Overview

Video diffusion models synthesize or transform frame sequences by denoising video latents, often in diffusion transformers, conditioned on text, images, reference clips, or structured controls. Researchers use temporal attention, autoregressive chunking, and pose, camera, depth, or physics cues to improve motion, identity, geometry, and interaction; related systems support editing, restoration, avatars, and world-model prediction. Evaluation spans prompt alignment and visual quality, using benchmarks such as VBench and FVD, as well as controllability, physical plausibility, and downstream action performance. A central challenge is balancing fidelity and precise control over long videos with the speed and memory needed for streaming, while limiting drift and physically implausible events.

Latest papers

812 papers72 papers in the last four weeks, down 36% on the four weeks before. 0.5% of all new papers.

Nov 12, 2026

K2V: Generative Keyframing For Physically Plausible Video Generation

Video diffusion models are able to generate samples with high-quality appearance and textural qualities. However, content produced by these methods is known to exhibit significant physical errors. This is because, in a classical video generation setting, all physical reasoning must take place simultaneously and exclusively within the diffusion process.We propose to resolve these issues by redistributing the task of physical reasoning onto a language model-backed keyframe generation module, which allows for explicit reasoning in language, and fine-grained control over the appearance of the video at important moments in the video. We then condition a video generation model with these keyframes, reducing the role of the diffusion model to the simpler task of interpolating between physically-correct intermediate states, rather than inferring an entire sequence with diffusion. Results demonstrate that our approach significantly outperforms existing approaches, achieving state-of-the-art performance on the Physics-IQ Verified benchmark and PhyGenBench.
Nov 12, 2026

TokenStream: Multi-Subject Personalized Video Generation with Token-wise LoRA and Guided Attention

We introduce TokenStream, an optimization-based framework that composes independently learned subject identities in text-to-video generation without a reference-video training dataset or joint optimization of each new subject combination. Learning from a single reference image makes subject appearance difficult to separate from static-image bias, while merging independently trained adapters can mix identities. Token-wise LoRA (ToRA) restricts low-rank updates to the cross-attention value projections of subject-associated text tokens, providing a separate adaptation path for each subject. A T2V-guided stream copies queries and keys from the frozen base model during early denoising to reuse its layout and motion priors while retaining personalized values. We also introduce Multi-DB, a 30-subject, 122-prompt benchmark with appearance-free prompts and varied interactions. On Multi-DB at 1.3B, TokenStream improves identity preservation from 0.422 for DreamBooth to 0.467 and textual alignment from 0.652 to 0.688.
Nov 12, 2026

Selective Generative Restoration for Video Deraining via Uncertainty-Aware Diffusion

Video deraining is challenging due to rain streaks, occlusions, and atmospheric effects. Existing methods mainly follow a deterministic restoration paradigm, assuming that clean scenes can be uniquely recovered from rainy observations. However, rain degradation is spatially heterogeneous. Reliable regions require faithful reconstruction, whereas severely occluded regions contain ambiguous content that require stronger prior-driven inference. Therefore, conventional deterministic restoration often produces over-smoothed results, while global generative restoration may introduce unnecessary hallucinations. We introduce a Selective Generative Restoration paradigm for video deraining, where generative priors are adaptively activated according to restoration uncertainty. Based on this principle, we propose Selective Generative Video Deraining (SGVD), an uncertainty-aware diffusion framework that enables adaptive fidelity preservation and generative recovery. Specifically, SGVD exploits stochastic diversity between dual one-step diffusion predictions to estimate region-level restoration uncertainty, which determines where observation fidelity should be preserved and where generative inference should be activated. Furthermore, an anchor-guided trajectory consistency mechanism is proposed to regularize stochastic diffusion evolution across frames, reducing temporal generation drift and improving video coherence. To achieve practical efficiency, SGVD adopts one-step diffusion inference while maintaining uncertainty-aware generative capability. Extensive experiments demonstrate that SGVD achieves superior restoration performance, perceptual quality, and temporal consistency compared with state-of-the-art video deraining methods.
Nov 11, 2026

TRACE: Trajectory Synchronization for Training-Free Resolution Extension in Video Generation

More text-to-video models are available, but many remain limited to low resolutions. Training them for higher resolutions is costly, while existing training-free methods still struggle with fine detail, scene structure, and efficiency. We propose TRACE, a training-free method that synchronizes low- and high-resolution generation trajectories. During early denoising, TRACE aligns predicted scene content at corresponding noise levels. It computes high-resolution updates from low-resolution predictions and the current high-resolution state, without high-resolution model inference. The high-resolution trajectory retains fine-grained variation for generating new details. Local window attention then refines fine details in a shared high-resolution latent state. TRACE checks suspected duplicate subjects against the corresponding low-resolution layout and corrects selected regions and their solver history while preserving detail elsewhere. Experiments across all kinds of video generation backbones and resolutions show improved perceptual quality and text-video alignment over FreeSwim, a state-of-the-art training-free method, with 2.0-2.9x end-to-end speedups. Our code and models will be released at https://github.com/libozhu03/TRACE.
Nov 11, 2026

Track4D: Representing Dense 3D Tracking for Video Diffusion Models

We study how to repurpose video diffusion models (VDMs), pretrained on millions of videos, for dense 3D point tracking, where training data is scarce. Doing so is non-trivial, as it requires representing 3D tracks as video-like signals that may lie outside the natural-video distribution on which the VDM was trained. We thus systematically assess the effectiveness of two such video representations. The first, Residual 3D Tracking Video (RTV), encodes per-pixel 3D offsets relative to a query frame and allows recovery of 3D trajectories directly from the generated video. The second, Normalized Coordinate Map Video (NCMV), labels each pixel with a canonical coordinate invariant through time. This is closer to a natural video and easier for the VDM to learn, but requires additional post-processing with an off-the-shelf 2D tracker to recover 3D tracks. RTV tracks more accurately, while NCMV shows that a VDM can generate dense correspondence as a video. We instantiate both representations with LoRA adaptation of a frozen Wan backbone, conditioned on point maps from an off-the-shelf 3D reconstructor as a geometric scaffold. Trained with a single dynamic 3D tracking dataset, Track4D achieves competitive zero-shot tracking on four real and synthetic benchmarks. A regression readout that bypasses the video VAE and uses additional training data brings further improvement, demonstrating the potential of this direction for dense 3D tracking. We hope our analysis will inspire future work that leverages VDM priors for dense 3D tracking and 4D reconstruction. Project page: https://nirvanalan.github.io/projects/track4d
Nov 11, 2026

Reframing Video Inverse Tone Mapping as Blind Posterior Inference via Latent Luminance Supervision

High-dynamic-range (HDR) displays are now common, but most video content, such as archival footage, broadcast material, and the output of generative video models, exists only in standard dynamic range (SDR). The need is sharpest for generated video, which has no HDR master at all and is graded alongside live-action and visual-effects material finished in HDR. Converting SDR to HDR means undoing a tone-mapping operation that was never recorded, and clipping destroys the radiance wherever it acts, so HDR recovery is a blind inverse problem. Methods built on generative video priors fill clipped regions with plausible structure, but their reconstructions are systematically biased: in every public method we test, saturated content leans too bright or too dark, and the direction of the lean moves with the tone operator behind the input. We call this failure the luminance error, measure both its magnitude and its sign, and make it the focus of this work. Reframing ITM as posterior inference under an unknown operator locates the failure in two places and yields two corrections. A fixed affine functional of the video-diffusion latent reads log-luminance without decoding; on that coordinate an asymmetric loss selects an expectile of the posterior, applied only where clipping leaves luminance unconstrained. We also synthesize training SDR through an explicit family of tone operators, so the model marginalizes over what the observation cannot show. Trained this way, our luminance error stays flatter across operators and smaller on average than any public method, and the full system achieves the best perceptual quality on all three benchmarks.
Nov 10, 2026

BirdDreams: Joint Perspective and BEV Video Generation for a Plausible, Reactive Driving World Model

Driving world models aim to simulate how a scene unfolds under different ego actions. This requires plausible motion of surrounding vehicles as well as precise ego-trajectory control. Learning vehicle motion from perspective video alone is challenging because motion is entangled with appearance and vehicles can be occluded. Prior work therefore conditions video generation on layouts that specify vehicle motion, providing control over surrounding vehicles but requiring their future trajectories as inputs. We introduce BirdDreams, an autoregressive driving world model that jointly generates perspective video and a schematic bird's-eye-view (BEV) video, making scene layout part of the predicted future. The BEV, rendered from scene annotations for training, depicts roads and vehicles as simple shapes at a fixed metric scale, providing a top-down view of their positions and motion with reduced occlusion between vehicles. Perspective video, by contrast, retains scene details that the BEV omits, such as traffic signals and object appearance. Joint attention between two video diffusion transformers, one per stream, lets each stream inform the other during generation. The predicted BEV also provides a geometric interface for ego control: we translate and rotate its latest frame along an input ego trajectory and use the resulting frames as a soft condition that shows how the scene would appear to move if only the ego vehicle moved. At inference, BirdDreams generates both videos from a single camera image without requiring an HD map or vehicle-state inputs. On PhysicalAI-AV and nuScenes, BirdDreams follows input ego trajectories more faithfully than prior driving world models and generates more plausible ego motion without ego-action conditioning. Human evaluation shows that BirdDreams outperforms a perspective-only baseline in generating realistic surrounding-vehicle behavior.
Nov 10, 2026

When to Start Denoising? Adaptive Denoising Entry for Generative Video Compression

Generative video compression offers a promising solution for compressing and reconstructing videos under extremely low bitrate constraints, among which the diffusion-based solutions still achieve the best performance by providing the powerful generative latent prior. However, their effectiveness critically depends on when the compressed latent is introduced into the diffusion process, which directly governs the trade-off between reconstruction fidelity and perceptual realism. To address this limitation, we identify denoising start-timestep selection as a key factor governing this trade-off and formulate it as a geometric optimization problem. Based on this formulation, we propose a novel adaptive diffusion entry method for generative video compression (ADE-GVC), the first timestep-aware generative video compression framework, featuring a lightweight predictor that estimates an adaptive denoising start point from the compressed latent. A temporal adapter further extracts compression-aware conditions and injects them into a pretrained video diffusion model, enabling the reconstruction of videos with both high fidelity and realism. More importantly, we observe a strong quantitative relationship between the predicted timesteps and the distortion between the original and compressed latents, providing empirical support for our optimization formulation. Extensive experiments demonstrate that ADE-GVC achieves the state-of-the-art performance on perceptual metrics and substantially improves reconstruction realism over existing generative video compression methods. Notably, our ADE-GVC achieves an average bitrate savings of 40.99% (on DISTS) and 95.24% (on PSNR) compared to the latest open-source baseline GLC-Video.
Nov 10, 2026

Dark Matters: VAE-Aware HDR Encoding for SDR-to-HDR Video Reconstruction

SDR-to-HDR reconstruction is an ill-posed problem in which clipped highlights, crushed shadows, and quantization artifacts must be recovered while preserving the original content. Latent diffusion models provide powerful generative priors for restoring missing detail, but their effectiveness is constrained by how HDR signals pass through the pretrained VAE. We find that common log-like encodings suppress low-luminance chromatic differences before generation, producing a dark-side collapse in the frozen-VAE latent. We introduce EquivFlow, a compact, invertible exposure-equivariant encoding that maps exposure changes to additive shifts while retaining a scale-independent chromatic component, without modifying the VAE. Encoding-level tests across pretrained image and video VAEs show that its exposure structure and deep-dark chroma transport persist across the tested hosts. Integrated into SDR-to-HDR video diffusion, EquivFlow delivers consistent chromatic-fidelity gains across the tested settings while retaining competitive reconstruction fidelity.
Nov 9, 2026

MPEG-Diffusion: Mixed-Resolution Video Generation

Video diffusion transformers (DiTs) generate realistic videos, but their heavy computational cost limits broader real-world applications. Mixed-resolution generation reduces this cost by generating the visually important region at high resolution (HR) and the rest at low resolution (LR), yet existing methods suffer a severe quality drop on videos and require knowing the HR region in advance. We show that these failures are not inherent to mixed resolution, and propose MPEG-Diffusion\textbf{MPEG-Diffusion} to address their three root causes: (1) dual-resolution attention to mitigate the curse of mixed resolution, where each query reads the K/V of its own resolution, (2) a mixed-resolution decoder to blend the HR and LR latents before decoding and resolve distortion, and (3) a self-selected HR region using the DiT's knowledge to find the visually important region without requiring it in advance. Processing only 30%30\% of full-resolution tokens, MPEG-Diffusion speeds up denoising by 4.8×4.8\times on Wan2.1 14B (3.8×3.8\times on 1.3B), outperforming prior methods on nearly every metric. In the user study, we demonstrate the high quality of MPEG-Diffusion, with raters preferring it over prior mixed-resolution methods in over 90%90\% of pairs and rating it close to full resolution.
Nov 7, 2026

ESTRA: Enabling Extreme Attention Sparsity for Training-Free Video Diffusion Acceleration

Video diffusion transformers generate high-quality videos, but attention over long spatiotemporal sequences makes inference computationally expensive. Sparse attention reduces this cost by computing only a subset of query-key interactions exactly, and recent training-free methods further approximate the remaining interactions. These methods, however, degrade quickly at high sparsity, which limits the achievable speedup. We present ESTRA (Extreme Sparsity Through Refined Attention), a training-free acceleration method that preserves generation quality even at extreme sparsity. ESTRA concentrates exact attention computation on a small set of important key-value tokens and approximates the remaining contributions with cluster representatives. To identify these tokens efficiently, it groups similar queries and uses their representatives to score individual keys. A variance-based correction further refines the attention weights of the cluster representatives, improving the accuracy of the approximation. Experiments on Wan2.1-14B and HunyuanVideo-13B show that ESTRA establishes a new quality-efficiency frontier for training-free sparse attention, achieving 3.26×3.26\times and 4.71×4.71\times end-to-end speedups at 5% attention density while surpassing Sol-Attn and SVG-EAR in fidelity to dense generation and on most VBench dimensions. Even at 1% density, ESTRA remains more faithful to dense generation than Sol-Attn at 12.5%, while reaching up to 5.39×5.39\times speedup.
Nov 6, 2026

AR-Edit: Training-Free Autoregressive Video Editing without Inversion

Editing a video stream in real time, without access to future frames, is essential for interactive applications. Existing training-free video editing methods, however, assume full offline access to the entire sequence, while streaming approaches either rely on additional training or are limited to specific editing techniques. We show that this limitation can be removed by analyzing the self-attention dynamics of DMD-distilled autoregressive (AR) video diffusion models. We uncover a key asymmetry: queries and keys remain stable and aligned with their source counterparts across denoising timesteps, while values diverge and encode the evolving content. Inherited from the non-distilled bidirectional teacher, this asymmetry is strongest at high noise, where few-step streaming models operate. The resulting mechanism is simple: a single forward pass suffices to capture source identity in the value features, which can be reused throughout generation. Starting from pure Gaussian noise and injecting only these source values reconstructs the input with fidelity matching or surpassing inversion-based methods, effectively eliminating inversion. Based on this insight, we introduce AR-Edit, a training-free, inversion-free method that caches source values once and selectively injects them during denoising. Across multiple AR backbones, AR-Edit enables real-time streaming editing with superior source preservation and tonal fidelity at minimal overhead.
Nov 6, 2026

Trajectory Forcing: Exploiting Diffusion Trajectories for Autoregressive Long Video Generation

Autoregressive (AR) modeling is arguably the only way to extend a fixed-length video diffusion model to longer or even streaming videos. However, existing AR diffusion methods use only the previously predicted chunk as context for the next, discarding the intermediate predictions that constitute the chunk's generation history. In contrast, conventional AR models such as LLMs condition on the entire history of preceding tokens. To this end, we propose Trajectory Forcing (TraF), an AR diffusion method that exploits the denoising trajectories of previous video chunks as context. TraF operates under two training regimes: (1) with ground-truth videos, where pseudo-trajectories are constructed from Gaussian interpolations and re-predicted by the model, and (2) with a teacher model, where real trajectories are obtained from the student's own denoising rollout. The first regime naturally provides a strong initialization for the second. On the 5-second VBench benchmark, TraF achieves a Total score of 85.15, outperforming Self Forcing (+0.84) and Causal Forcing (+1.11), with Quality and Semantic improving simultaneously. On 30-second generation---6×\times the training horizon---TraF achieves the highest overall quality among all compared methods.
Nov 6, 2026

UniRGBA: Unifying RGB and Alpha in Pretrained Video Diffusion Models

Recent advances in video diffusion models have achieved remarkable photorealistic RGB synthesis, yet generating videos with a native alpha channel remains an open challenge. Existing approaches often rely on matting-style data and non-native RGBA designs, causing them to specialize in real-object scenes and to degrade severely on visual-effect content whose semi-transparency is spatially non-local and non-elliptical. We present UniRGBA, a unified framework whose key insight is that RGBA generation does not require a dedicated alpha encoder or complex latent-space restructuring---a minimal, targeted extension of the existing RGB pathway is sufficient. UniRGBA-VAE extends a causal video VAE to 4-channel RGBA via zero-initialized channel expansion, symmetric selective layer unfreezing (5% trainable), and a six-term progressive reconstruction loss anchored by a KL reference loss that prevents latent drift, followed by a two-stage schedule that introduces a lightweight GAN objective. We then apply standard DoRA to the pretrained DiT, reusing the unified 4-channel latent without any alpha-specific architectural changes. On a visual-effect RGBA benchmark, UniRGBA substantially outperforms Wan-Alpha on RGBA reconstruction while preserving RGB fidelity. For generation, UniRGBA outperforms Wan-Alpha on RGBA text-to-video tasks and matches or surpasses the original pretrained Wan2.1 backbone on RGB-only metrics, demonstrating that native alpha support can be integrated without compromising the underlying RGB prior or the model's architectural versatility.
Nov 6, 2026

MotionCore: A Latent Neural Simulator for Video Generation with Rigid-Body Dynamics

Video generation models can produce highly realistic videos while generating physically inconsistent object motion. This limitation suggests that visual modeling alone may not provide a transferable representation of physical dynamics. We introduce MotionCore, a framework for learning physical dynamics via a learnable partial differential equation solver and focuses on rigid-body motion. We introduce a latent neural simulator learning to predict the evolution of object-level physical states from observed motion. Its predictions are then converted into spatially aligned latent controls for a controllable video diffusion model. The generator serves as a visual interface: it provides appearance and scene detail, while the learned dynamics model specifies the physical trajectory. To preserve this trajectory through the interface, MotionCore combines a diffusion-matching objective with a low-capacity decodability constraint on object identity and position. We evaluate MotionCore on public physical-consistency benchmarks and on paired data with known trajectories. Results indicate that MotionCore improves physical consistency and trajectory obedience in the evaluated rigid-body settings, while maintaining comparable visual quality. A controlled ablation protocol compares learned dynamics with an analytic prior and studies the choice of latent evolution processor.
Nov 5, 2026

InstaGen: Group-Shared Importance Sampling for Fast Video Generation

Diffusion Transformers (DiTs) are the dominant architecture for video generation, but their quadratic attention complexity results in substantial computational cost and latency. Permuted block-sparse attention (PBSA) offers a promising approach to reducing this cost by selectively computing blocks of a semantically permuted attention matrix. However, existing approaches typically select only the top few key blocks for each query block, following a top-pp paradigm. We find that top-pp selection can be highly sample inefficient in high-entropy attention regimes, which are common in video generation. In these regimes, we show that sampling can provide substantially better approximations under the same computational budget. More generally, we cast block selection as importance sampling, where the top-scoring keys form an exact set that is computed in full and the remaining keys are represented by a sample drawn in proportion to their estimated attention and reweighted by inverse sampling probability. Building on this formulation, we present InstaGen, which can be parameterized by any PBSA method: it replaces the method's block selection with an exact set plus a weighted sample. Each query cluster draws one sample shared by all its queries, so the sampled keys form blocks amenable to efficient GPU execution. In this paper, we take the SVG family of PBSA methods and show that, on Wan2.1-T2V-1.3B and Wan2.1-T2V-14B, InstaGen parameterized by SVG2 improves PSNR over SVG2 by 3.4 dB and 2 dB, respectively, at matched attention density, and by 1.6 dB and 2.2 dB at matched wall-clock time. On Wan2.1-T2V-14B, InstaGen also beats SVG-EAR by 1.1 dB PSNR at nearly the same density while running faster (1.65x against 1.48x over dense).
Nov 5, 2026

Training-Free Predictive Spatio-Temporal Token Pruning for Few-Step Autoregressive Video Diffusion

Few-step autoregressive (AR) video diffusion models generate videos chunk by chunk and have made efficient video generation practical. However, further accelerating these models toward real-time generation remains challenging, as few-step distillation leaves little cross-step redundancy for existing training-free caching methods to exploit. We revisit this regime and identify two token-level redundancy patterns: cross-frame redundancy from static content shared across adjacent frames, and cross-step redundancy from tokens that converge early during denoising. Crucially, we find that early clean estimates along the model's own denoising trajectory provide an effective online signal for locating both patterns, requiring no retraining, auxiliary predictor, or reference video. Based on these findings, we propose PreSTP (Predictive Spatio-Temporal Token Pruning), a training-free framework that constructs a unified token-wise redundancy mask from cross-frame similarity and cross-step convergence. Tokens marked as redundant skip subsequent computation under a consistent last-computed-value restoration rule, and a content-adaptive mask reuse strategy further reduces dense probing cost. On 5-second generation, PreSTP achieves 1.52x-3.20x speedups and a better efficiency-fidelity trade-off than coarse-grained caching baselines such as TeaCache and FlowCache. On 30-second generation, it reaches up to 1.96x speedup while improving VBench-Long and suppressing long-horizon drift. Notably, it continues to reduce inference cost even in the extreme one-step regime, where cross-step redundancy vanishes entirely and coarse step-wise caching is no longer applicable.
Nov 5, 2026

Householder Motion Embedding: Noise Modulation for Motion-Controllable Video Generation

Diffusion models for video generation have achieved remarkable success, yet precise motion control under explicit user guidance remains challenging. Recent noise-warping methods introduce temporal correlations by warping the noise according to the user-provided motion signal. However, standard spatial warping generally disrupts the noise's i.i.d. Gaussian structure, requiring distribution correction, such as Jacobian-based correction, but at a substantial computational overhead. We propose Householder Motion Embedding (HME), which encodes the motion signal into the noise through a sequence of Householder reflections that together form an orthogonal transformation applied to each location's channel vector. HME exactly preserves the spatial Gaussian structure within each frame while introducing cross-frame correlations that promote temporally consistent video. HME is agnostic to the network architecture and integrates directly into standard video diffusion fine-tuning pipelines. Experiments on different scenarios show that HME matches or outperforms the state of the art, while achieving a 2.7x speedup over noise warping. These results demonstrate that channel-space Householder transformations offer a principled and efficient alternative to noise warping for motion-controllable video generation.
Nov 4, 2026

VA-TDM: A Unified View for Improving Few-Step Diffusion Models by Velocity Adversarial Distillation

Few-step distillation offers efficient generation with diffusion and flow models, but preserving quality with only a few sampling steps remains challenging. Distribution-matching methods via score distillation provide a strong foundation, yet they rely on accurately tracking the student's changing distribution. At intermediate generation steps in few-step sampling, student-score estimation can be difficult, and commonly used estimators may describe a different distribution from the one used to update the student. Flow matching allows a more flexible approach to supervision through velocity fields constructed from the student's sampling process. Building on this perspective, we introduce Velocity Adversarial Distillation (VAD), which uses a learned critic to compare student and teacher fields and turn their disagreement into feedback for learning. VAD unifies score distillation, Fisher-based matching, and moment matching, also explaining existing heuristic MMD updates and delivering new training objectives. We further develop VA-TDM, which follows trajectory distribution matching: providing supervision for partial denoising along the student trajectory. Beyond this, VA-TDM combines transition-aware online velocity learning and dual-form VAD supervision across intermediate student transitions. Extensive results show that VA-TDM is powerful for learning few-step generators, achieving state-of-the-art four-step performance in both text-to-image and text-to-video generation across various backbones and scaling effectively to 33B MiniMax-H3. In particular, our VA-TDM-H3 notably outperforms the few-step Fast-H3 and is closer to the original many-step H3 according to human preference.
Nov 4, 2026

DecoupAttn: Decoupled Sparse Attention for Training-Free Acceleration of Video Diffusion

Sparse attention accelerates few-step video diffusion transformers by computing a few key blocks exactly and compensating for the rest. However, existing sparse attention still faces key challenges, including identification overhead from finding critical blocks at every denoising step, suboptimal budget allocation that sets sparsity budgets uniformly or by local metrics, and cross-step error accumulation as reused masks or clusters go stale. Observing that the errors left on skipped blocks cancel each other once every key is compensated, we propose DecoupAttn, a new paradigm of decoupled sparse attention: semantic clusters compensate every key, and exact attention corrects their error on a few blocks fixed once offline by position. To support this paradigm, we introduce Distortion Budget, a rate-distortion objective for budget allocation that places exact computation across timesteps, layers, and heads where it removes the most distortion that compensation leaves in the final latent. Built on this budget, we further propose Reassign Cache, in which each step reassigns the drifting keys to cached centroids and refreshes compensation while the budgeted blocks stay fixed. Moreover, DecoupAttn is training-free: a few calibration clips compile one plan offline, and inference runs standard block-sparse kernels with one fused compensation kernel. Our code will be released after submission.
Nov 4, 2026

DualCast: Dual-Image Conditioned Diffusion Transformers for Digital Human E-Commerce Video Generation

Digital human e-commerce video generation requires preserving product appearance, human identity, and overall video style during dynamic demonstrations. Although recent models support multiple reference images, their general conditioning schemes do not explicitly distinguish the asymmetric roles of "what is presented" and "who presents it." Consequently, generated videos may favor a single reference, resulting in product distortion, identity drift, or single-reference collapse. To address this issue, we propose DualCast, a role-addressed dual-image conditioning framework for video generation. By explicitly modeling the distinct roles of product and human conditions, DualCast reduces role ambiguity and feature competition between references. First, a dual-spatiotemporal slot injection mechanism VAE-encodes the product and human images separately and places their latent features at the front and rear boundaries of the DiT sequence, enabling sequence-level condition separation and directed injection. Second, a mask-guided region-weighted loss strengthens supervision over product, face, and product-face overlap regions, improving generation fidelity in critical areas. Third, image condition dropout enables a single model to support dual-image, product-only, and human-only generation. In addition, we construct a curated dataset of 10K e-commerce videos for model training. Experimental results demonstrate the strong performance of DualCast.
Nov 3, 2026

DreamX: Video Diffusion Model is a Universal Latent Action Learner

Learning a universal action representation from video requires modeling dynamics that are consistent across different executions of the same action yet invariant to changes in embodiment and appearance. Existing latent action models infer actions from visual transitions, but their representations often entangle the underlying action with irrelevant visual variations, limiting transfer across embodiments. We introduce DreamX, which repurposes a pretrained video diffusion model as a universal latent action learner. Our approach extracts compact latent actions from the video model's rich spatiotemporal features, compressing each latent video frame by 300×\boldsymbol{300\times} (1,200→4\boldsymbol{1{,}200 \rightarrow 4} tokens). In addition, by leveraging the video model's generative prior, these latent actions capture action-relevant dynamics disentangled from visual appearance. The resulting latent space serves as a shared action interface across embodiments (human and robot), environments, and action parameterizations. On LARY-Bench, DreamX achieves state-of-the-art performance among latent action models. Moreover, actions encoded in one setting remain effective in another: the latent space supports any-to-any video transfer and zero-shot action transfer across diverse embodiments and environments. Furthermore, this shared interface also enables policy learning in a compact latent action space, reducing the computational cost of video-diffusion-based policies.
Nov 3, 2026

FloMo: 3D Scene Flow as a World Action Model Intermediate for Learning from Human Video

Pretrained video generation models distill internet-scale motion priors into robot policies, but full video is a redundant signal for action prediction. We argue that dense 3D motion is the medium through which video pretraining best transfers to robot actions, and that predicting motion rather than video affords better sample-efficiency for action prediction. FloMo is a 5B video-generation model fine-tuned to predict 3D scene-flow and action tokens. Rendering scene flow as RGB and encoding it into the same latent space as video enables a pretrained video generation backbone to efficiently adapt to predict 3D scene flow. Co-trained on action-free human video and action-labeled robot teleoperation data, our model benefits from human-video pretraining and generalizes to unseen objects, motion primitives, and tasks. Replacing motion prediction with a video prediction objective degrades out-of-distribution generalization, while adding a video prediction target alongside motion lowers policy success rather than improving it. When motion, video, and action are predicted jointly, action tokens attend far more to motion than to video, suggesting that predicting video wastes model capacity. Together, these results indicate that motion effectively mediates the transfer from video pretraining to action. More information can be found at flomo-wam.github.io.
Nov 3, 2026

ReXMo: End-to-End Video Character Replacement with Expression and Motion Preservation

Video character replacement aims to replace actors while preserving facial expressions, motions, and scene dynamics. Existing methods typically rely on masked generation with pose guidance, limiting fine-grained expression and motion preservation in complex scenes. We propose ReXMo, an end-to-end framework that supports single-, multi-, and selective character replacement scenarios. ReXMo introduces paired triplet supervision, where synthesized driving videos and original videos share identical performances but different identities, enabling scalable training through synthetic data generation. An In-Context Diffusion Transformer learns identity transformation by jointly conditioning on driving videos, reference identities, and text prompts, while an Expression Enhancement Adapter improves facial expression preservation in multi-person scenes. Extensive experiments demonstrate that ReXMo outperforms existing methods in identity fidelity, motion consistency, expression preservation, and visual quality. Our dataset and code will be publicly released.
Nov 2, 2026

HIVE-Diff: High Frame Rate HDR Video Reconstruction from a Single LDR Frame and Events

High frame rate HDR video is important in applications such as autonomous driving and scientific imaging, where rapid motion and extreme illumination often occur simultaneously. Conventional frame cameras face a trade-off between temporal resolution and dynamic range, while event cameras provide complementary high-speed brightness-change measurements over a wide illumination range. These properties have motivated event-based HDR reconstruction, with existing methods broadly following event-only and hybrid event-frame paradigms. Event-only HDR reconstruction, however, suffers from brightness and color drift due to the lack of absolute intensity references. Hybrid event-frame methods alleviate this ambiguity using low-dynamic-range (LDR) frames as anchors, but most perform one-to-one HDR reconstruction at available LDR timestamps. In this work, we investigate single-reference HDR temporal expansion, reconstructing N HDR frames from one LDR reference and events instead of N aligned LDR observations. We develop HIVE-Diff, an event-conditioned video diffusion framework that decomposes the task into event-guided temporal generation and HDR content completion. Because paired real HDR-event data are scarce, we train on synthetic data and adapt the model to real event domains using LDR-event sequences as proxy supervision with teacher-student distillation. Experiments on synthetic and multiple real-world datasets demonstrate strong reconstruction quality and robust real-data adaptation.
Nov 2, 2026

VidDub: Replacement-Aware Supervision for Two-Step Video Dubbing

Video dubbing aims to synthesize lip movements that match target speech while preserving the speaker identity, head pose, and background of a source video. Existing editors are predominantly trained via self-reconstruction under original paired audio, even though the source mouth becomes a conflicting target when driving speech is replaced. We introduce replacement-aware cross-audio supervision: paired-audio samples use full-frame reconstruction, whereas unpaired replacement-audio samples mask out source-mouth reconstruction losses and supervise the mouth solely through audio--visual synchronization while preserving unmasked facial regions and background. Because offline dubbing provides the complete source video and target speech beforehand, we present \method, a full-context diffusion editor that uses matched trajectory regression to transfer CFG-guided teacher trajectories to a two-step student and distribution matching to recover visual quality after compression. We will release the curated training dataset, VidDub-Bench, fixed paired- and cross-audio evaluation manifests, preprocessing outputs, and unified evaluation code. Across three training seeds, masking out conflicting source-mouth reconstruction improves cross-audio Sync-C by 0.833. At two NFE, the final student provides the strongest cross-audio synchronization and visual-speech recognition among the evaluated methods, with competitive visual fidelity and source preservation.
Nov 2, 2026

Disen-Forcing: Disentangling Semantic Anchoring from Motion for Autoregressive Video Diffusion

Autoregressive video diffusion models (AR-VDMs) enable real-time, long-horizon video generation but degrade rapidly once the inference horizon outruns the training horizon, due to compounding exposure bias. To stabilize long rollouts, recent methods heuristically adopt frame sink, keeping initial frames as a persistent model context, inspired by attention sinks in LLMs. We revisit this design and find that the mechanism by which initial frames stabilize long-range consistency in AR-VDMs is not an LLM-style attention sink; instead, long-range consistency is anchored through a sparse subset of attention heads, while the majority of heads predominantly capture local temporal dynamics. This observation reveals an over-conditioning pathology in existing frame-sink strategies: by universally exposing every head to the pinned initial-frame context, the model becomes prone to collapsing motion toward the low-dynamic modes of the teacher during model distillation. We propose Disen-Forcing, a distillation-time plug-in for AR-VDMs that disentangles and reinforces this head specialization through disentangled context routing and motion-disentangling regularization. On long-horizon video generation benchmarks, Disen-Forcing simultaneously improves long-range consistency and motion dynamics of base AR-VDMs, a trade-off that sink-based methods fail to balance.
Nov 1, 2026

dKV Forcing: Separating Generation and Memory Representations through a Differentiable KV Cache

A causal video generator produces two key--value (KV) states for every segment: the KV it uses while denoising that segment, and the KV it writes into the cache for later segments to read. In a pretrained generator the two already differ substantially, later segments attend to them differently, and replacing the written KV with the generation KV degrades the continuation. We propose dKV Forcing (dKV denotes differentiable KV) to learn a dedicated memory representation from its future use: a writer-specific low-rank residual on the shared backbone separates memory writing from generation and is trained through the future reads of what it writes. Learning a write from its future use needs gradients through the cache, which the mutable cache in common use cannot provide. dKV Forcing replaces the training-time cache by persistent differentiable pages with local rematerialization and, in one temporal traversal, computes exactly the gradient that a two-pass reconstruction of the same objective computes. One traversal computes the full memory gradient at 7%7\% generator-update overhead over detached-write training, with 28%28\% less resident memory and 20.8%20.8\% shorter training cycles than the two-pass reconstruction. Over Self Gradient Forcing (SGF), the same objective trained with the same traversal, writer adaptation raises VBench-Long aesthetic quality by 6.66.6 points at 60,s and 8.58.5 points at 240,s with the highest subject and background consistency, and rollout dropout of the residual keeps most of this gain while restoring the motion of training without write gradients.
Nov 1, 2026

Recoverability-Aware Evaluation of Diffusion Video Inpainting and Object Removal

Video inpainting for object removal is scored over the whole hole, although the background behind the object is visible in other frames of the shot for some hole pixels and in no frame for others. A whole-hole score therefore mixes copying with generation and cannot show which of the two a remover does well. To separate the two, we propose recoverability-stratified evaluation, in which a point tracker run on the clean video labels each hole pixel by whether and how far away it is observed, so that fidelity and temporal stability can be reported per stratum against a copy reference. To reuse the observed content, we also propose evidence-guided sampling (EGS), which blends tracked colors into the clean-latent prediction of a video diffusion remover without training. On DAVIS, the perceptual advantage of three diffusion removers over flow propagation shrinks where the content is visible a few frames away (LPIPS difference-in-differences +0.046 to +0.056 against E2FGVI-HQ), and real paired captures show the same pattern for two of them. Their DAVIS hole warping error is two to four times that of ProPainter, and copying along clean-video tracks lowers their recoverable LPIPS by a quarter to two fifths. EGS closes 43% and 56% of the gap to this copy reference on held-out DAVIS when its colors come from clean-video tracks. With colors tracked on the masked input, a pre-registered confirmation finds no gain on MiniMax-Remover and a loss on ROSE. Diffusion removers thus leave observed content unused on DAVIS, and training-free reuse does not yet hold with tracked evidence.
Oct 31, 2026

Wave Forcing: Towards Speed-of-Light Streaming Video Generation

Real-time long-video generation requires continuous output with consistent appearance and natural motion. Existing chunk-wise autoregressive methods face challenges in both generation efficiency and long-video quality: serialized denoising and history-cache refresh require two forward passes even with one-step denoising, and longer rollouts can degrade quality and cause camera drift. We propose Wave Forcing (WF), which co-designs generation, training, and execution for continuous video output with multi-step denoising. WF uses chunk-wise causal attention to access earlier, cleaner chunks before their denoising is complete, and distributes staggered denoising and cache-refresh stages across full-model replicas to achieve 1F1C: one output chunk per concurrent forward round after pipeline fill. To adapt the model to this continuous generation process, we develop a three-stage training procedure that successively introduces mixed-noise denoising, motion supervision, and causal adaptation using self-generated history. To execute this process efficiently, Wave Runtime (WaveRT) uses Wave Parallelism (WP) for early K/V delivery and communication-computation overlap, together with optimized kernels and pipelined VAE decoding. Experiments on Wan2.1-T2V-1.3B show that 5-step WF maintains overall quality comparable to Rolling Forcing while achieving higher Dynamic Degree on both short and long videos. Our system efficiency experiments show that 4-step WF achieves 147.6 DiT FPS and 134.8 end-to-end FPS on eight H200 GPUs, yielding 7.07x the single-GPU end-to-end throughput. In DiT-only tests with matched SageAttention+W8A8 backends, denoising steps, and GPU counts, WP delivers up to 2.53x the throughput of the faster of the Ulysses and Ring context-parallel baselines across six model scales (1.3B-14B).