Pre-Trained Video Diffusion Models

Test-Time Tuning (TTT) on pretrained diffusion models has emerged as a powerful paradigm for video editing. However, there exists a foundational mismatch between the distribution-mapping nature of generative models and the single-point optimization of standard TTT. In this paper, we demonstrate that this mismatch triggers \textit{Prior Collapse}, a degenerate state where the model discards the text conditions and spatial latents, collapsing generations to the source video, or entangling the features of distinct regions. To resolve this, we propose \textbf{ElasticTTT}, a novel framework that preserves the prior generative distribution and rescues generative elasticity. Specifically, we propose \textit{Target Distribution Regularization} to prevent sharp memorization minima, \textit{Contrastive CFG} to guide inference away from source biases, and \textit{Asynchronous Noise Schedule} to preserve unedited regions. Extensive evaluations, supported by theoretical analysis, demonstrate that ElasticTTT successfully preserves the generative prior of the base model, achieving state-of-the-art performance on one-shot video editing.

Discrete flow matching provides a flexible framework for generative modeling on discrete structures. However, the standard factorized training objective exposes the model to targets of varying difficulty, mixing well-conditioned, predictable tokens with ambiguous, high-entropy ones. We empirically demonstrate that the uncertainty over the value of each token is closely related to the density of available context in its neighborhood. Motivated by this observation, we propose a simple modification to the underlying continuous-time Markov chain (CTMC) that incorporates local context information. Our context-weighted sampler improves generation quality with negligible computational overhead, while our scaled cross-entropy loss function reweights the training signal from different tokens and reduces generative perplexity by up to 63% on OpenWebText. Moreover, our approach matches a strong semi-autoregressive block diffusion baseline in quality while retaining the ability to perform generation in any order. These results highlight the role of local context as an important factor in discrete generative modeling and show that simple context-aware modifications can significantly improve both sampling and training efficiency.

Current identity customized video generation methodologies are predominantly limited to single-identity scenarios, as the lack of explicit identity separation mechanisms often leads to identity confusion in multi-identity settings. Existing multi-identity approaches, which directly extend single-identity frameworks by concatenating face images as input conditions, frequently result in unnatural facial expressions and motions, manifesting as the "copy-paste" phenomenon. To overcome these limitations, we introduce GroupVideo, a novel framework that leverages multiple individual photographs to generate identitycustomized video. Built upon Video Diffusion Transformers, GroupVideo incorporates multimodal identity alignment: visual alignment jointly encodes multiple face images to provide robust identity references, while semantic alignment introduces a semantic perceiver to enhance the naturalness of motions. An ID localization module with spatial guidance is introduced to address identity blending and enhance identity fidelity, along with bounding box constraints and mask regularization loss, to focus on facial regions and improve training efficiency. In response to the shortage of multi-ID video datasets, we have curated a comprehensive high-quality dataset of 20,000 videos, thereby establishing a crucial resource to advance future research in multi-ID video generation. Extensive experiments demonstrate that GroupVideo outperforms existing methods in generating multi-character videos with consistent identities and natural motions.

This paper proposes the Distribution-Alignment Bridge (DAB), a framework that reconceptualizes text-to-video retrieval as a distribution alignment task rather than traditional deterministic point matching. By modeling both text and video embeddings as Gaussian distributions defined by mean and variance, DAB explicitly accounts for modality-specific uncertainty. We employ a deterministic, diffusion-inspired bridge to iteratively refine text distributions toward their target video distributions through a truncated refinement process. This approach unifies probabilistic embedding and distributional transformation into a cohesive, end-to-end trainable system. To optimize cross-modal similarity, we introduce a distribution-aware contrastive loss based on Kullback-Leibler divergence. Extensive evaluations on MSR-VTT, MSVD, and VATEX benchmarks confirm that DAB significantly outperforms existing probabilistic and diffusion-based baselines, while providing calibrated uncertainty-aware ranking through bridge-induced distributional margins.

Streaming video diffusion models have made substantial progress toward interactive and dynamic world simulation, but the nested autoregressive and denoising loops of conventional next-frame generation hinder real-time deployment. Recent rolling-window methods pipeline denoising across multiple consecutive frames at different noise levels, improving throughput and long-horizon stability. However, they tokenize every state at the same fine spatial granularity, leaving substantial noise-dependent redundancy in the joint denoising window. We propose Ms.Forcing, an efficient streaming video generation paradigm that adapts spatial granularity to each state's noise level. Its Multi-Scale Patchification (MSP) assigns coarser patches to noisier states, reducing the active-window token count by 45%, while Multi-Scale Self-Attention (MSSA) matches the density of visible non-sink keys and values to each query scale to further reduce attention cost. Because both schedules are fixed by window position, Ms.Forcing retains a static, hardware-friendly computation graph. We further introduce Homogeneous-Noise-Level DMD (H-DMD), which assembles each fake video from clean predictions sharing the same source noise level, thereby reducing the mismatch between DMD training sequences and inference-time rollouts. The multi-scale design helps offset the additional training cost of backpropagating through overlapping windows. We include both quantitative and qualitative experiments to show that Ms.Forcing reaches 22.84 FPS on a single H200 GPU, 39.6% faster than Rolling Forcing, while significantly improving VBench scores in both short video and long video generation setting.

Real-world video deblurring remains challenging due to diverse motion patterns, complex degradations, and the scarcity of realistic training data, yet robust restoration is critical for downstream pipelines such as mobile imaging and 3D reconstruction. This work presents \textbf{RealVDeblur}, an efficient generative framework designed to improve in-the-wild robustness under diverse real capture conditions. First, a large-scale, physically grounded blur synthesis pipeline is constructed from scene-level 3D Gaussian Splatting (3DGS) assets and high-frame-rate videos, providing realistic training data covering both camera-induced and object-motion blur. Second, a video diffusion prior is leveraged for restoration; to better accommodate frame-dependent blur variations, temporal compression in the VAE is disabled and a frame-wise encoding scheme is adopted. For practical deployment on long videos, multi-step diffusion sampling is distilled into an efficient one-step generator, and a training-free Temporal Window Mask stabilizes inference beyond the training horizon with constant memory usage. Extensive experiments on diverse real-world benchmarks demonstrate strong perceptual quality, semantic fidelity, and temporal consistency on unseen videos, as well as improved robustness in downstream 3D reconstruction under severe motion blur. Project page: https://rbjin.github.io/RealVDeblur

Recent autoregressive video diffusion methods are increasingly built upon Self Forcing, where the student is trained on histories produced by its own rollout rather than ground-truth video contexts. This reduces exposure bias, but the historical key-value cache is still used by future frames only as frozen rollout state. As a result, future losses cannot supervise how earlier generated latents should be written into more useful keys and values for later video-latent generation. We call this the historical context-gradient gap. We propose Self Gradient Forcing (SGF), a two-pass training strategy that restores this missing supervision signal without backpropagating through the full serial rollout. Pass 1 performs a no-gradient autoregressive rollout matching inference and, at a sampled denoising exit step, records both the self-generated context and the noisy latents fed to the model. Pass 2 performs parallel context-gradient reconstruction for the recorded exit step. The generated context is used as stop-gradient clean-latent input, while the model recomputes the context KV representations and future-to-context causal attention. Thus, SGF provides the missing memory-writing supervision within the native autoregressive training objective, using losses on future video latents to train the model to encode context into more effective causal memory. Across extensive long-horizon frame-wise and chunk-wise experiments under different initializations, SGF achieves stronger native long-video extrapolation than Self Forcing, especially in subject identity, background/layout consistency, and temporal stability. Remarkably, using only a 5-second training window, SGF can extrapolate to videos lasting several minutes. Code and models will be released to advance research on autoregressive video generation.

Autoregressive (AR) video diffusion models have become a promising paradigm for long and streaming video synthesis, but the continuously growing Key-Value (KV) cache makes attention the dominant inference cost, especially at high resolution where each frame contributes many tokens. Existing remedies either evict the cache with coarse heuristics that cause inter-frame flickering, or require model re-training. We propose HeadCast, a training-free, plug-and-play acceleration framework built on the observation that a pre-trained AR model's attention heads exhibit stable, heterogeneous behaviors. After a short warm-up, HeadCast performs a one-time classification at the maximum-noise step that sorts every head into one of four archetypes: Sink, Dummy, Spatial, and Global, and restructures the monolithic KV cache into head-specific pathways. Crucially, it retains the Global heads that preserve the long-range temporal consistency aggressive eviction destroys. Because the Spatial pathway operates on a fixed-size grid, its savings grow with resolution: across state-of-the-art AR models, HeadCast accelerates inference by up to 1.62x at 720P and 1.95x at 1080P, while keeping VBench quality on par with full attention and largely flicker-free. Code is available at https://github.com/sjlgaga/HeadCast .

Text-guided video editing with diffusion models is impractically slow, hindered by costly multi-step sampling and inversion. We present OSVE, the first framework to successfully adapt one-step Text-to-Image (T2I) models for high-quality video editing, addressing the core challenges of inversion, editability, and temporal consistency. To bypass slow iterative inversion, we train a learnable encoder that predicts the initial noise for each frame in a single forward pass. This encoder is trained with a novel Structure-Aware Editing (SAE) loss on a curated dataset of structurally-aligned image pairs, teaching it to preserve the source video's geometry during edits. For temporal coherence, we introduce Unified-Frame Editing (UFE), a technique that concatenates frame latents to facilitate cross-frame attention in a single generation step. Furthermore, for long videos, a sliding-window strategy with an anchor frame maintains global consistency. Our extensive experiments demonstrate that OSVE achieves editing quality comparable or superior to state-of-the-art multi-step methods, while operating approximately 155--171 times faster. This breakthrough paves the way for practical, real-time video editing applications. Code is available at https://github.com/KU-VGI/OSVE.

VLMs are increasingly deployed in AD systems, creating an urgent need for rigorous safety evaluation under rare yet safety-critical scenarios. Among these, interactions with vulnerable road users represent a major source of real-world failures. However, existing safety-critical scenario generation methods predominantly rely on simulator-based pipelines, which suffer from a substantial sim-to-real gap and often fail to capture realistic, diverse, and unforeseen human-vehicle interaction dynamics. We present SafeGen, a goal-conditioned diffusion framework for safety-critical scenario generation in VLMADs. Our key insight is to formulate scenario generation as a goal-conditioned diffusion process, where a predefined catastrophic end-state serves as a strong supervisory signal, guiding the generation of temporally coherent video trajectories that naturally evolve toward safety-critical outcomes. Building on this formulation, we introduce Context Grounded End State Reasoning, which leverages VLMs to analyze benign driving contexts and infer latent vulnerabilities in human-vehicle interactions, producing structured end-state specifications that induce high-risk scenarios. Conditioned on these targets, we further propose End State Conditioned Video Evolution, which grounds semantic threats into physically plausible visual dynamics. Specifically, we instantiate high-risk agents within the scene via depth-aware geometric projection, followed by boundary-conditioned diffusion to generate intermediate frames with consistent motion patterns and temporal coherence. Extensive experiments across 3 VLMADs demonstrate that SafeGen increases the Judge Overall Score, a metric using a VLM judge to evaluate VLMADs' understanding and decision-making, by 24.25% on average compared to SoTA baselines. Furthermore, fine-tuning a VLMAD improves performance in real-world driving scenes by an average of 15.9%.

Text-to-video generation has advanced significantly over the past five years through scaling of model size, data, and compute. Unlike model architecture, training data is often underexplored. Real-world data curation is complex and non-trivial, involving clip selection from raw videos and captioning to create video-text pairs for learning text-to-video mappings. We study how data distribution and caption quality impact text-to-video models. To enable controlled experiments, we introduce Moving Alphabet, a procedural testbed that renders letters with varying fonts, colors, sizes, and positions, moving in different directions and speeds against a black background. This design allows precise control over data distribution and caption quality by corrupting ground-truth metadata. Our experiments yield three findings: a) a diverse and balanced distribution of video content and duration is critical for generalization; b) caption quality significantly affects both model performance and training efficiency, suggesting that text-to-video models are bounded by video understanding capabilities; and c) classifier-free guidance and fine-tuning on high-quality data provide partial recovery from models trained on corrupted captions, but cannot fully compensate for poor pre-training data. We believe these insights can inform the development of large-scale text-to-video models, and we advocate for greater attention to the science of pre-training data.

Most existing video compression algorithms follow a paradigm of transformation and quantization, optimizing the trade-off between distortion and bitrate. However, extremely low-bitrate compression remains an underexplored frontier where perceptual quality optimization under severely constrained coding resources has not been adequately addressed. In this paper, we propose a unified generative framework that leverages pre-trained Diffusion Transformer (DiT) priors to achieve high perceptual quality at extremely low bitrates. We first introduce a flexible Group-of-Latents (GoL) strategy within the latent space of a causal tokenizer, explicitly partitioning the latent stream into intra $I$-latents and inter $P$-latents. The Deep Compression Module (I-DCM) then encodes key $I$-latents to preserve perceptual anchors with minimal overhead. Building upon these anchors, the DiT-based Unified Latent Denoising Module (U-LDM) refines intra-frame textures and synthesizes $P$-latents from noise, reconstructing temporal dynamics at zero additional bitrate cost. Extensive experiments demonstrate that our method uniquely operates in the extreme-low-bitrate regime (e.g., (<0.005) bpp), achieving state-of-the-art perceptual fidelity with rich spatial details and robust temporal consistency. The code will be made publicly available.

Streaming autoregressive diffusion makes minute-scale video synthesis practical, but its bounded context and fixed denoising schedule allocate resources uniformly across a highly non-stationary sequence. A rolling key-value cache forgets distant visual evidence even when that evidence remains important, while every generated chunk receives the same number of denoising passes irrespective of its actual difficulty. We introduce Surprise Forcing, a training-free framework that treats both limitations as online resource-allocation problems. A Surprise-Gated Memory Bank summarizes evicted frames with value-token descriptors, evaluates them using complementary global-deviation and nearest-neighbor novelty signals, and regulates admission through a feedback-controlled budget in normalized score space. Priority-based replacement and relevance-aware routing then keep the external memory compact and useful. In parallel, Surprise-Aware Denoising estimates chunk difficulty from the maximum adjacent-frame cosine distance after the first denoising pass and uses a local percentile scheduler to skip intermediate steps for comparatively easy chunks. Experiments on VBench, VBench-Long, and VBench-2.0 show that the proposed allocation strategy improves long-horizon consistency and visual quality while retaining real-time streaming throughput.

In line with the prevailing direction of vision research, we explore the integration of both generation and editing capabilities for video and image modalities within a single model. Current approaches to collecting video editing data typically depend on labour-intensive, time-consuming curated procedures--involving object mask annotation, the use of error-introducing pair synthesis via I2V model and ControlNet-like guidance, and VLM-based quality filtering or refinement--and demonstrate limited task scalability. As a result, the diversity of editing tasks remains substantially narrower than that available for image editing models. We develop a pixel-pair temporal warped flow field that can directly generate corresponding video editing samples in real time from image editing samples, and we demonstrate across multiple levels of video editing tasks that a model can learn video editing using only such data. We regard the image modality as a particular form of the video modality. Accordingly, we design a modality mimic generation loss and a modality mimic editing loss to relatively align the capabilities--and thereby the output distributions--of the two modalities through mutual imitation. Moreover, language-based visual editing entails the comprehension of the editing instruction and the reference visual content, the localization of the region corresponding to that instruction within the reference visual contents, and the modification of that region alone. Existing approaches predominantly rely on external aids, such as fine-tuning an additional MLLM or explicitly supplying a mask sequence as auxiliary input during inference. In contrast, we aspire for the model to internalize this capability. To that end, we introduce sense-related tasks--for instance, referring expression segmentation--along with corresponding editing-region-aware latent-level loss and attention-level loss.

Recently, text-to-video (T2V) models have been widely deployed, sparking growing concerns over their robustness against jailbreak attacks. Existing jailbreak methods, mostly adapted from text-to-image attacks, suffer notable drawbacks when applied to T2V systems. They fail to fully leverage temporal consistency, an inherent characteristic of video generation. Besides, these methods demand heavy video query optimization, which is infeasible in practical black-box scenarios. Their adversarial prompt search is also driven by heuristic local signals, lacking principled structured exploration strategies. To tackle these limitations, we propose BSB, a structured, query-efficient jailbreak framework for T2V models. BSB harnesses temporal consistency by encoding harmful intent as the transition between two individually harmless boundary states. Under this paradigm, the attack targets boundary-state pairs whose interpolation tends to produce unsafe intermediate frames during video generation. Directly evaluating all candidate pairs within the video space incurs prohibitive computation cost. Instead, BSB conducts Monte Carlo Tree Search (MCTS) in a cheaper textual proxy space and regularly calibrates search outcomes with sparse video-level evaluations. We conduct comprehensive experiments on mainstream commercial T2V models including Veo 3.1, Sora 2, Seedance and Kling v1. Results show BSB surpasses all existing jailbreak baselines, delivering an average 18.6% relative gain in attack success rate over the strongest competitor across evaluated models. Our findings identify temporal consistency as an understudied yet vital attack surface for T2V models and verify that structured search facilitates effective vulnerability discovery under constrained query budgets.

Text-to-video (T2V) generation models have achieved strong visual realism, but improving physical plausibility can come at the cost of semantic consistency with the input text. This tension arises because physical preference is typically determined by comparing dynamics between two videos, without accounting for whether either video faithfully depicts the scene specified by the prompt, making physical-semantic conflict a systematic tendency under this supervision paradigm. We formulate this challenge as a constrained preference optimization problem and propose Physical and Semantic Direct Preference Optimization (PSDPO), which modulates each preference pair's contribution based on the agreement between its physical and semantic signals. A gradient-level analysis shows that PSDPO bounds the semantic drift from conflicting pairs to a controllable residual, and further motivates a staged optimization protocol that provably reduces cumulative drift. The resulting method operates entirely within the standard DPO framework, requiring no auxiliary models or additional loss terms. Experiments show that PSDPO improves physical plausibility by up to $2\times$ over the baseline on VideoPhy-2, while maintaining strong semantic consistency on VBench, achieving a more reliable balance than existing preference-based methods.

High-resolution video diffusion models built on Diffusion Transformers (DiTs) deliver strong fidelity but quickly exhaust the memory budget of a single workstation. A 100 billion-plus parameter DiT easily requires over a terabyte of persistent state, while naive spatiotemporal self-attention grows quadratically in sequence length. These two walls -- parameter memory and activation memory -- prevent researchers from adapting massive generative models without large GPU clusters. We revisit this problem from a systems perspective and introduce MegaSlide-DiT, a prototype that demonstrates how a pre-trained 105B DiT can be adapted on a single H200 GPU with 1.5 TB of host RAM. Our key insight is that the GPU need not own the model state: all persistent weights, master weights and optimizer moments remain in host memory, while only transient shards are streamed to the GPU on demand. Simultaneously, we replace quadratic global attention with 3D Deformable Slide Attention (3D-DSA), a motion-adaptive local attention operator that reduces both memory and computational complexity to linear in the sequence length. We report detailed memory accounting, execution traces and evaluation results to substantiate our design. MegaSlide-DiT does not claim to train a 105B model from scratch on a single GPU, nor does it magically solve bandwidth limits; rather, it offers a pragmatic path for full-parameter adaptation of massive video diffusion models on high-end workstations.

Modern video generation models are increasingly hailed as emerging world models with an internalized grasp of physical law. Yet existing benchmarks largely evaluate physical plausibility only at the output level, without verifying whether the model arrives there through a faithful, law-grounded reasoning process. We introduce Apple-PI, the first benchmark that anchors video-model evaluation explicitly in physical laws. Apple-PI comprises three components. 1) Orchard: a dataset of 400 videos covering ten canonical tasks in classical mechanics. It separates single-law tasks for confounder-free diagnosis from multi-law tasks for probing generalization. 2) Benchmark Protocol: a three-stage protocol based on scientific reasoning, including Perception, Formulation, and Deduction. It uses chain-of-frames prompting on infographic-annotated first frames, treating the generated video as the model's visible reasoning trace. 3) Evaluation Suite: a hybrid evaluation suite that combines MLLM-based subjective scoring with physics-law-grounded objective measures. This enables stage-resolved diagnosis of not only whether a model fails, but where it fails. Benchmarking 11 models shows that current video models remain far from reliable law-grounded world simulators, with the best video model scoring only 0.473. Our stage-, pillar-, and source-resolved analyses further expose a Perception-to-Formulation-to-Deduction bottleneck, weak multi-law state transfer, and a persistent Sim-to-Real gap. These findings position Apple-PI as a diagnostic foundation for guiding future video models toward world models with law-grounded physical intelligence.

Video Diffusion Transformers process long spatio-temporal sequences, making self-attention the main bottleneck in high-resolution video generation. Training-free sparse attention reduces this cost, but adaptive Top-$p$ routing creates uneven per-head workloads under multi-GPU sequence parallelism. The resulting workload heterogeneity turns sparse attention into a rank-level straggler problem. We present \method{}, a training-free sparse-attention system that improves the distributed execution efficiency of adaptive sparse attention under multi-GPU sequence parallelism. \method{} uses Top-$p$ routing, a Top-$k$ safety floor, and video-aware block organization as the sparse-routing frontend, then repairs the materialized mask at runtime. Runtime Load Balancing migrates a small number of heavy heads via P2P communication to shorten the current critical path. Slack-Aware Sparse Augmentation fills residual non-critical-rank slack with additional high-value blocks, while overlap hides scheduling and migration overhead behind existing computation. On step-distilled Wan2.2 I2V, \method{} reduces average load imbalance from 1.34 to 1.08 and delivers a $4.41\times$ attention speedup over FlashAttention, while achieving a $2.02$--$2.11\times$ DiT inference speedup with competitive video quality.

Autoregressive video diffusion models have enabled the generation of arbitrarily long videos by removing conditioning on future frames, thus greatly improving computational efficiency. Yet, they suffer from error accumulation over time, as the denoised sequence gradually drifts away from the conditioning distribution seen during training. Recent advances attempt to reduce this error by anchoring each generated frame to the learned manifold of real ones. However, even when all generated individual frames lie close to the real manifold, there are trajectories which the model lacks sufficient knowledge to continue without exiting it, thus reaching a terminal point. To prevent the model from being trapped in terminal points, we start from the hypothesis that for well-modeled future trajectories the distribution of the predicted noise should match the one of the forward noising process. To enforce such a prior at test time, we introduce Terminal points Avoidance through Noise Guided Optimization (TANGO), which uses the diffusion model as a critic of its own outputs, by predicting one step forward and requiring an isotropic Gaussian noise prediction. We use the deviation from this expected noise distribution to search for an alternative trajectory that does not lead to a terminal point. Our approach achieves a $3.1\%$ absolute improvement on VBench over state-of-the-art, while reducing Fréchet Video Distance by $28.3\%$ on average across $15$s videos. Our code is available on https://mever-team.github.io/tango.

Maintaining recurring character identities across scene transitions and long temporal gaps is a central challenge in narrative long video generation. Methods targeting global consistency often retrieve memory using cues that are not aligned with character identity preservation, while recent character-centric variants still rely on coarse frame-level kv memory that entangles identity with incidental visual factors and lacks a continuous update mechanism under limited memory capacity. To address these limitations, we propose \textbf{SlotMem}, a character-addressable internal memory framework for multi-character narrative long video generation. Specifically, SlotMem uses a Character-Semantic Probe to localize character-relevant visual tokens from cross-attention responses, and a Memory Encoder to compress DiT tokens into compact role-wise slot memory. As generation proceeds, a Memory Writer conservatively updates each character's memory with new observations, while Character-Wise Cross-Attention retrieves the role memory and injects it only into localized tokens of the same character. Experiments on multiple narrative long video generation benchmarks show that SlotMem improves long-range character consistency over existing baselines, while maintaining comparable video quality. Our code is available at https://github.com/YilaiLiu-HKU/SlotMem.

Recent advances in physics-grounded video generation leverage physics simulation as a physical prior to guide video synthesis toward physically plausible outcomes. The simulation process is controlled by physical specifications, which are typically generated by a vision-language model in a single pass. Such one-shot prediction often fails to accurately translate user intent into executable simulations, particularly for fine-grained object dynamics, complex motion trajectories, and temporally structured interactions. In this paper, we propose PhysAgent, a reflective agentic framework that closes the loop among physical program generation, physics simulation, stage-specific verification, and targeted program repair. Beyond improving the control of coupled physical parameters, our framework enables the agent to progressively realize complex trajectories, multi-stage interactions, and precise event outcomes by treating each physical program as an executable hypothesis. In addition, we design a set of physics-control APIs to support more stable and complex motion behaviors. Extensive experiments demonstrate that PhysAgent produces more physically plausible videos, achieves better prompt alignment, and generalizes more effectively across diverse physical scenarios.

Diffusion Transformers have recently achieved strong performance in video generation, yet controlling scene geometry under viewpoint changes and camera motion remains challenging. In this work, we revisit the role of positional encoding in video diffusion transformers and show that it provides a useful spatial bias for geometry-aware control. Specifically, if reference tokens are encoded according to their projected locations in the target view, the denoising model is encouraged to retrieve content from position aligned regions of the input video. Building on this observation, we introduce a geometry-aware cross-attention mechanism that enables target video latent tokens to attend to structured context tokens derived from reference images or frames. To establish correspondence between the reference content and the target camera trajectory, we equip the context tokens with a projected positional encoding scheme that combines target-view 2D reprojection with depth-aware disambiguation. At the same time, we preserve the original spatiotemporal positional encoding of the generated video latent, allowing geometric guidance to be injected while maintaining consistency with the video model's native latent structure. The resulting framework provides a simple and effective approach for controllable video generation. It improves spatial controllability in viewpoint-dependent editing tasks, including camera re-trajectory, novel-view video synthesis, and geometry-aware video editing, while preserving the generative prior of the underlying video diffusion model. The code is available at: https://github.com/MTLab/PE-Field.

Video object removal is a fundamental yet challenging task in video editing. Despite recent progress, existing methods typically fall into two categories. Traditional approaches based on optical flow or attention mechanisms often introduce noticeable artifacts and yield unnatural results. In contrast, diffusion-based methods improve visual realism but demand multiple denoising steps, limiting their practicality. To address these issues, we propose From-Draft-to-Draft-Free (D2DF), a framework that distills the ability of transforming coarse drafts into refined videos into a one-step video generation model. Within D2DF, a teacher model is trained to refine low-quality removal results ("drafts") into high-fidelity videos by multiple steps. Then, through Prior-Privileged Consistency Distillation (PPCD), we distill this capability into a student model that performs one-step removal conditioned on the draft. To eliminate draft dependency, we introduce a Self-Guided Fast Planting (SGFP) module based on our Temporal Masked Transformer that autonomously generates scene-consistent pseudo-drafts in latent space, enabling a fully draft-free one-step model. Extensive experiments show that both draft-conditioned and draft-free versions achieve state-of-the-art performance on multiple metrics, surpassing traditional and multi-step generative methods in both quality and efficiency. The denoising process for a single video takes only about 1 second.

State-of-the-art flow based text-to-image (T2I) models exhibit remarkable generative abilities but remain vulnerable to producing unsafe content. Prior safety efforts range from concept erasure and prompt filtering to classifier-based gating. However, simple techniques like parameter efficient adaptations of the models easily bypass such guardrails. We introduce a unique principled approach that achieves safety by regulating the model's attention dynamics through inference-time introspection, exhibiting intrinsic robustness. Our method analyzes and rebalances attention activations throughout image synthesis, steering generations away from unsafe concepts while preserving semantic alignment. This introspective control ensures safety of deployed models. Across standard and adversarial safety benchmarks, our approach achieves remarkable safety scores while maintaining or even improving alignment and perceptual quality. Our results reveal that attention-space regulation offers a considerably more promising path to safer diffusion transformer based image generation than the existing concept erasing mechanism.Our code can be accessed at https://basim-azam.github.io/iam/

Music generation foundation models have recently attracted significant industry attention. However, achieving efficient generation and high-fidelity long-form audio while supporting controllability remains challenging. To address these needs, we present \textbf{WanSong}, a simple yet powerful approach for long-form, commercial-grade song generation. Unlike autoregressive (AR) and cascaded multi-stage pipelines (\eg, AR followed by diffusion), \textbf{WanSong} is a pure diffusion-based model that directly generates high-fidelity, multilingual songs up to 5 minutes and outputs dual stems (vocals and background music) in a single run. In addition, our diffusion framework enables faster inference through step-distillation, and offers an efficient pathway for fine-tuning and customization to support downstream editing tasks.

Recent diffusion-based video generation models have made significant progress in multi-reference image-conditioned video editing. However, existing methods still struggle to coordinate information from multiple visual sources accurately. We identify a critical deficiency in existing approaches. Existing editing instructions lack explicit reference relationships, and most multimodal large language models (MLLMs) cannot generate them reliably. To address this problem, we propose ReBind, a systematic framework that introduces semantic instructions with embedded reference tokens as the intermediate representation for multi-reference image-conditioned video editing. Our key insight is embedding reference tokens at semantic positions to eliminate ambiguity and establish precise bindings between visual attributes and their sources. We develop ReBind-Instruct, a specialized MLLM that learns to establish explicit bindings between visual attributes and their reference sources through a two-stage progressive scheme for precise reference relationships. We further develop ReBind-Edit, which enables lightweight adaptation of text-to-video models to coordinate multiple references by binding visual attributes to their designated sources. Extensive experiments demonstrate that ReBind substantially outperforms general-purpose MLLMs in instruction quality and achieves state-of-the-art performance among open-source methods on reference image conditioned video editing. Our project webpage: https://rebind-mrv2v.github.io/.

This work introduces FilmGPT, an autoregressive transformer designed to address the challenge of video montage--turning a collection of raw, "unwatchable" footage into coherent cinematic sequences. Inspired by language learning in modern LLMs, we train a long-context autoregressive transformer on a large corpus of movies. The aim is to implicitly capture the "grammar" of film directly from data rather than from hand-coded rules. Unlike other generative models, FilmGPT does not generate any new video frames. Instead, at inference time, we introduce a footage-constrained decoding algorithm to select the best next shot from the input raw footage according to the statistical patterns learned from films. We first evaluate these learned statistics directly by using the FilmGPT autoregressive model for next shot prediction on a standard benchmark of shot sequence ordering, outperforming the previous state of the art. We then evaluate our footage-constrained decoding algorithm on the full film editing task via a user study, and find that our FilmGPT-based editing significantly outperforms previous approaches. Finally, we demonstrate the applicability of FilmGPT to a wide range of applications in video montage, from automatic video segment trimming to human-in-the-loop film editing.

Large-scale video diffusion models (VDMs) deliver strong generation performance, but full fine-tuning for downstream tasks incurs prohibitive computational costs. Existing parameter-efficient fine-tuning (PEFT) methods have two critical flaws on billion-scale models: they still require substantial trainable parameters, and reward-based training suffers from noise-induced optimization instability in condition-guided tasks. We propose MagicPrompt, a lightweight framework that achieves extreme parameter efficiency and stable reward optimization. It first adopts Attention-Embedded Prompt Tuning, which steers generation via lightweight soft prompts with orders of magnitude fewer parameters while preserving pre-trained knowledge. It further introduces Dual-Space Reward Feedback Optimization, which uses self-supervised latent objectives to improve condition-guided reward training. Experiments show MagicPrompt reaches competitive performance with less than 1% trainable parameters and notably reduces training costs.

Driving-world generation has emerged as a core capability for scalable autonomous-driving simulation, yet existing methods remain limited in object-level controllability and long-horizon stability. We present M$^\text{4}$World, a Multi-view and Multimodal generative driving world model that synthesizes future surround-view video streams and synchronized LiDAR scans while supporting interactive object Manipulation and stable Minute-long streaming. Fine-grained object manipulation is realized through a flexible conditioning interface that supports explicit control over both the spatial layout and visual appearance of individual objects. Stable minute-long streaming, on the other hand, is achieved through a multi-stage training framework that enables online causal generation in only four denoising steps while maintaining coherent world dynamics throughout extended rollouts. Building on these components, we introduce an efficient few-clip post-training as well as a suite of visual reference-conditioned generation models, preserving general generation ability while allowing rare-case customization for long-tail controllability. To assess controllability beyond realism, we further introduce an automated VLM-based judging pipeline that evaluates scene-level condition adherence, view-wise object controllability, and cross-view object consistency. Comprehensive experiments show that M$^\text{4}$World consistently delivers high generation quality, precise controllability, and stable minute-long streaming. Together with downstream long-tail augmentation and scene editing, these results demonstrate the potential of M$^\text{4}$World for controllable, scalable driving simulation.

Text-to-video (T2V) generators can synthesize realistic and temporally coherent videos, but controllably removing a target concept from a generator remains difficult. Unlike text-to-image concept erasure, T2V unlearning must suppress a target concept that may persist across frames while preserving non-target subjects, actions, scenes, and temporal structure. We propose \textbf{SIRUS}, a training-free inference-time framework for concept-level T2V unlearning. Given textual aliases of a target concept, SIRUS localizes target-related prompt evidence and suppresses target expression during sampling, without updating the text encoder or denoising network. We further introduce a video-oriented evaluation framework for T2V unlearning that separately measures target forgetting, non-target preservation, video quality, jailbreak robustness, and efficiency, using video-level failure criteria, frame-level residue statistics, paired preservation analysis, VBench-based quality diagnostics, and deployment overhead measurement. Across five safety, object, and style concepts on CogVideoX, SIRUS reaches 70.4% average forgetting success and 25.7% average frame hit, compared with 44.4% / 47.2% for VideoEraser, while reducing the average VBench quality drop from -0.043 to -0.016, yielding the strongest forgetting-quality trade-off among fully evaluated baselines. Transfer experiments on Wan2.2 further suggest that SIRUS generalizes across modern T2V backbones.

Video diffusion transformers (vDiTs) generate high quality video but introduce extremely high compute cost due to the long diffusion timesteps and self attention computation. As diffusion timesteps are reduced, the computation cost of self attention becomes the dominant bottleneck. Existing acceleration approaches largely inherit sparse attention techniques from large language models, which fail to consider the unique spatiotemporal correlation of video data. This paper presents Kaleido, an algorithm hardware codesign that accelerates all operations in vDiTs by exploiting channel-wise spatiotemporal correlations in latent space. Based on this insight, we propose a lightweight channelwise reuse algorithm that skips redundant computations by reusing partial results while preserving higher generative quality than prior methods (>17 dB). To efficiently support this algorithm, we design a systolic array like accelerator with reconfigurable processing elements and a lightweight data dispatcher to mitigate irregular sparsity and data access patterns introduced by our reuse algorithm. Evaluations across three mainstream vDiT models show that Kaleido achieves up to 5.9x speedup and 16.0x energy savings over state of the art accelerators.

Sign language is a primary communication channel for millions of Deaf and hard-of-hearing people, yet text-to-signer video generation remains costly because video diffusion models are expensive to train and evaluate. This paper presents Text2Sign, a text-conditioned diffusion model for short sign-language clips that runs on a single NVIDIA L4 GPU. It combines a frozen vision-language text encoder with a 3D encoder-decoder and factorized spatiotemporal attention to reduce the cost of full-video attention while preserving motion coherence. We compare convolution-only and transformer-style backbones, frozen pretrained and task-specific text encoders, and factorized versus full attention. On a signer-disjoint How2Sign split, the best short-run ablation reaches a validation loss of 0.0648, while a longer-run checkpoint reaches 0.00999. On a compact evaluation slice, the latter achieves an SSIM of $0.2403 \pm 0.0238$, a PSNR of $15.11 \pm 0.42$ dB, and temporal consistency of $1.0000 \pm 0.0000$ using 8-step DDIM sampling with a guidance scale of 5.0. It generates a 32-frame, $64 \times 64$ clip in 12.60 seconds, or 2.54 frames per second, with peak inference memory of 3.12 GB. A held-out denoising audit shows only weak prompt sensitivity: removing text increases late-timestep loss from 0.9875 to 0.9891, while shuffled prompts perform similarly to correct prompts. Frozen text conditioning therefore improves short-budget validation loss, but prompt-specific separation remains limited. The system is restricted to low-resolution, short clips and lacks expert linguistic evaluation, so it should be viewed as a single-GPU research baseline rather than a complete sign-language production system. Code is available at https://github.com/xiaruize0911/text2sign.

While recent advances in 3D generation have enabled impressive visual synthesis, existing methods often rely on 2D diffusion supervision without explicit mechanisms for geometric consistency, leading to spatial hallucinations such as duplicated structures and misaligned geometry. These issues become more severe in 4D generation, where maintaining consistency across viewpoints and temporal evolution introduces additional challenges, including jitter, identity flicker, and structural drift. We present \textbf{Hallo4D}, a unified and model-agnostic framework for mitigating spatiotemporal hallucinations in 3D and 4D content generation. Hallo4D introduces a generation-detection-correction paradigm that leverages large multimodal language models (LMMs) to identify and summarize spatial and temporal inconsistencies from multi-view and multi-frame renderings. These insights guide a consensus-driven image-space consistency optimization, where an LMM-based selector evaluates candidate corrections through multi-model voting, without requiring retraining or architectural modifications. To further improve temporal consistency and optimization efficiency, Hallo4D incorporates motion-aware keyframe sampling, LMM-guided initialization, and appearance alignment. We additionally introduce exposure-aware optimization and visibility pruning to enhance robustness under challenging viewpoints. Extensive experiments demonstrate that Hallo4D consistently outperforms strong baselines across diverse 3D and 4D generation settings, providing a scalable and generalizable solution for consistency-aware content generation.

Autoregressive diffusion models have enabled high-quality video generation, yet their sequential nature inherently suffers from error accumulation. In long-horizon video synthesis, minor prediction deviations compound over time, inevitably leading to unconstrained generative drift, structural collapse, and severe visual degradation. To address this, we propose Cycle-World, a novel framework designed for stable and temporally consistent long-video generation. Our approach tackles error drift by enforcing strict temporal reversibility across both the training and inference phases. Theoretically, we demonstrate that forward generative drift can be strictly bottlenecked by a cycle-consistency objective. During training, we integrate an efficient reverse-prediction model to implicitly embed causal constraints into the forward generator, compelling it to produce reversible sequences that tightly adhere to the natural video manifold. At inference time, we repurpose this frozen reverse model as a runtime corrector. Through gradient-based cycle guidance, it iteratively refines the generated latent representations, actively suppressing accumulated errors before they are committed to the historical context. Extensive experiments on the VBench benchmark demonstrate that Cycle-World's dual-phase synergy significantly mitigates error drift, achieving state-of-the-art overall generation quality and long-horizon temporal consistency in 60-second synthesis.

Recent advances in text-to-image (T2I) generation have led to models capable of producing highly realistic images. Yet, reliably evaluating their outputs remains challenging, especially at scale. Existing automatic evaluators, often relying on a static prompt set, struggle to capture subtle failure modes such as partial prompt misalignment, compositional errors, or visually plausible but semantically incorrect generations. In this work, we introduce DynEval, a Dynamic Evaluation framework designed to jointly assess text-to-image alignment and image quality of T2I models. To support scalable training beyond limited human-annotated data, we construct two large datasets. First, we build GenDB, a collection of 500K prompt-image pairs generated from human-written prompts drawn from DiffusionDB using a tiered prompt-model generation strategy. Second, building upon GenDB, we construct DynEvalInstruct, a 250K instruction dataset comprising prompt-image-response triplets distilled from a structured evaluation pipeline that decomposes evaluation into text-image alignment and visual quality reasoning. Using this dataset, we perform full fine-tuning of a compact evaluator through a curriculum learning strategy to effectively distill the superior evaluation capabilities of a larger teacher vision-language model, resulting in DynEval-2B and DynEval-4B. In extensive comparisons against existing evaluators across 11 benchmarks, our evaluator achieves a higher overall correlation with human judgments. Furthermore, it provides fine-grained analysis of the capabilities and failure modes of 36 T2I models across 42 subcategories and 9 semantic dimensions.

Recent Video Super-Resolution (VSR) methods rely heavily on transformers and explicit optical flow, creating computational overhead and custom operations that hinder deployment on hardware accelerators like TensorRT. To address this, we introduce NanoVSR, a scalable, fully convolutional architecture designed for resource-constrained edge devices. Using structural reparameterization, NanoVSR collapses into standard convolutions during inference, ensuring seamless hardware compatibility and negligible runtime overhead. Furthermore, despite lacking explicit motion compensation, it maintains competitive restoration quality by implicitly learning spatio-temporal alignments through progressive training. Evaluated on the REDS4 benchmark, NanoVSR demonstrates an exceptional balance between accuracy and computational efficiency, significantly improving the trade-off for compact architectures. Our NanoVSR-644k baseline yields 28.64 dB PSNR while delivering 27.2 FPS on the NVIDIA Jetson Orin NX 16GB (25W), offering massive speed gains over heavier models. The scaled NanoVSR-1.7M variant reaches 29.15 dB with a throughput of 19.58 FPS, providing superior, edge-optimized upscaling. Code is available at https://github.com/filippawlicki/nanovsr.

Three-dimensional (3D) molecule generation has been dominated by diffusion models, which achieve strong generation quality but typically require the molecular size to be specified a priori. Recent autoregressive approaches have substantially narrowed the performance gap while naturally supporting variable-length generation and conditioning on partial molecular context. However, balancing unconditional and context-conditioned generation remains challenging. We introduce KRONOS, a latent autoregressive diffusion framework that generates molecules in the latent space of a pre-trained autoencoder, jointly modeling molecular graph topology and geometry, while retaining the flexibility of autoregressive generation. We further introduce a mixed training strategy inspired by Fill-in-the Middle (FIM) paradigm, enabling both unconditional and fragment-conditioned molecular generation within a single left-to-right autoregressive model. Experiments on QM9 and GEOM-Drugs demonstrate that KRONOS achieves leading unconditional generation performance among autoregressive methods, while remaining competitive with diffusion models. Moreover, fragment-conditioned generation is achieved with negligible impact on unconditional generation performance, demonstrating that both generation paradigms can be supported within a single architecture.

Generated videos provide useful visual motion priors for robot manipulation, but their visual plausibility does not imply physical executability. A generated video usually lacks metric geometry, grasp grounding, robot kinematic feasibility, and execution-time feedback, which makes direct trajectory replay unreliable in real-world manipulation. This paper presents GenVid2Robot, a rigid-geometric consistency framework that converts generated video motion into executable real-robot manipulation trajectories. Given an initial RGB-D observation and a task instruction, GenVid2Robot samples task-relevant semantic anchors from the real first frame, tracks these anchors through generated video candidates, and verifies whether the resulting 2D motion can be explained by first-frame RGB-D anchors under a sparse relative $SE(3)$ model. In this way, generated videos are treated as uncertain visual motion hypotheses rather than direct robot demonstrations. Only geometrically consistent motion is transferred to the robot. The accepted relative motion is then applied to the real grasp-time TCP pose selected by mask-constrained grasping, producing a grasp-conditioned execution trajectory that is consistent with both the visual motion prior and the physical grasp configuration. To reduce execution mismatch caused by RGB-D noise, calibration residuals, and small contact-induced displacement, a bounded depth-compensation module corrects local depth-direction errors without assuming full online replanning. Real-robot experiments demonstrate that GenVid2Robot improves the reliability of generated-video-guided manipulation by grounding visual motion priors with sparse metric geometry, grasp constraints, robot feasibility checking, and bounded execution feedback.

Driven by next-token prediction, NLP shifted from task-specific models into powerful generalist foundation models. What, then, is the equivalent catalyst needed to achieve a general-purpose model in computer vision? In this paper, we contend that large-scale text-to-video generation serves as a strong pre-training paradigm for computer vision, providing the necessary spatiotemporal priors, vision-language alignment, and scalability required for general visual intelligence. We introduce GenCeption, which leverages a pre-trained video generative diffusion backbone to define a feed-forward perception model, capable of performing various vision tasks steered by text instructions. Empirical results demonstrate that GenCeption achieves state-of-the-art performance across a diverse suite of tasks, including depth, surface normal, and camera pose estimation, expression-referring segmentation, and 3D keypoint prediction, often matching or surpassing specialized models (e.g. DepthAnything3, SAM3, D4RT, VGGT-Omega, Sapiens, David, Genmo, and Lotus-2). Furthermore, the video generative pretrained backbone outperforms alternative pretraining paradigms (e.g., V-JEPA, and Video MAE) under comparable settings. Importantly, GenCeption exhibits preliminary data and model scaling properties along with exceptional data efficiency, where it achieves comparable performance with leading models like D4RT and VGGT-Omega with 7 to 500 less training data. Finally, GenCeption also exhibits intriguing emergent behaviors: a model trained exclusively on synthetic human videos generalizes to real-world footage and out-of-distribution object categories (e.g., animals and robots). These findings suggest that video generation is not merely a synthesis tool, but a foundational path toward generalist vision intelligence for the physical world. Project page: https://genception.github.io

Recovering high-quality video from sparse event streams is a challenging task. Regression methods often blur textures, while existing generative models struggle with long-term stability. We propose LongE2V, a novel approach that leverages pre-trained video diffusion priors to jointly handle event-based video reconstruction, prediction, and frame interpolation. By fine-tuning a foundational video model, our approach achieves high data efficiency and superior perceptual quality. We introduce Autoregressive Unrolling and Adaptive Context Switching to mitigate temporal drift in extremely long sequences. We also propose Reencoding Alignment with Cross Residual Correction to ensure precise bidirectional consistency during frame interpolation. Furthermore, Event Voxel Density Augmentation ensures robustness across varying sensor resolutions. Extensive experiments on real-world benchmarks demonstrate that LongE2V outperforms state-of-the-art methods across all three tasks, exhibiting exceptional temporal coherence and zero-shot generalization. Project page: https://cdfan0627.github.io/LongE2V-page/

We propose OPSD-V, an on-policy self-distillation paradigm for post-training few-step autoregressive (AR) video diffusion models. Existing few-step AR video generators can produce long videos with low latency, but still suffer from error accumulation and weakened motion dynamics during long autoregressive rollout. OPSD-V reduces long-horizon degradation while preserving the original few-step inference path. The key idea is to introduce real long-video data as temporal context during training and use it to provide dense trajectory-level supervision. Specifically, the student follows the exact inference-time rollout, generating each chunk conditioned on its own previously generated KV cache. In parallel, the teacher is evaluated at the same student-visited denoising states, but uses a cleaner AR-consistent temporal cache in which older history can be replaced by real-video context. This provides dense denoising-level corrective targets under on-policy AR cache dynamics, without changing the sampler, number of denoising steps, or inference-time cache mechanism. We apply OPSD-V to representative few-step AR video models, including Self-Forcing and LongLive. Experiments show consistent improvements in visual quality, motion dynamics, and VBenchLong scores. A user study with 10 participants comparing 20 video pairs shows that OPSD-V is preferred over the base models in 66.0% of overall-preference judgments (82.5% excluding ties).

Generating realistic 3D human motions in real-time within interactive applications is key for animation, simulation, and humanoid robotics. While recent offline motion generation approaches offer precise control via text and kinematic constraints, they lack the inference speed required for interactive settings. Conversely, existing online methods enable real-time synthesis but often sacrifice controllability or struggle with complex text semantics and long-horizon goals due to limited context windows. In this work, we introduce ARDY, a streaming generation framework that bridges this gap by enabling high-fidelity motion generation controllable via online text prompts and flexible kinematic constraints. ARDY employs a hybrid representation that combines explicit root features with a latent body embedding, balancing precise trajectory control with efficient generative learning. We propose a two-stage autoregressive transformer denoiser that features variable history context and supports conditioning on flexible, long-horizon kinematic constraints. By training on a large-scale motion capture dataset and being directly conditioned on text labels and kinematic constraints sampled from ground truth poses, ARDY natively learns controllable generation that supports online prompting and flexible long-horizon goals. Extensive evaluations on the HumanML3D benchmark and the large-scale, high-fidelity Bones Rigplay dataset demonstrate ARDY's high motion quality and constraint adherence, validating the efficacy of our key architectural decisions. Finally, we demonstrate the method's practical versatility through an interactive demo featuring dynamic text control, diverse keyframe pose constraints, path following, and interactive locomotion control via mouse and keyboard. Supplementary video results, code, and model releases can be found at https://research.nvidia.com/labs/sil/projects/ardy/.

We study 4D generation to synthesize temporally coherent sequences of 3D geometry for animation and content creation. In contrast to existing SDS-based optimization methods and video-driven animation approaches, we adopt a skeleton-driven animation framework aligned with standard industrial pipelines, which enables explicit control and editing. To this end, we propose SkelGen4D, a weakly supervised feed-forward framework for text-driven mesh animation that generates explicit skeleton motions without requiring per-frame skeleton annotations. SkelGen4D first recovers temporally consistent pseudo-skeletons from animated meshes via differentiable fitting, and then generates text-conditioned skeleton motion sequences in a feed-forward manner, further refined with Motion-GRPO to ensure temporally coherent, physically plausible, and articulated animation. We evaluate our method on two large-scale benchmarks, Truebones Zoo and Diffusion4D. Our results show that our weakly supervised skeleton modeling matches or surpasses fully supervised baselines while scaling to diverse object categories for high-quality text-driven mesh animation. Further, our method supports flexible motion editing and is aligned with standard animation production pipelines.

Generative video foundation models exhibit strong compositional priors, yet world-action models (WAMs) and video-action models (VAMs) often lose these priors after finetuning on robotic action data. We refer to this discrepancy as the video-action generalization gap. In this paper, we systematically investigate this gap by evaluating a comprehensive design space of VAMs, demonstrating that standard design choices yield no emergent explanation pattern. To explain this behavior, we introduce the Temporal Ratio (TR), an attention-based measure of how strongly the action head relies on future latent rollouts relative to the anchored current frame. TR has two key properties: first, a model's structural reliance on future-predictive latents, measured via TR, acts as a predictor of its compositional generalization capacity; second, it natively fluctuates based on task phase, shifting attention to future frames during planning and reverting to the present frame for precise manipulation. Finally, based on these findings, we propose an inference-time adaptive guidance method, which exploits this intrinsic feature attention pattern to dynamically amplify compositional video conditioning signals precisely when the policy relies on future rollouts. Evaluated on the LIBERO benchmark and real-world tasks, our approach mitigates the OOD-ID compositional generalization gap. More details: https://umishra.me/temporal-ratio/

Autoregressive video diffusion enables efficient streaming and long-horizon video generation, but repeatedly reusing generated latents as causal context can amplify temporal errors, resulting in flickering, motion jitter, and structural drift. In this paper, we investigate this failure mode from a spectral kinematic perspective and identify discrete latent acceleration as an effective signal for revealing unstable high-frequency temporal perturbations. To this end, we propose SAGA, a training-free \textbf{\textit{s}}table \textbf{\textit{a}}cceleration \textbf{\textit{g}}uidance approach for \textbf{\textit{a}}utoregressive video generation. SAGA integrates an acceleration domain spectral guidance objective based on finite-window Slepian projections with a structured autoregressive noise initialization strategy that suppresses short-range temporal correlations while preserving long-range motion structure. Without retraining or modifying the backbone, SAGA can be directly applied to existing chunk-wise autoregressive diffusion models, which is the prevalent setting for high-quality generation. Extensive experiments show that SAGA consistently improves temporal quality across multiple autoregressive diffusion models. On Self-Forcing, SAGA improves Temporal Quality from 97.30 to 97.91 and Image Quality from 69.60 to 70.51. Moreover, spectral analysis and human preference studies demonstrate that SAGA reduces temporal instability while maintaining visual fidelity.

Video relighting requires balancing long-form temporal consistency with a physically grounded understanding of light transport, which depends on accurate estimation of intrinsic scene properties such as materials, geometry, and illumination. Existing methods follow two paradigms: (1) reconstruct a video's photometric properties via inverse rendering and relight them to a target illumination via forward rendering, using physically-based rendering (PBR) or a neural renderer; these suffer from noisy reconstructions and struggle with hard-to-model effects such as global illumination. (2) Frame the task as generative video-to-video translation conditioned on relighting targets (a target environment map or text); this limits relighting control and temporal stability, since diffusion models struggle to translate long-form videos, and is constrained by the availability of input/relit training pairs. We propose LightCrafter, a hybrid pipeline that reformulates video relighting as video translation of a proxy video: rather than translating the input video directly to the target, we translate a PBR rendering of the input under the target illumination to the final target. This bakes illumination targets into the PBR proxy, removing the need to teach the diffusion model illumination concepts like environment maps, and enables more intricate lighting control while naturally providing long-form temporal consistency. We show PBR renders alone already outperform some prior art but struggle with effects like global illumination; to capture these, we leverage photometric priors in video generation models by post-training CogVideoX on synthetic video pairs and real-world unpaired videos. We outperform prior state-of-the-art on existing real-world relighting benchmarks and contribute a synthetic benchmark for further analysis. We will release our dataset, benchmark, metrics, and code.

World models are typically trained to predict discrete-time physical dynamics with a fixed step size baked into the model weights, preventing prediction at variable temporal resolutions. This matters for hierarchical planning, sim-to-real transfer, and scientific or game-engine applications that must query the same dynamics at multiple timescales. Hamiltonian Generative Networks (HGN) offer a principled path forward, grounding predictions in a continuous-time energy function that is, in principle, independent of the observation frame rate. In practice, however, their temporal generalization breaks down in non-conservative settings. We show that in externally forced, dissipative environments, HGN rollouts at step sizes beyond the training regime fail due to distinct failure modes, including latent magnitude growth driven by an unconstrained action-force map, and global truncation error accumulation from an under-resolved integrator. We identify a targeted fix for each mechanism and demonstrate stable dynamics prediction at temporal resolutions well outside the training distribution. In a detailed analysis, we recommend several strategies for enabling temporal generalization in continuous-time video generation.

Prior work suggests that diffusion representations capture low-level geometry but struggle with high-level semantics. We demonstrate that state-of-the-art video diffusion models overcome this limitation. By systematically probing their intermediate activations using recent mutual-kNN alignment metrics, we reveal a highly structured latent space where visual representations evolve across both network depth and noise levels. We show that while moderate noise levels yield linearly separable global semantics, fine-grained details persist at lower noise levels but become spatially scattered, requiring attention mechanisms to decode. Building on these insights, we introduce Gen4U (Generation for Understanding), a framework that repurposes these generative representations with a single forward pass. Our experiments establish that frozen, large-scale video diffusion models function as highly competitive video encoders across a wide spectrum of tasks, spanning semantic and non-semantic objectives (video classification, depth estimation, camera pose estimation, image and video captioning). Bypassing fine-tuning, Gen4U unifies the generation and understanding paradigms, achieving strong perception performance while fully preserving the model's ability to generate high-quality video.

Tracking the six-degree-of-freedom (6-DoF) pose of objects and surfaces from monocular video is a long-standing problem in computer vision. To tackle this problem, existing methods require inputs beyond the video itself-such as 3D models, depth maps, object masks, or task-specific learned features-and they struggle with textureless, transparent, reflective, or deformable surfaces. Here, we introduce ProxyPose, which recasts 6-DoF pose tracking as video-to-video translation. Given only a video and a single marked pixel in the first frame, a fine-tuned video diffusion model translates the input into a proxy video-a synthetic video depicting a colored polyhedron undergoing the same local rigid-body motion as the surface region at the marked pixel. Because the proxy's geometry and appearance are known by construction, recovering its full 6-DoF trajectory reduces to classical pose estimation with off-the-shelf solvers. This formulation leverages large-scale video pre-training to absorb the hardest aspects of pose tracking-handling challenging materials, occlusions, and deformations-into the translation step, while operating at the pixel level with no assumptions about object identity, boundaries, or global rigidity. ProxyPose achieves state-of-the-art 6-DoF pose tracking accuracy without the additional inputs required by competing methods and after fine-tuning the video model only on synthetic data. We further demonstrate that ProxyPose extends to face tracking, camera pose estimation, and challenging in-the-wild scenes that are beyond the reach of existing approaches. Project page: https://ruihangzhang97.github.io/proxypose/.

Recent advances in video diffusion have been driven by scaling transformer-based architectures to billions of parameters, substantially improving visual fidelity and motion coherence. In contrast, existing mobile video diffusion models remain limited to relatively small parameter budgets, typically 0.4-1.8B, restricting generation quality. In this work, we show that high-quality mobile video generation does not require small models. Instead, we demonstrate that a server-scale 5B-parameter video diffusion transformer can be deployed efficiently on memory-constrained mobile hardware through recurrent reformulation and structured compression. Starting from Wan2.2-5B, we rely on a recurrence distillation framework that converts video generation into a chunk-wise autoregressive process with constant-memory attention computation. Combined with causal linear attention, the model operates as an RNN at inference time while preserving temporal coherence across chunks. We further propose a learnable attention head pruning method based on binary per-head gates optimized end-to-end using a noise-biased sparsity objective and distillation-based finetuning. Together with sampling-step distillation and memory-optimized VAE decoding, MobileWan becomes the first 5B-scale video diffusion model deployable on a commercial mobile device. Our system generates 5-second 480x832 videos at 16 FPS in 20 seconds end-to-end latency, achieving a VBench score of 83.79 and establishing a new state of the art in mobile video generation. Project page: https://qualcomm-ai-research.github.io/mobilewan

Human-Object Interaction (HOI) video generation aims to synthesize realistic videos of humans manipulating diverse objects, serving as a promising avenue for AI-driven live streaming e-commerce. A primary obstacle in this domain lies in the complexity of modeling fine-grained physical dynamics and the intricate spatial-temporal coordination between human hands and objects. Existing approaches to this problem typically rely on dense temporal guidance, e.g., frame-wise hand-object pose sequences, to strictly control the interaction process. However, such dense guidance incurs high annotation costs and affects motion synthesis diversity. To overcome these limitations, we introduce SparseCtrl-HOI, a novel sparse temporal control framework for HOI video generation. It requires only a few keyframes that capture interaction states at designated timestamps. Specifically, we employ a Time-Controlled Rotary Positional Embedding (TiRoPE) mechanism to temporally anchor these keyframes while preserving their spatial integrity. Subsequently, to govern the dynamics across intermediate frames, we propose a Motion Prior Injection Module that leverages Multimodal Large Language Models (MLLMs) to extract high-level motion priors. This empowers the model to hallucinate logically and physically plausible transitions. Furthermore, we build SparseHOI-5K, a high-quality and richly annotated dataset for HOI video generation with sparse temporal control. Comprehensive evaluations confirm that our method substantially reduces annotation overhead while synthesizing superior live-streaming e-commerce videos. Both our code and dataset are publicly available at https://mpi-lab.github.io/SparseCtrl-HOI.

Recently, diffusion models have achieved considerable success in the text-guided video editing domain. However, existing works often struggle to balance the trade-off between temporal consistency and editability in video editing, with consistency and editability typically being inversely related. To address this, we propose a high-quality video editing framework enhanced for consistency and editability, named EquiEdit, which improves coordinatively the temporal consistency and editability of the edited videos while achieving a balance between the two. In terms of temporal consistency, the proposed temporal Mamba module with a tailored temporal-aware scanning scans fused video sequences following four designed directions, effectively enhancing the inter-frame consistency of edited videos. For editability, we design a noise injection strategy based on the spectral transformation to increase editing flexibility, where the Fourier transform is used to preserve the hidden structure in the initial latent noise used for editing, ensuring inter-frame consistency of the edited video and fidelity to the input video. Extensive qualitative and quantitative experiments demonstrate the effectiveness of our method in terms of temporal consistency and editability, as well as its great fidelity to the input video itself.

While modern video diffusion models excel in visual fidelity, maintaining long-range physical consistency remains a formidable challenge. Conventional pixel-reconstruction objectives mainly focus on appearance details and often fail to capture the underlying dynamics of a scene. To mitigate this, recent efforts have integrated auxiliary modalities (e.g., optical flow) to introduce physics priors via joint training with video appearance. However, these methods have three main limitations: (1) they do not distinguish the different motion patterns of different entity types; (2) joint modeling of visual and auxiliary modalities can cause capacity conflicts and weaken the pretrained visual prior; and (3) auxiliary modalities may accumulate errors during inference. To address these issues, we propose \textbf{VPT}, a fine-tuning framework for improving physical consistency in video diffusion models. VPT introduces a role-aware signal that groups entities into agents, controlled objects, passive objects, and background, so that different physical roles can be modeled more clearly. We further propose a modality-decoupled denoising strategy, where the visual and auxiliary channels are assigned independent noise levels. Together with a loss-weight decay strategy, this design makes auxiliary modalities serve as soft constraints rather than strong dependencies, mitigating recursive prediction errors during inference. We also introduce cross-step auto-guidance to further strengthen physical dynamics. Experiments show that VPT improves physical consistency while preserving visual quality, achieving relative gains of 39.4% in SA and 17.9% in PC on VideoPhy benchmark over Wan2.1-T2V-1.3B, and consistent improvements on VideoPhy-2 benchmark. The project page is available at https://tom-zgt.github.io/VPT.

Achieving simultaneous preference alignment and distillation acceleration in video diffusion models remains an open challenge. Existing methods optimize the two objectives over mismatched representation spaces, where improving one objective often compromises the other. To overcome this, we propose Reward Lightning, a unified framework that aligns and accelerates a video diffusion model within a single shared representation. Its central principle is homology: both objectives are evaluated on identical latent features, which mitigates the gradient conflicts that arise when they are optimized over disjoint representations. As a foundational component, we first introduce a latent reward model (LRM) that scores videos directly in the latent space, without decoding back to the pixel space. Building on the LRM, homologous preference distillation (HPD) reuses this shared backbone to perform adversarial distillation and preference alignment jointly, yielding few-step generators that remain faithful and well aligned. Extensive experiments demonstrate that the LRM surpasses pixel-level and latent-level reward baselines by $11.0\%$ and $14.7\%$ in preference accuracy, and that Reward Lightning generates high-fidelity videos in merely $1$ to $4$ steps, improving the average VBench score by $2.1\%$ while leading in text alignment, motion quality, and visual quality. Project page: https://reward-lightning.github.io.

We propose Perceptual Flow Matching (PFM), a simple yet effective framework for few-step generation in flow-matching models. Rather than performing velocity regression in the conventional VAE latent space, PFM supervises flow matching in a perceptual feature space using pretrained perceptual models. This simple change substantially improves the few-step generation capability of flow-matching models, reducing the number of sampling steps from 35-50 to 4-8 while preserving generation quality. Unlike existing acceleration and distillation approaches, PFM requires neither teacher models nor auxiliary score networks and can be integrated into standard flow-matching training pipelines with minimal modifications. Extensive experiments on image generation, video generation, and image editing tasks demonstrate that PFM consistently produces high-quality results while producing fewer artifacts than existing distillation-based methods. We further show that perceptual supervision shifts the regression minimizer from mean-seeking to mode-seeking, biasing predictions toward on-manifold modes that remain accurate under coarse few-step integration. Our results reveal that standard flow-matching training can naturally yield high-quality few-step generators when supervised in an appropriate representation space. We hope this insight inspires future research into representation-aware objectives for efficient generative modeling.

Recent progress in large-scale generative models has substantially advanced video generation, yet existing methods remain constrained by a rigid inference paradigm. Bidirectional diffusion models excel at global coherence and visual fidelity but suffer from slow inference, while autoregressive models offer efficient and streaming generation at the cost of long-range consistency and exposure bias. We introduce Flex-Forcing, a unified training and inference framework that enables a video diffusion model to seamlessly operate under both bidirectional and autoregressive generation regimes. The core idea is a flexible chunking mechanism jointly defined over the temporal axis and denoising steps. This design allows the model to (1) perform flexible chunking according to different device budgets, (2) perform bidirectional inference across chunks for global structure planning, while generating frames autoregressively within each chunk for efficient and fine-grained synthesis, and (3) perform any-order, any-timestep autoregressive generation without the strict causal constraint. Extensive experiments on multiple video generation benchmarks demonstrate that Flex-Forcing achieves consistently better video quality, long-video stability than strong baselines with a rigid inference schedule, while offering faster inference.

Video Diffusion Transformers (VDiTs) have demonstrated significant capabilities in high-fidelity video generation. However, their ability to produce long-duration videos is fundamentally constrained by the quadratic complexity of the self-attention mechanism. Recent clustering-based sparse attention methods improve the quality-speed trade-off by grouping semantically similar tokens, but their practical efficiency remains limited by two bottlenecks: substantial clustering overhead and low CTA utilization caused by irregular cluster-induced blocks. We propose HyperVAttention (HVA), a training-free sparse attention framework that addresses both bottlenecks jointly. To reduce clustering overhead, we introduce 3D local-window clustering, which exploits the spatio-temporal locality of video tokens to restrict centroid search to fixed local neighborhoods, and implement it with a custom Triton kernel for efficient execution. We further propose a hybrid clustering strategy that performs full clustering only at anchor steps and updates only subset tokens at intermediate steps, leveraging the temporal stability of cluster assignments across denoising steps. To improve CTA utilization, we present hardware-aware cluster merging that minimizes CTA-aligned execution cost through parallel agglomerative merging, improving block density and approximation fidelity by utilizing idle tile capacity. Together, these components reduce clustering overhead, avoid redundant updates, and better align sparse attention with the fixed tile structure of modern GPU kernels. Experiments on Text-to-Video generation show that HVA establishes a new Pareto frontier for training-free sparse attention in video diffusion, reducing end-to-end latency by up to $2.13\times$ while improving fidelity over existing training-free sparse attention baselines.

Dense video captioning aims to generate temporally grounded descriptions of video events, benefiting both event-level video understanding and generation. In this domain, autoregressive video large language models have emerged as a prevalent paradigm due to their strong generative and cross-modal modeling capacity. However, generating dense captions under the token-by-token paradigm severely limits inference efficiency and hinders scalability as video length and event density increase. In this work, we propose a parallelized autoregressive framework that not only improves generation efficiency but also enhances temporally grounded captioning performance. Our key insight is to exploit the weak local dependencies across temporally distinct events to restructure the causal dependency graph, thereby enabling lossless parallel generation. Specifically, tokens with weak cross-event dependencies can be decoded in parallel, while tightly coupled tokens within each event retain sequential decoding to preserve local semantic coherence. To realize this insight, we introduce two key components for lossless parallel decoding: (1) a latent global planning mechanism that automatically learns the event-level structure and produces compact tokens encoding global inter-event causality while adaptively aggregating event-level audio-visual semantics, guiding subsequent dependency restructuring and parallel decoding; and (2) an event-factorized parallel decoding mechanism that effectively balances local focus with global inter-event awareness. Experiments on various benchmarks demonstrate the clear advantage of our approach in both efficiency and performance in omni-modal event grounding and captioning. Project website: https://github.com/showlab/PadCaptioner.

Modern image-and-text-to-video diffusion models can synthesize highly realistic videos by iteratively denoising an initial Gaussian noise tensor conditioned on reference image and text inputs. However, existing approaches still lack precise and unified controllability over both object motion and camera motion within a single generation process. We present UniCaMo, a unified framework that enables simultaneous control of object trajectories and camera viewpoints by directly constructing the input noise of the diffusion model. Specifically, UniCaMo builds a shared 3D-grounded motion-consistent noise space across latent video frames. Sparse 3D point tracks are used to warp the Gaussian noise of the reference frame along desired object trajectories, while a virtual spherical noise representation provides globally consistent noise values for newly revealed scene regions under camera motion. By combining local track-guided noise warping with global sphere-based noise sampling, UniCaMo maintains geometric and temporal consistency under both object movement and viewpoint changes. Because UniCaMo modifies only the input noise, it requires no auxiliary adapters, control branches, or architectural changes to the underlying video diffusion model. With lightweight LoRA fine-tuning on large pretrained video diffusion models, including Wan 2.1 (14B), UniCaMo achieves state-of-the-art results in both video quality and motion controllability on standard controllable video generation benchmarks.