which demonstrate the following merits of our model: 1) superior efficiency than other methods, particularly advantageous in challenging scenarios requiring long-term interactions (up to 4.25x faster), while achieving favorable segmentation performance; 2) strong applicability to various methods serving as a universal enhancement technique; 3) well cross-task generalizability, e.g., to medical image segmentation, and robustness against misleading user guidance.

Changlin Li, Guangyang Wu, Yanan Sun et al.

Capitalizing on the rapid development of neural networks, recent video frame interpolation (VFI) methods have achieved notable improvements. However, they still fall short for real-world videos containing large motions. Complex deformation and/or occlusion caused by large motions make it an extremely difficult problem in video frame interpolation. In this paper, we propose a simple yet effective solution, H-VFI, to deal with large motions in video frame interpolation. H-VFI contributes a hierarchical video interpolation transformer (HVIT) to learn a deformable kernel in a coarse-to-fine strategy in multiple scales. The learnt deformable kernel is then utilized in convolving the input frames for predicting the interpolated frame. Starting from the smallest scale, H-VFI updates the deformable kernel by a residual in succession based on former predicted kernels, intermediate interpolated results and hierarchical features from transformer. Bias and masks to refine the final outputs are then predicted by a transformer block based on interpolated results. The advantage of such a progressive approximation is that the large motion frame interpolation problem can be decomposed into several relatively simpler sub-tasks, which enables a very accurate prediction in the final results. Another noteworthy contribution of our paper consists of a large-scale high-quality dataset, YouTube200K, which contains videos depicting a great variety of scenarios captured at high resolution and high frame rate. Extensive experiments on multiple frame interpolation benchmarks validate that H-VFI outperforms existing state-of-the-art methods especially for videos with large motions.

Jiehui Huang, Yuechen Zhang, Xu He et al.

Recent video generation models demonstrate impressive synthesis capabilities but remain limited by single-modality conditioning, constraining their holistic world understanding. This stems from insufficient cross-modal interaction and limited modal diversity for comprehensive world knowledge representation. To address these limitations, we introduce UnityVideo, a unified framework for world-aware video generation that jointly learns across multiple modalities (segmentation masks, human skeletons, DensePose, optical flow, and depth maps) and training paradigms. Our approach features two core components: (1) dynamic noising to unify heterogeneous training paradigms, and (2) a modality switcher with an in-context learner that enables unified processing via modular parameters and contextual learning. We contribute a large-scale unified dataset with 1.3M samples. Through joint optimization, UnityVideo accelerates convergence and significantly enhances zero-shot generalization to unseen data. We demonstrate that UnityVideo achieves superior video quality, consistency, and improved alignment with physical world constraints. Code and data can be found at: https://github.com/dvlab-research/UnityVideo

Longtao Jiang, Jiangmin Bao, Zhendong Wang et al.

Normalizing flows (NFs) provide exact likelihoods and deterministic invertible sampling, but have historically lagged behind diffusion models for large-scale image generation. We identify a key obstacle: NFs are required to learn a single invertible transport over the full ambient space, making them highly sensitive to high-dimensional representations. This leads to a semantic-capacity mismatch in modern visual representation spaces, where semantic information is compact but encoded in overcomplete features. We propose SRC-Flow, which introduces a Semantic Representation Compressor (SRC) to compact high-dimensional RAE features into a low-dimensional semantic space before flow modeling and preserve reconstruction through the frozen RAE decoder. This compact space reduces the modeling burden of NFs and enables effective likelihood-based generation in semantic representation space. We further adopt constant noise regularization tailored to the fixed unconditional bijection learned by flows. On ImageNet $256 \times 256$ and $512 \times 512$, SRC-Flow achieves state-of-the-art generation quality among normalizing flow methods, with gFID scores of 1.65 and 2.07 under classifier-free guidance, while retaining exact likelihood computation in the compact semantic representation space and deterministic invertible sampling at the flow level. Codes and models will be available at https://github.com/longtaojiang/SRC-Flow.

Tongcheng Fang, Hanling Zhang, Ruiqi Xie et al. · tsinghua

Diffusion Transformers have recently demonstrated remarkable performance in video generation. However, the long input sequences result in high computational latency due to the quadratic complexity of full attention. Various sparse attention mechanisms have been proposed. Training-free sparse attention is constrained by limited sparsity and thus offers modest acceleration, whereas training-based methods can reach much higher sparsity but demand substantial data and computation for training. In this work, we propose SALAD, introducing a lightweight linear attention branch in parallel with the sparse attention. By incorporating an input-dependent gating mechanism to finely balance the two branches, our method attains 90% sparsity and 1.72x inference speedup, while maintaining generation quality comparable to the full attention baseline. Moreover, our finetuning process is highly efficient, requiring only 2,000 video samples and 1,600 training steps with a batch size of 8.

Donglin Yang, Yongxing Zhang, Xin Yu et al.

While flow matching is elegant, its reliance on single-sample conditional velocities leads to high-variance training targets that destabilize optimization and slow convergence. By explicitly characterizing this variance, we identify 1) a high-variance regime near the prior, where optimization is challenging, and 2) a low-variance regime near the data distribution, where conditional and marginal velocities nearly coincide. Leveraging this insight, we propose Stable Velocity, a unified framework that improves both training and sampling. For training, we introduce Stable Velocity Matching (StableVM), an unbiased variance-reduction objective, along with Variance-Aware Representation Alignment (VA-REPA), which adaptively strengthen auxiliary supervision in the low-variance regime. For inference, we show that dynamics in the low-variance regime admit closed-form simplifications, enabling Stable Velocity Sampling (StableVS), a finetuning-free acceleration. Extensive experiments on ImageNet $256\times256$ and large pretrained text-to-image and text-to-video models, including SD3.5, Flux, Qwen-Image, and Wan2.2, demonstrate consistent improvements in training efficiency and more than $2\times$ faster sampling within the low-variance regime without degrading sample quality. Our code is available at https://github.com/linYDTHU/StableVelocity.

Yuanyang Yin, Yaqi Zhao, Yajie Zhang et al.

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities by integrating visual and textual inputs, yet modality alignment remains one of the most challenging aspects. Current MLLMs typically rely on simple adapter architectures and pretraining approaches to bridge vision encoders with large language models (LLM), guided by image-level supervision. We identify this paradigm often leads to suboptimal alignment between modalities, significantly constraining the LLM's ability to properly interpret and reason with visual features particularly for smaller language models. This limitation degrades overall performance-particularly for smaller language models where capacity constraints are more pronounced and adaptation capabilities are limited. To address this fundamental limitation, we propose Supervised Embedding Alignment (SEA), a token-level supervision alignment method that enables more precise visual-text alignment during pretraining. SEA introduces minimal computational overhead while preserving language capabilities and substantially improving cross-modal understanding. Our comprehensive analyses reveal critical insights into the adapter's role in multimodal integration, and extensive experiments demonstrate that SEA consistently improves performance across various model sizes, with smaller models benefiting the most (average performance gain of 7.61% for Gemma-2B). This work establishes a foundation for developing more effective alignment strategies for future multimodal systems.

Xin Tao, Mladen Čičić, Jonas Mårtensson

With the development of vehicular technologies on automation, electrification, and digitalization, vehicles are becoming more intelligent while being exposed to more complex, uncertain, and frequently occurring faults. In this paper, we look into the maintenance planning of an operating vehicle under fault condition and formulate it as a multi-criteria decision-making problem. The maintenance decisions are generated by route searching in road networks and evaluated based on risk assessment considering the uncertainty of vehicle breakdowns. Particularly, we consider two criteria, namely the risk of public time loss and the risk of mission delay, representing the concerns of the public sector and the private sector, respectively. A public time loss model is developed to evaluate the traffic congestion caused by a vehicle breakdown and the corresponding towing process. The Pareto optimal set of non-dominated decisions is derived by evaluating the risk of the decisions. We demonstrate the relevance of the problem and the effectiveness of the proposed method by numerical experiments derived from real-world scenarios. The experiments show that neglecting the risk of vehicle breakdown on public roads can cause a high risk of public time loss in dense traffic flow. With the proposed method, alternate decisions can be derived to reduce the risks of public time loss significantly with a low increase in the risk of mission delay. This study aims at catalyzing public-private partnership through collaborative decision-making between the private sector and the public sector, thus archiving a more sustainable transportation system in the future.

Tao Zhang, Xingye Tian, Yikang Zhou et al.

We present the \textbf{D}ecoupled \textbf{VI}deo \textbf{S}egmentation (DVIS) framework, a novel approach for the challenging task of universal video segmentation, including video instance segmentation (VIS), video semantic segmentation (VSS), and video panoptic segmentation (VPS). Unlike previous methods that model video segmentation in an end-to-end manner, our approach decouples video segmentation into three cascaded sub-tasks: segmentation, tracking, and refinement. This decoupling design allows for simpler and more effective modeling of the spatio-temporal representations of objects, especially in complex scenes and long videos. Accordingly, we introduce two novel components: the referring tracker and the temporal refiner. These components track objects frame by frame and model spatio-temporal representations based on pre-aligned features. To improve the tracking capability of DVIS, we propose a denoising training strategy and introduce contrastive learning, resulting in a more robust framework named DVIS++. Furthermore, we evaluate DVIS++ in various settings, including open vocabulary and using a frozen pre-trained backbone. By integrating CLIP with DVIS++, we present OV-DVIS++, the first open-vocabulary universal video segmentation framework. We conduct extensive experiments on six mainstream benchmarks, including the VIS, VSS, and VPS datasets. Using a unified architecture, DVIS++ significantly outperforms state-of-the-art specialized methods on these benchmarks in both close- and open-vocabulary settings. Code:~\url{https://github.com/zhang-tao-whu/DVIS_Plus}.

Yixuan Zhu, Jiaqi Feng, Wenzhao Zheng et al.

Recent advances in diffusion transformers have empowered video generation models to generate high-quality video clips from texts or images. However, world models with the ability to predict long-horizon futures from past observations and actions remain underexplored, especially for general-purpose scenarios and various forms of actions. To bridge this gap, we introduce Astra, an interactive general world model that generates real-world futures for diverse scenarios (e.g., autonomous driving, robot grasping) with precise action interactions (e.g., camera motion, robot action). We propose an autoregressive denoising architecture and use temporal causal attention to aggregate past observations and support streaming outputs. We use a noise-augmented history memory to avoid over-reliance on past frames to balance responsiveness with temporal coherence. For precise action control, we introduce an action-aware adapter that directly injects action signals into the denoising process. We further develop a mixture of action experts that dynamically route heterogeneous action modalities, enhancing versatility across diverse real-world tasks such as exploration, manipulation, and camera control. Astra achieves interactive, consistent, and general long-term video prediction and supports various forms of interactions. Experiments across multiple datasets demonstrate the improvements of Astra in fidelity, long-range prediction, and action alignment over existing state-of-the-art world models.

Guangyang Wu, Xin Tao, Changlin Li et al.

Previous methods for Video Frame Interpolation (VFI) have encountered challenges, notably the manifestation of blur and ghosting effects. These issues can be traced back to two pivotal factors: unavoidable motion errors and misalignment in supervision. In practice, motion estimates often prove to be error-prone, resulting in misaligned features. Furthermore, the reconstruction loss tends to bring blurry results, particularly in misaligned regions. To mitigate these challenges, we propose a new paradigm called PerVFI (Perception-oriented Video Frame Interpolation). Our approach incorporates an Asymmetric Synergistic Blending module (ASB) that utilizes features from both sides to synergistically blend intermediate features. One reference frame emphasizes primary content, while the other contributes complementary information. To impose a stringent constraint on the blending process, we introduce a self-learned sparse quasi-binary mask which effectively mitigates ghosting and blur artifacts in the output. Additionally, we employ a normalizing flow-based generator and utilize the negative log-likelihood loss to learn the conditional distribution of the output, which further facilitates the generation of clear and fine details. Experimental results validate the superiority of PerVFI, demonstrating significant improvements in perceptual quality compared to existing methods. Codes are available at \url{https://github.com/mulns/PerVFI}

Yuanhui Huang, Wenzhao Zheng, Yuan Gao et al.

Video generation models (VGMs) have received extensive attention recently and serve as promising candidates for general-purpose large vision models. While they can only generate short videos each time, existing methods achieve long video generation by iteratively calling the VGMs, using the last-frame output as the condition for the next-round generation. However, the last frame only contains short-term fine-grained information about the scene, resulting in inconsistency in the long horizon. To address this, we propose an Omni World modeL (Owl-1) to produce long-term coherent and comprehensive conditions for consistent long video generation. As videos are observations of the underlying evolving world, we propose to model the long-term developments in a latent space and use VGMs to film them into videos. Specifically, we represent the world with a latent state variable which can be decoded into explicit video observations. These observations serve as a basis for anticipating temporal dynamics which in turn update the state variable. The interaction between evolving dynamics and persistent state enhances the diversity and consistency of the long videos. Extensive experiments show that Owl-1 achieves comparable performance with SOTA methods on VBench-I2V and VBench-Long, validating its ability to generate high-quality video observations. Code: https://github.com/huang-yh/Owl.

Yuechen Zhang, Jinbo Xing, Bin Xia et al.

Despite the remarkable generation quality of video Diffusion Transformer (DiT) models, their practical deployment is severely hindered by extensive computational requirements. This inefficiency stems from two key challenges: the quadratic complexity of self-attention with respect to token length and the multi-step nature of diffusion models. To address these limitations, we present Jenga, a novel inference pipeline that combines dynamic attention carving with progressive resolution generation. Our approach leverages two key insights: (1) early denoising steps do not require high-resolution latents, and (2) later steps do not require dense attention. Jenga introduces a block-wise attention mechanism that dynamically selects relevant token interactions using 3D space-filling curves, alongside a progressive resolution strategy that gradually increases latent resolution during generation. Experimental results demonstrate that Jenga achieves substantial speedups across multiple state-of-the-art video diffusion models while maintaining comparable generation quality (8.83$\times$ speedup with 0.01\% performance drop on VBench). As a plug-and-play solution, Jenga enables practical, high-quality video generation on modern hardware by reducing inference time from minutes to seconds -- without requiring model retraining. Code: https://github.com/dvlab-research/Jenga

Luozhou Wang, Zhifei Chen, Yihua Du et al.

Large-scale video generation models have demonstrated emergent physical coherence, positioning them as potential world models. However, a gap remains between contemporary "stateless" video architectures and classic state-centric world model theories. This work bridges this gap by proposing a novel taxonomy centered on two pillars: State Construction and Dynamics Modeling. We categorize state construction into implicit paradigms (context management) and explicit paradigms (latent compression), while dynamics modeling is analyzed through knowledge integration and architectural reformulation. Furthermore, we advocate for a transition in evaluation from visual fidelity to functional benchmarks, testing physical persistence and causal reasoning. We conclude by identifying two critical frontiers: enhancing persistence via data-driven memory and compressed fidelity, and advancing causality through latent factor decoupling and reasoning-prior integration. By addressing these challenges, the field can evolve from generating visually plausible videos to building robust, general-purpose world simulators.

Cheng Liang, Haoxian Chen, Liang Hou et al.

The quadratic complexity of the attention mechanism severely limits the context scalability of Video Diffusion Transformers (DiTs). We find that the highly sparse spatio-temporal attention patterns exhibited in Video DiTs can be naturally represented by the Monarch matrix. It is a class of structured matrices with flexible sparsity, enabling sub-quadratic attention via an alternating minimization algorithm. Accordingly, we propose VMonarch, a novel attention mechanism for Video DiTs that enables efficient computation over the dynamic sparse patterns with structured Monarch matrices. First, we adapt spatio-temporal Monarch factorization to explicitly capture the intra-frame and inter-frame correlations of the video data. Second, we introduce a recomputation strategy to mitigate artifacts arising from instabilities during alternating minimization of Monarch matrices. Third, we propose a novel online entropy algorithm fused into FlashAttention, enabling fast Monarch matrix updates for long sequences. Extensive experiments demonstrate that VMonarch achieves comparable or superior generation quality to full attention on VBench after minimal tuning. It overcomes the attention bottleneck in Video DiTs, reduces attention FLOPs by a factor of 17.5, and achieves a speedup of over 5x in attention computation for long videos, surpassing state-of-the-art sparse attention methods at 90% sparsity.

Sihui Ji, Xi Chen, Shuai Yang et al.

The core challenge for streaming video generation is maintaining the content consistency in long context, which poses high requirement for the memory design. Most existing solutions maintain the memory by compressing historical frames with predefined strategies. However, different to-generate video chunks should refer to different historical cues, which is hard to satisfy with fixed strategies. In this work, we propose MemFlow to address this problem. Specifically, before generating the coming chunk, we dynamically update the memory bank by retrieving the most relevant historical frames with the text prompt of this chunk. This design enables narrative coherence even if new event happens or scenario switches in future frames. In addition, during generation, we only activate the most relevant tokens in the memory bank for each query in the attention layers, which effectively guarantees the generation efficiency. In this way, MemFlow achieves outstanding long-context consistency with negligible computation burden (7.9% speed reduction compared with the memory-free baseline) and keeps the compatibility with any streaming video generation model with KV cache.

Tianxiong Zhong, Xingye Tian, Boyuan Jiang et al.

Modern video generation frameworks based on Latent Diffusion Models suffer from inefficiencies in tokenization due to the Frame-Proportional Information Assumption. Existing tokenizers provide fixed temporal compression rates, causing the computational cost of the diffusion model to scale linearly with the frame rate. The paper proposes the Duration-Proportional Information Assumption: the upper bound on the information capacity of a video is proportional to the duration rather than the number of frames. Based on this insight, the paper introduces VFRTok, a Transformer-based video tokenizer, that enables variable frame rate encoding and decoding through asymmetric frame rate training between the encoder and decoder. Furthermore, the paper proposes Partial Rotary Position Embeddings (RoPE) to decouple position and content modeling, which groups correlated patches into unified tokens. The Partial RoPE effectively improves content-awareness, enhancing the video generation capability. Benefiting from the compact and continuous spatio-temporal representation, VFRTok achieves competitive reconstruction quality and state-of-the-art generation fidelity while using only 1/8 tokens compared to existing tokenizers. The code and weights are released at: https://github.com/KwaiVGI/VFRTok.

Guibao Shen, Yihua Du, Wenhang Ge et al.

The rapid growth of stereoscopic displays, including VR headsets and 3D cinemas, has led to increasing demand for high-quality stereo video content. However, producing 3D videos remains costly and complex, while automatic Monocular-to-Stereo conversion is hindered by the limitations of the multi-stage ``Depth-Warp-Inpaint'' (DWI) pipeline. This paradigm suffers from error propagation, depth ambiguity, and format inconsistency between parallel and converged stereo configurations. To address these challenges, we introduce UniStereo, the first large-scale unified dataset for stereo video conversion, covering both stereo formats to enable fair benchmarking and robust model training. Building upon this dataset, we propose StereoPilot, an efficient feed-forward model that directly synthesizes the target view without relying on explicit depth maps or iterative diffusion sampling. Equipped with a learnable domain switcher and a cycle consistency loss, StereoPilot adapts seamlessly to different stereo formats and achieves improved consistency. Extensive experiments demonstrate that StereoPilot significantly outperforms state-of-the-art methods in both visual fidelity and computational efficiency. Project page: https://hit-perfect.github.io/StereoPilot/.

Wenhang Ge, Guibao Shen, Jiawei Feng et al.

Recent advances in camera-controlled video diffusion models have significantly improved video-camera alignment. However, the camera controllability still remains limited. In this work, we build upon Reward Feedback Learning and aim to further improve camera controllability. However, directly borrowing existing ReFL approaches faces several challenges. First, current reward models lack the capacity to assess video-camera alignment. Second, decoding latent into RGB videos for reward computation introduces substantial computational overhead. Third, 3D geometric information is typically neglected during video decoding. To address these limitations, we introduce an efficient camera-aware 3D decoder that decodes video latent into 3D representations for reward quantization. Specifically, video latent along with the camera pose are decoded into 3D Gaussians. In this process, the camera pose not only acts as input, but also serves as a projection parameter. Misalignment between the video latent and camera pose will cause geometric distortions in the 3D structure, resulting in blurry renderings. Based on this property, we explicitly optimize pixel-level consistency between the rendered novel views and ground-truth ones as reward. To accommodate the stochastic nature, we further introduce a visibility term that selectively supervises only deterministic regions derived via geometric warping. Extensive experiments conducted on RealEstate10K and WorldScore benchmarks demonstrate the effectiveness of our proposed method. Project page: \href{https://a-bigbao.github.io/CamPilot/}{CamPilot Page}.

Kaixin Ding, Yang Zhou, Xi Chen et al.

Recent advances in Text-to-Image (T2I) generative models, such as Imagen, Stable Diffusion, and FLUX, have led to remarkable improvements in visual quality. However, their performance is fundamentally limited by the quality of training data. Web-crawled and synthetic image datasets often contain low-quality or redundant samples, which lead to degraded visual fidelity, unstable training, and inefficient computation. Hence, effective data selection is crucial for improving data efficiency. Existing approaches rely on costly manual curation or heuristic scoring based on single-dimensional features in Text-to-Image data filtering. Although meta-learning based method has been explored in LLM, there is no adaptation for image modalities. To this end, we propose **Alchemist**, a meta-gradient-based framework to select a suitable subset from large-scale text-image data pairs. Our approach automatically learns to assess the influence of each sample by iteratively optimizing the model from a data-centric perspective. Alchemist consists of two key stages: data rating and data pruning. We train a lightweight rater to estimate each sample's influence based on gradient information, enhanced with multi-granularity perception. We then use the Shift-Gsampling strategy to select informative subsets for efficient model training. Alchemist is the first automatic, scalable, meta-gradient-based data selection framework for Text-to-Image model training. Experiments on both synthetic and web-crawled datasets demonstrate that Alchemist consistently improves visual quality and downstream performance. Training on an Alchemist-selected 50% of the data can outperform training on the full dataset.

Kaixin Ding, Xi Chen, Sihui Ji et al.

The demand for high-resolution video generation is growing rapidly. However, the generation resolution is severely constrained by slow inference speeds. For instance, Wan2.1 requires over 50 minutes to generate a single 720p video. While previous works explore accelerating video generation from various aspects, most of them compromise the distinctive signatures (e.g., layout, semantic, motion) of the original model. In this work, we propose SURF, an efficient framework for generating high-resolution videos, while maximally keeping the signatures. Specifically, SURF divides video generation into two stages: First, we leverage the pretrained model to infer at optimal resolution and downsample latent to generate low-resolution previews in fast speed; then we design a Refiner to upscale the preview. In the preview stage, we identify that directly inferring a model (trained with higher resolution) on lower resolution causes severe losses in signatures. So we introduce noise reshifting, a training-free technique that mitigates this issue by conducting initial denoising steps on the original resolution and switching to low resolution in later steps. In the refine stage, we establish a mapping relationship between the preview and the high-resolution target, which significantly reduces the denoising steps. We further integrate shifting windows and carefully design the training paradigm to get a powerful and efficient Refiner. In this way, SURF enables generating high-resolution videos efficiently while maximally closer to the signatures of the given pretrained model. SURF is conceptually simple and could serve as a plug-in that is compatible with various base model and acceleration methods. For example, it achieves 12.5x speedup for generating 5-second, 16fps, 720p Wan 2.1 videos and 8.7x speedup for generating 5-second, 24fps, 720p HunyuanVideo.

Junhao Cheng, Liang Hou, Xin Tao et al.

While language models have become impactful in many real-world applications, video generation remains largely confined to entertainment. Motivated by video's inherent capacity to demonstrate physical-world information that is difficult to convey through language alone (e.g., imagine teaching someone to tie a tie using only text), we identify an underutilized opportunity to extend video as a new answer modality for Next-Event Prediction (NEP), formalized as Video-Next-Event Prediction (VNEP). While the established NEP task takes a video with a procedural or predictive question as input to predict the next event in text, VNEP requires dynamic video responses. This shift from telling to showing unlocks more intuitive and customized answers for procedural learning and creative exploration. However, this task remains challenging for existing models, as it demands an understanding of multimodal input, instruction-conditioned reasoning, and the generation of video with visual and semantic consistency. To address this, we introduce VANS, a model that leverages reinforcement learning to align a Vision-Language Model (VLM) with a Video Diffusion Model (VDM) for VNEP. The core of VANS is our proposed Joint-GRPO that orchestrates the VLM and VDM to function as a unit. Driven by a shared reward on their respective output, it optimizes the VLM to produce captions that are both accurate and friendly to visualize, while guiding the VDM to generate videos that are faithful to these captions and the input visual context. To enable this learning, we craft VANS-Data-100K, a dedicated dataset for the VNEP task. Experiments on procedural and predictive benchmarks demonstrate that VANS achieves state-of-the-art performance in both video event prediction and visualization. Codes are released in https://github.com/KlingTeam/VANS.

Jia Jun Cheng Xian, Muchen Li, Haotian Yang et al.

Recent advances in diffusion-based text-to-image (T2I) models have led to remarkable success in generating high-quality images from textual prompts. However, ensuring accurate alignment between the text and the generated image remains a significant challenge for state-of-the-art diffusion models. To address this, existing studies employ reinforcement learning with human feedback (RLHF) to align T2I outputs with human preferences. These methods, however, either rely directly on paired image preference data or require a learned reward function, both of which depend heavily on costly, high-quality human annotations and thus face scalability limitations. In this work, we introduce Text Preference Optimization (TPO), a framework that enables "free-lunch" alignment of T2I models, achieving alignment without the need for paired image preference data. TPO works by training the model to prefer matched prompts over mismatched prompts, which are constructed by perturbing original captions using a large language model. Our framework is general and compatible with existing preference-based algorithms. We extend both DPO and KTO to our setting, resulting in TDPO and TKTO. Quantitative and qualitative evaluations across multiple benchmarks show that our methods consistently outperform their original counterparts, delivering better human preference scores and improved text-to-image alignment. Our Open-source code is available at https://github.com/DSL-Lab/T2I-Free-Lunch-Alignment.

Yukai Shi, Jiarong Ou, Rui Chen et al.

In visual generation tasks, the responses and combinations of complex concepts often lack stability and are error-prone, which remains an under-explored area. In this paper, we attempt to explore the causal factors for poor concept responses through elaborately designed experiments. We also design a concept-wise equalization loss function (IMBA loss) to address this issue. Our proposed method is online, eliminating the need for offline dataset processing, and requires minimal code changes. In our newly proposed complex concept benchmark Inert-CompBench and two other public test sets, our method significantly enhances the concept response capability of baseline models and yields highly competitive results with only a few codes released at https://github.com/KwaiVGI/IMBA-Loss.

Ye Tian, Ling Yang, Haotian Yang et al.

Diffusion models have demonstrated great success in text-to-video (T2V) generation. However, existing methods may face challenges when handling complex (long) video generation scenarios that involve multiple objects or dynamic changes in object numbers. To address these limitations, we propose VideoTetris, a novel framework that enables compositional T2V generation. Specifically, we propose spatio-temporal compositional diffusion to precisely follow complex textual semantics by manipulating and composing the attention maps of denoising networks spatially and temporally. Moreover, we propose an enhanced video data preprocessing to enhance the training data regarding motion dynamics and prompt understanding, equipped with a new reference frame attention mechanism to improve the consistency of auto-regressive video generation. Extensive experiments demonstrate that our VideoTetris achieves impressive qualitative and quantitative results in compositional T2V generation. Code is available at: https://github.com/YangLing0818/VideoTetris

Xiaohong Liu, Xiongkuo Min, Guangtao Zhai et al.

This paper reports on the NTIRE 2024 Quality Assessment of AI-Generated Content Challenge, which will be held in conjunction with the New Trends in Image Restoration and Enhancement Workshop (NTIRE) at CVPR 2024. This challenge is to address a major challenge in the field of image and video processing, namely, Image Quality Assessment (IQA) and Video Quality Assessment (VQA) for AI-Generated Content (AIGC). The challenge is divided into the image track and the video track. The image track uses the AIGIQA-20K, which contains 20,000 AI-Generated Images (AIGIs) generated by 15 popular generative models. The image track has a total of 318 registered participants. A total of 1,646 submissions are received in the development phase, and 221 submissions are received in the test phase. Finally, 16 participating teams submitted their models and fact sheets. The video track uses the T2VQA-DB, which contains 10,000 AI-Generated Videos (AIGVs) generated by 9 popular Text-to-Video (T2V) models. A total of 196 participants have registered in the video track. A total of 991 submissions are received in the development phase, and 185 submissions are received in the test phase. Finally, 12 participating teams submitted their models and fact sheets. Some methods have achieved better results than baseline methods, and the winning methods in both tracks have demonstrated superior prediction performance on AIGC.

Zhaokun Zhou, Qiulin Wang, Bin Lin et al.

As an alternative to expensive expert evaluation, Image Aesthetic Assessment (IAA) stands out as a crucial task in computer vision. However, traditional IAA methods are typically constrained to a single data source or task, restricting the universality and broader application. In this work, to better align with human aesthetics, we propose a Unified Multi-modal Image Aesthetic Assessment (UNIAA) framework, including a Multi-modal Large Language Model (MLLM) named UNIAA-LLaVA and a comprehensive benchmark named UNIAA-Bench. We choose MLLMs with both visual perception and language ability for IAA and establish a low-cost paradigm for transforming the existing datasets into unified and high-quality visual instruction tuning data, from which the UNIAA-LLaVA is trained. To further evaluate the IAA capability of MLLMs, we construct the UNIAA-Bench, which consists of three aesthetic levels: Perception, Description, and Assessment. Extensive experiments validate the effectiveness and rationality of UNIAA. UNIAA-LLaVA achieves competitive performance on all levels of UNIAA-Bench, compared with existing MLLMs. Specifically, our model performs better than GPT-4V in aesthetic perception and even approaches the junior-level human. We find MLLMs have great potential in IAA, yet there remains plenty of room for further improvement. The UNIAA-LLaVA and UNIAA-Bench will be released.

Luozhou Wang, Ziyang Mai, Guibao Shen et al.

In this work, we present a novel approach for motion customization in video generation, addressing the widespread gap in the exploration of motion representation within video generative models. Recognizing the unique challenges posed by the spatiotemporal nature of video, our method introduces Motion Embeddings, a set of explicit, temporally coherent embeddings derived from a given video. These embeddings are designed to integrate seamlessly with the temporal transformer modules of video diffusion models, modulating self-attention computations across frames without compromising spatial integrity. Our approach provides a compact and efficient solution to motion representation, utilizing two types of embeddings: a Motion Query-Key Embedding to modulate the temporal attention map and a Motion Value Embedding to modulate the attention values. Additionally, we introduce an inference strategy that excludes spatial dimensions from the Motion Query-Key Embedding and applies a differential operation to the Motion Value Embedding, both designed to debias appearance and ensure the embeddings focus solely on motion. Our contributions include the introduction of a tailored motion embedding for customization tasks and a demonstration of the practical advantages and effectiveness of our method through extensive experiments.

Guibao Shen, Luozhou Wang, Jiantao Lin et al.

Recent advancements in text-to-image generation have been propelled by the development of diffusion models and multi-modality learning. However, since text is typically represented sequentially in these models, it often falls short in providing accurate contextualization and structural control. So the generated images do not consistently align with human expectations, especially in complex scenarios involving multiple objects and relationships. In this paper, we introduce the Scene Graph Adapter(SG-Adapter), leveraging the structured representation of scene graphs to rectify inaccuracies in the original text embeddings. The SG-Adapter's explicit and non-fully connected graph representation greatly improves the fully connected, transformer-based text representations. This enhancement is particularly notable in maintaining precise correspondence in scenarios involving multiple relationships. To address the challenges posed by low-quality annotated datasets like Visual Genome, we have manually curated a highly clean, multi-relational scene graph-image paired dataset MultiRels. Furthermore, we design three metrics derived from GPT-4V to effectively and thoroughly measure the correspondence between images and scene graphs. Both qualitative and quantitative results validate the efficacy of our approach in controlling the correspondence in multiple relationships.

Yuanyang Yin, Yaqi Zhao, Mingwu Zheng et al.

Achieving optimal performance of video diffusion transformers within given data and compute budget is crucial due to their high training costs. This necessitates precisely determining the optimal model size and training hyperparameters before large-scale training. While scaling laws are employed in language models to predict performance, their existence and accurate derivation in visual generation models remain underexplored. In this paper, we systematically analyze scaling laws for video diffusion transformers and confirm their presence. Moreover, we discover that, unlike language models, video diffusion models are more sensitive to learning rate and batch size, two hyperparameters often not precisely modeled. To address this, we propose a new scaling law that predicts optimal hyperparameters for any model size and compute budget. Under these optimal settings, we achieve comparable performance and reduce inference costs by 40.1% compared to conventional scaling methods, within a compute budget of 1e10 TFlops. Furthermore, we establish a more generalized and precise relationship among validation loss, any model size, and compute budget. This enables performance prediction for non-optimal model sizes, which may also be appealed under practical inference cost constraints, achieving a better trade-off.

Minglei Shi, Ziyang Yuan, Haotian Yang et al.

Diffusion models have demonstrated remarkable success in various image generation tasks, but their performance is often limited by the uniform processing of inputs across varying conditions and noise levels. To address this limitation, we propose a novel approach that leverages the inherent heterogeneity of the diffusion process. Our method, DiffMoE, introduces a batch-level global token pool that enables experts to access global token distributions during training, promoting specialized expert behavior. To unleash the full potential of the diffusion process, DiffMoE incorporates a capacity predictor that dynamically allocates computational resources based on noise levels and sample complexity. Through comprehensive evaluation, DiffMoE achieves state-of-the-art performance among diffusion models on ImageNet benchmark, substantially outperforming both dense architectures with 3x activated parameters and existing MoE approaches while maintaining 1x activated parameters. The effectiveness of our approach extends beyond class-conditional generation to more challenging tasks such as text-to-image generation, demonstrating its broad applicability across different diffusion model applications. Project Page: https://shiml20.github.io/DiffMoE/

Jianzong Wu, Liang Hou, Haotian Yang et al.

The quadratic complexity of full attention mechanisms poses a significant bottleneck for Video Diffusion Models (VDMs) aiming to generate long-duration, high-resolution videos. While various sparse attention methods have been proposed, many are designed as training-free inference accelerators or do not optimally capture the unique spatio-temporal characteristics inherent in video data when trained natively. This paper introduces Video Mixture of Block Attention (VMoBA), a novel sparse attention mechanism specifically adapted for VDMs. Motivated by an in-depth analysis of attention patterns within pre-trained video transformers, which revealed strong spatio-temporal locality, varying query importance, and head-specific concentration levels, VMoBA enhances the original MoBA framework with three key modifications: (1) a layer-wise recurrent block partition scheme (1D-2D-3D) to dynamically adapt to diverse spatio-temporal attention patterns and improve efficiency; (2) global block selection to prioritize the most salient query-key block interactions across an entire attention head; and (3) threshold-based block selection to dynamically determine the number of attended blocks based on their cumulative similarity. Extensive experiments demonstrate that VMoBA significantly accelerates the training of VDMs on longer sequences, achieving 2.92x FLOPs and 1.48x latency speedup, while attaining comparable or even superior generation quality to full attention. Furthermore, VMoBA exhibits competitive performance in training-free inference, offering 2.40x FLOPs and 1.35x latency speedup for high-res video generation.

Bo Zhang, Zhuotao Tian, Xin Tao et al.

Amodal segmentation is a challenging task that aims to predict the complete geometric shape of objects, including their occluded regions. Although existing methods primarily focus on amodal segmentation within the training domain, these approaches often lack the generalization capacity to extend effectively to novel object categories and unseen contexts. This paper introduces Amodal SAM, a unified framework that leverages SAM (Segment Anything Model) for both amodal image and amodal video segmentation. Amodal SAM preserves the powerful generalization ability of SAM while extending its inherent capabilities to the amodal segmentation task. The improvements lie in three aspects: (1) a lightweight Spatial Completion Adapter that enables occluded region reconstruction, (2) a Target-Aware Occlusion Synthesis (TAOS) pipeline that addresses the scarcity of amodal annotations by generating diverse synthetic training data, and (3) novel learning objectives that enforce regional consistency and topological regularization. Extensive experiments demonstrate that Amodal SAM achieves state-of-the-art performance on standard benchmarks, while simultaneously exhibiting robust generalization to novel scenarios. We anticipate that this research will advance the field toward practical amodal segmentation systems capable of operating effectively in unconstrained real-world environments.

Shiyuan Yang, Zheng Gu, Liang Hou et al.

Video inpainting involves modifying local regions within a video, ensuring spatial and temporal consistency. Most existing methods focus primarily on scene completion (i.e., filling missing regions) and lack the capability to insert new objects into a scene in a controllable manner. Fortunately, recent advancements in text-to-video (T2V) diffusion models pave the way for text-guided video inpainting. However, directly adapting T2V models for inpainting remains limited in unifying completion and insertion tasks, lacks input controllability, and struggles with long videos, thereby restricting their applicability and flexibility. To address these challenges, we propose MTV-Inpaint, a unified multi-task video inpainting framework capable of handling both traditional scene completion and novel object insertion tasks. To unify these distinct tasks, we design a dual-branch spatial attention mechanism in the T2V diffusion U-Net, enabling seamless integration of scene completion and object insertion within a single framework. In addition to textual guidance, MTV-Inpaint supports multimodal control by integrating various image inpainting models through our proposed image-to-video (I2V) inpainting mode. Additionally, we propose a two-stage pipeline that combines keyframe inpainting with in-between frame propagation, enabling MTV-Inpaint to effectively handle long videos with hundreds of frames. Extensive experiments demonstrate that MTV-Inpaint achieves state-of-the-art performance in both scene completion and object insertion tasks. Furthermore, it demonstrates versatility in derived applications such as multi-modal inpainting, object editing, removal, image object brush, and the ability to handle long videos. Project page: https://mtv-inpaint.github.io/.

Ouxiang Li, Yuan Wang, Xinting Hu et al.

Text-to-image (T2I) generation aims to synthesize images from textual prompts, which jointly specify what must be shown and imply what can be inferred, which thus correspond to two core capabilities: composition and reasoning. Despite recent advances of T2I models in both composition and reasoning, existing benchmarks remain limited in evaluation. They not only fail to provide comprehensive coverage across and within both capabilities, but also largely restrict evaluation to low scene density and simple one-to-one reasoning. To address these limitations, we propose T2I-CoReBench, a comprehensive and complex benchmark that evaluates both composition and reasoning capabilities of T2I models. To ensure comprehensiveness, we structure composition around scene graph elements (instance, attribute, and relation) and reasoning around the philosophical framework of inference (deductive, inductive, and abductive), formulating a 12-dimensional evaluation taxonomy. To increase complexity, driven by the inherent real-world complexities, we curate each prompt with higher compositional density for composition and greater reasoning intensity for reasoning. To facilitate fine-grained and reliable evaluation, we also pair each evaluation prompt with a checklist that specifies individual yes/no questions to assess each intended element independently. In statistics, our benchmark comprises 1,080 challenging prompts and around 13,500 checklist questions. Experiments across 28 current T2I models reveal that their composition capability still remains limited in high compositional scenarios, while the reasoning capability lags even further behind as a critical bottleneck, with all models struggling to infer implicit elements from prompts.

Ruotong Wang, Mingli Zhu, Jiarong Ou et al.

Text-to-video (T2V) generative models have rapidly advanced and found widespread applications across fields like entertainment, education, and marketing. However, the adversarial vulnerabilities of these models remain rarely explored. We observe that in T2V generation tasks, the generated videos often contain substantial redundant information not explicitly specified in the text prompts, such as environmental elements, secondary objects, and additional details, providing opportunities for malicious attackers to embed hidden harmful content. Exploiting this inherent redundancy, we introduce BadVideo, the first backdoor attack framework tailored for T2V generation. Our attack focuses on designing target adversarial outputs through two key strategies: (1) Spatio-Temporal Composition, which combines different spatiotemporal features to encode malicious information; (2) Dynamic Element Transformation, which introduces transformations in redundant elements over time to convey malicious information. Based on these strategies, the attacker's malicious target seamlessly integrates with the user's textual instructions, providing high stealthiness. Moreover, by exploiting the temporal dimension of videos, our attack successfully evades traditional content moderation systems that primarily analyze spatial information within individual frames. Extensive experiments demonstrate that BadVideo achieves high attack success rates while preserving original semantics and maintaining excellent performance on clean inputs. Overall, our work reveals the adversarial vulnerability of T2V models, calling attention to potential risks and misuse. Our project page is at https://wrt2000.github.io/BadVideo2025/.

Zhen Yang, Guibao Shen, Minyang Li et al.

Diffusion models have achieved remarkable progress across various visual generation tasks. However, their performance significantly declines when generating content at resolutions higher than those used during training. Although numerous methods have been proposed to enable high-resolution generation, they all suffer from inefficiency. In this paper, we propose RectifiedHR, a straightforward and efficient solution for training-free high-resolution synthesis. Specifically, we propose a noise refresh strategy that unlocks the model's training-free high-resolution synthesis capability and improves efficiency. Additionally, we are the first to observe the phenomenon of energy decay, which may cause image blurriness during the high-resolution synthesis process. To address this issue, we introduce average latent energy analysis and find that tuning the classifier-free guidance hyperparameter can significantly improve generation performance. Our method is entirely training-free and demonstrates efficient performance. Furthermore, we show that RectifiedHR is compatible with various diffusion model techniques, enabling advanced features such as image editing, customized generation, and video synthesis. Extensive comparisons with numerous baseline methods validate the superior effectiveness and efficiency of RectifiedHR.

Cong Liu, Liang Hou, Mingwu Zheng et al.

Resolution generalization in image generation tasks enables the production of higher-resolution images with lower training resolution overhead. However, a significant challenge in resolution generalization, particularly in the widely used Diffusion Transformers, lies in the mismatch between the positional encodings encountered during testing and those used during training. While existing methods have employed techniques such as interpolation, extrapolation, or their combinations, none have fully resolved this issue. In this paper, we propose a novel two-dimensional randomized positional encodings (RPE-2D) framework that focuses on learning positional order of image patches instead of the specific distances between them, enabling seamless high- and low-resolution image generation without requiring high- and low-resolution image training. Specifically, RPE-2D independently selects positions over a broader range along both the horizontal and vertical axes, ensuring that all position encodings are trained during the inference phase, thus improving resolution generalization. Additionally, we propose a random data augmentation technique to enhance the modeling of position order. To address the issue of image cropping caused by the augmentation, we introduce corresponding micro-conditioning to enable the model to perceive the specific cropping patterns. On the ImageNet dataset, our proposed RPE-2D achieves state-of-the-art resolution generalization performance, outperforming existing competitive methods when trained at a resolution of $256 \times 256$ and inferred at $384 \times 384$ and $512 \times 512$, as well as when scaling from $512 \times 512$ to $768 \times 768$ and $1024 \times 1024$. And it also exhibits outstanding capabilities in low-resolution image generation, multi-stage training acceleration and multi-resolution inheritance.

Tianxiong Zhong, Xingye Tian, Xuebo Wang et al.

Existing latent diffusion models typically couple scale with content complexity, using more latent tokens to represent higher-resolution images or higher-frame rate videos. However, the latent capacity required to represent visual data primarily depends on content complexity, with scale serving only as an upper bound. Motivated by this observation, we propose DCS-LDM, a novel paradigm for visual generation that decouples information complexity from scale. DCS-LDM constructs a hierarchical, scale-independent latent space that models sample complexity through multi-level tokens and supports decoding to arbitrary resolutions and frame rates within a fixed latent representation. This latent space enables DCS-LDM to achieve a flexible computation-quality tradeoff. Furthermore, by decomposing structural and detailed information across levels, DCS-LDM supports a progressive coarse-to-fine generation paradigm. Experimental results show that DCS-LDM delivers performance comparable to state-of-the-art methods while offering flexible generation across diverse scales and visual qualities.

Xunzhi Xiang, Xingye Tian, Guiyu Zhang et al.

Variational autoencoders (VAEs) typically encode images into a compact latent space, reducing computational cost but introducing an optimization dilemma: a higher-dimensional latent space improves reconstruction fidelity but often hampers generative performance. Recent methods attempt to address this dilemma by regularizing high-dimensional latent spaces using external vision foundation models (VFMs). However, it remains unclear how high-dimensional VAE latents affect the optimization of generative models. To our knowledge, our analysis is the first to reveal that redundant high-frequency components in high-dimensional latent spaces hinder the training convergence of diffusion models and, consequently, degrade generation quality. To alleviate this problem, we propose a spectral self-regularization strategy to suppress redundant high-frequency noise while simultaneously preserving reconstruction quality. The resulting Denoising-VAE, a ViT-based autoencoder that does not rely on VFMs, produces cleaner, lower-noise latents, leading to improved generative quality and faster optimization convergence. We further introduce a spectral alignment strategy to facilitate the optimization of Denoising-VAE-based generative models. Our complete method enables diffusion models to converge approximately 2$\times$ faster than with SD-VAE, while achieving state-of-the-art reconstruction quality (rFID = 0.28, PSNR = 27.26) and competitive generation performance (gFID = 1.82) on the ImageNet 256$\times$256 benchmark.

Yuanhui Huang, Weiliang Chen, Wenzhao Zheng et al.

World models have garnered increasing attention for comprehensive modeling of the real world. However, most existing methods still rely on pixel-aligned representations as the basis for world evolution, neglecting the inherent 3D nature of the physical world. This could undermine the 3D consistency and diminish the modeling efficiency of world models. In this paper, we present Terra, a native 3D world model that represents and generates explorable environments in an intrinsic 3D latent space. Specifically, we propose a novel point-to-Gaussian variational autoencoder (P2G-VAE) that encodes 3D inputs into a latent point representation, which is subsequently decoded as 3D Gaussian primitives to jointly model geometry and appearance. We then introduce a sparse point flow matching network (SPFlow) for generating the latent point representation, which simultaneously denoises the positions and features of the point latents. Our Terra enables exact multi-view consistency with native 3D representation and architecture, and supports flexible rendering from any viewpoint with only a single generation process. Furthermore, Terra achieves explorable world modeling through progressive generation in the point latent space. We conduct extensive experiments on the challenging indoor scenes from ScanNet v2. Terra achieves state-of-the-art performance in both reconstruction and generation with high 3D consistency.

Sihui Ji, Xi Chen, Xin Tao et al.

Video generation models nowadays are capable of generating visually realistic videos, but often fail to adhere to physical laws, limiting their ability to generate physically plausible videos and serve as ''world models''. To address this issue, we propose PhysMaster, which captures physical knowledge as a representation for guiding video generation models to enhance their physics-awareness. Specifically, PhysMaster is based on the image-to-video task where the model is expected to predict physically plausible dynamics from the input image. Since the input image provides physical priors like relative positions and potential interactions of objects in the scenario, we devise PhysEncoder to encode physical information from it as an extra condition to inject physical knowledge into the video generation process. The lack of proper supervision on the model's physical performance beyond mere appearance motivates PhysEncoder to apply reinforcement learning with human feedback to physical representation learning, which leverages feedback from generation models to optimize physical representations with Direct Preference Optimization (DPO) in an end-to-end manner. PhysMaster provides a feasible solution for improving physics-awareness of PhysEncoder and thus of video generation, proving its ability on a simple proxy task and generalizability to wide-ranging physical scenarios. This implies that our PhysMaster, which unifies solutions for various physical processes via representation learning in the reinforcement learning paradigm, can act as a generic and plug-in solution for physics-aware video generation and broader applications.