Daily Papers

Our aim is to build autonomous agents that can solve tasks in environments like Minecraft. To do so, we used an imitation learning-based approach. We formulate our control problem as a search problem over a dataset of experts' demonstrations, where the agent copies actions from a similar demonstration trajectory of image-action pairs. We perform a proximity search over the BASALT MineRL-dataset in the latent representation of a Video PreTraining model. The agent copies the actions from the expert trajectory as long as the distance between the state representations of the agent and the selected expert trajectory from the dataset do not diverge. Then the proximity search is repeated. Our approach can effectively recover meaningful demonstration trajectories and show human-like behavior of an agent in the Minecraft environment.

General-purpose robots require decision-making models that generalize across diverse tasks and environments. Recent works build robot foundation models by extending multimodal large language models (MLLMs) with action outputs, creating vision-language-action (VLA) systems. These efforts are motivated by the intuition that MLLMs' large-scale language and image pretraining can be effectively transferred to the action output modality. In this work, we explore an alternative paradigm of using large-scale video pretraining as a primary modality for building robot foundation models. Unlike static images and language, videos capture spatio-temporal sequences of states and actions in the physical world that are naturally aligned with robotic behavior. We curate an internet-scale video dataset of human activities and task demonstrations, and train, for the first time at a foundation-model scale, an open video model for generative robotics planning. The model produces zero-shot video plans for novel scenes and tasks, which we post-process to extract executable robot actions. We evaluate task-level generalization through third-party selected tasks in the wild and real-robot experiments, demonstrating successful physical execution. Together, these results show robust instruction following, strong generalization, and real-world feasibility. We release both the model and dataset to support open, reproducible video-based robot learning. Our website is available at https://www.boyuan.space/large-video-planner/.

Behavioural cloning uses a dataset of demonstrations to learn a behavioural policy. To overcome various learning and policy adaptation problems, we propose to use latent space to index a demonstration dataset, instantly access similar relevant experiences, and copy behavior from these situations. Actions from a selected similar situation can be performed by the agent until representations of the agent's current situation and the selected experience diverge in the latent space. Thus, we formulate our control problem as a search problem over a dataset of experts' demonstrations. We test our approach on BASALT MineRL-dataset in the latent representation of a Video PreTraining model. We compare our model to state-of-the-art Minecraft agents. Our approach can effectively recover meaningful demonstrations and show human-like behavior of an agent in the Minecraft environment in a wide variety of scenarios. Experimental results reveal that performance of our search-based approach is comparable to trained models, while allowing zero-shot task adaptation by changing the demonstration examples.

Constructing AI models that respond to text instructions is challenging, especially for sequential decision-making tasks. This work introduces an instruction-tuned Video Pretraining (VPT) model for Minecraft called STEVE-1, demonstrating that the unCLIP approach, utilized in DALL-E 2, is also effective for creating instruction-following sequential decision-making agents. STEVE-1 is trained in two steps: adapting the pretrained VPT model to follow commands in MineCLIP's latent space, then training a prior to predict latent codes from text. This allows us to finetune VPT through self-supervised behavioral cloning and hindsight relabeling, bypassing the need for costly human text annotations. By leveraging pretrained models like VPT and MineCLIP and employing best practices from text-conditioned image generation, STEVE-1 costs just $60 to train and can follow a wide range of short-horizon open-ended text and visual instructions in Minecraft. STEVE-1 sets a new bar for open-ended instruction following in Minecraft with low-level controls (mouse and keyboard) and raw pixel inputs, far outperforming previous baselines. We provide experimental evidence highlighting key factors for downstream performance, including pretraining, classifier-free guidance, and data scaling. All resources, including our model weights, training scripts, and evaluation tools are made available for further research.

Pretraining on noisy, internet-scale datasets has been heavily studied as a technique for training models with broad, general capabilities for text, images, and other modalities. However, for many sequential decision domains such as robotics, video games, and computer use, publicly available data does not contain the labels required to train behavioral priors in the same way. We extend the internet-scale pretraining paradigm to sequential decision domains through semi-supervised imitation learning wherein agents learn to act by watching online unlabeled videos. Specifically, we show that with a small amount of labeled data we can train an inverse dynamics model accurate enough to label a huge unlabeled source of online data -- here, online videos of people playing Minecraft -- from which we can then train a general behavioral prior. Despite using the native human interface (mouse and keyboard at 20Hz), we show that this behavioral prior has nontrivial zero-shot capabilities and that it can be fine-tuned, with both imitation learning and reinforcement learning, to hard-exploration tasks that are impossible to learn from scratch via reinforcement learning. For many tasks our models exhibit human-level performance, and we are the first to report computer agents that can craft diamond tools, which can take proficient humans upwards of 20 minutes (24,000 environment actions) of gameplay to accomplish.

Large language models (LLMs) have demonstrated that large-scale pretraining enables systems to adapt rapidly to new problems with little supervision in the language domain. This success, however, has not translated as effectively to the visual domain, where models, including LLMs, continue to struggle with compositional understanding, sample efficiency, and general-purpose problem-solving. We investigate Video Diffusion Models (VDMs) as a promising direction for bridging this gap. Pretraining on spatiotemporal data endows these models with strong inductive biases for structure and dynamics, which we hypothesize can support broad task adaptability. To test this, we design a controlled evaluation in which both a pretrained LLM and a pretrained VDM are equipped with lightweight adapters and presented with tasks in their natural modalities. Across benchmarks including ARC-AGI, ConceptARC, visual games, route planning, and cellular automata, VDMs demonstrate higher data efficiency than their language counterparts. Taken together, our results indicate that video pretraining offers inductive biases that support progress toward visual foundation models.

Non-human primates are our closest living relatives, and analyzing their behavior is central to research in cognition, evolution, and conservation. Computer vision could greatly aid this research, but existing methods often rely on human-centric pretrained models and focus on single datasets, which limits generalization. We address this limitation by shifting from a model-centric to a data-centric approach and introduce PriVi, a large-scale primate-centric video pretraining dataset. PriVi contains 424 hours of curated video, combining 174 hours from behavioral research across 11 settings with 250 hours of diverse web-sourced footage, assembled through a scalable data curation pipeline. We pretrain V-JEPA on PriVi to learn primate-specific representations and evaluate it using a lightweight frozen classifier. Across four benchmark datasets, ChimpACT, BaboonLand, PanAf500, and ChimpBehave, our approach consistently outperforms prior work, including fully finetuned baselines, and scales favorably with fewer labels. These results demonstrate that primate-centric pretraining substantially improves data efficiency and generalization, making it a promising approach for low-label applications. Code, models, and the majority of the dataset will be made available.

The scarcity of manipulation data has motivated the use of pretrained large models from other modalities in robotics. In this work, we build upon autoregressive video generation models to propose a Physical Autoregressive Model (PAR), where physical tokens combine frames and actions to represent the joint evolution of the robot and its environment. PAR leverages the world knowledge embedded in video pretraining to understand physical dynamics without requiring action pretraining, enabling accurate video prediction and consistent action trajectories. It also adopts a DiT-based de-tokenizer to model frames and actions as continuous tokens, mitigating quantization errors and facilitating mutual enhancement. Furthermore, we incorporate a causal mask with inverse kinematics, parallel training, and the KV-cache mechanism to further improve performance and efficiency. Experiments on the ManiSkill benchmark show that PAR achieves a 100\% success rate on the PushCube task, matches the performance of action-pretrained baselines on other tasks, and accurately predicts future videos with tightly aligned action trajectories. These findings underscore a promising direction for robotic manipulation by transferring world knowledge from autoregressive video pretraining. The project page is here: https://hcplab-sysu.github.io/PhysicalAutoregressiveModel/

Video-based pretraining offers immense potential for learning strong visual representations on an unprecedented scale. Recently, masked video modeling methods have shown promising scalability, yet fall short in capturing higher-level semantics due to reconstructing predefined low-level targets such as pixels. To tackle this, we present Sinkhorn-guided Masked Video Modelling (SIGMA), a novel video pretraining method that jointly learns the video model in addition to a target feature space using a projection network. However, this simple modification means that the regular L2 reconstruction loss will lead to trivial solutions as both networks are jointly optimized. As a solution, we distribute features of space-time tubes evenly across a limited number of learnable clusters. By posing this as an optimal transport problem, we enforce high entropy in the generated features across the batch, infusing semantic and temporal meaning into the feature space. The resulting cluster assignments are used as targets for a symmetric prediction task where the video model predicts cluster assignment of the projection network and vice versa. Experimental results on ten datasets across three benchmarks validate the effectiveness of SIGMA in learning more performant, temporally-aware, and robust video representations improving upon state-of-the-art methods. Our project website with code is available at: https://quva-lab.github.io/SIGMA.

Egocentric human video offers a scalable alternative to robot data for pretraining, yet models pretrained on such video consistently underperform those pretrained on robot data. We attribute this gap to a missing signal, the active perception behavior in egocentric videos, where humans continuously reposition their viewpoint during manipulation, inducing camera motion that standard pipelines treat as noise. To address this, we present ActiveMimic, a pretraining framework that recovers synchronized camera and wrist trajectories from a single body-worn RGB camera, models camera motion as a viewpoint action, and jointly learns active perception and manipulation from in-the-wild egocentric human video before adapting to a target robot. Empirically, real-world experiments across tasks with diverse active perception demands show that ActiveMimic consistently surpasses baselines pretrained on human video and matches state-of-the-art models pretrained on robot data. Further analysis provides evidence that active perception capability originates from egocentric human video pretraining rather than robot-specific fine-tuning, confirming active perception as the key to unlocking egocentric human video for robot pretraining.

The scarcity of large-scale robotic data has motivated the repurposing of foundation models from other modalities for policy learning. In this work, we introduce PhysGen (Learning Physics from Pretrained Video Generation Models), a scalable continuous and sequential world interaction framework that leverages autoregressive video generation to solve robotic manipulation tasks. By treating the pretrained video model as a proxy for a physics simulator, PhysGen models the dynamic interplay between the external environment and robot actions. We introduce a multimodal continuous representation that unifies video and action into shared physical tokens, bridging the gap between discrete video generation and continuous robotic control. This approach enables the seamless transfer of implicit physical knowledge-such as object permanence and dynamics-from video pretraining to downstream manipulation.To ensure efficient convergence, we incorporate causal masking, inverse kinematics, Lookahead Multi-Token Prediction (L-MTP), and key-value (KV) caching. Experimental results on the Libero and ManiSkill benchmarks demonstrate that PhysGen consistently outperforms robust baselines, surpassing OpenVLA and WorldVLA by margins of 13.8% and 8.8%, respectively. Notably, in real-world scenarios, PhysGen matches the performance of large-scale action-pretrained models like π_0 without requiring prior action-specific pretraining, demonstrating superior capability in physically complex tasks such as grasping transparent objects. These findings validate the potential of extracting physical intuition from pretrained video generators to facilitate generalizable robotic manipulation.

We present Stable Video Diffusion - a latent video diffusion model for high-resolution, state-of-the-art text-to-video and image-to-video generation. Recently, latent diffusion models trained for 2D image synthesis have been turned into generative video models by inserting temporal layers and finetuning them on small, high-quality video datasets. However, training methods in the literature vary widely, and the field has yet to agree on a unified strategy for curating video data. In this paper, we identify and evaluate three different stages for successful training of video LDMs: text-to-image pretraining, video pretraining, and high-quality video finetuning. Furthermore, we demonstrate the necessity of a well-curated pretraining dataset for generating high-quality videos and present a systematic curation process to train a strong base model, including captioning and filtering strategies. We then explore the impact of finetuning our base model on high-quality data and train a text-to-video model that is competitive with closed-source video generation. We also show that our base model provides a powerful motion representation for downstream tasks such as image-to-video generation and adaptability to camera motion-specific LoRA modules. Finally, we demonstrate that our model provides a strong multi-view 3D-prior and can serve as a base to finetune a multi-view diffusion model that jointly generates multiple views of objects in a feedforward fashion, outperforming image-based methods at a fraction of their compute budget. We release code and model weights at https://github.com/Stability-AI/generative-models .

Robotic manipulation requires understanding both the 3D spatial structure of the environment and its temporal evolution, yet most existing policies overlook one or both. They typically rely on 2D visual observations and backbones pretrained on static image--text pairs, resulting in high data requirements and limited understanding of environment dynamics. To address this, we introduce MV-VDP, a multi-view video diffusion policy that jointly models the 3D spatio-temporal state of the environment. The core idea is to simultaneously predict multi-view heatmap videos and RGB videos, which 1) align the representation format of video pretraining with action finetuning, and 2) specify not only what actions the robot should take, but also how the environment is expected to evolve in response to those actions. Extensive experiments show that MV-VDP enables data-efficient, robust, generalizable, and interpretable manipulation. With only ten demonstration trajectories and without additional pretraining, MV-VDP successfully performs complex real-world tasks, demonstrates strong robustness across a range of model hyperparameters, generalizes to out-of-distribution settings, and predicts realistic future videos. Experiments on Meta-World and real-world robotic platforms demonstrate that MV-VDP consistently outperforms video-prediction--based, 3D-based, and vision--language--action models, establishing a new state of the art in data-efficient multi-task manipulation.

Generating visual instructions in a given context is essential for developing interactive world simulators. While prior works address this problem through either text-guided image manipulation or video prediction, these tasks are typically treated in isolation. This separation reveals a fundamental issue: image manipulation methods overlook how actions unfold over time, while video prediction models often ignore the intended outcomes. To this end, we propose ShowMe, a unified framework that enables both tasks by selectively activating the spatial and temporal components of video diffusion models. In addition, we introduce structure and motion consistency rewards to improve structural fidelity and temporal coherence. Notably, this unification brings dual benefits: the spatial knowledge gained through video pretraining enhances contextual consistency and realism in non-rigid image edits, while the instruction-guided manipulation stage equips the model with stronger goal-oriented reasoning for video prediction. Experiments on diverse benchmarks demonstrate that our method outperforms expert models in both instructional image and video generation, highlighting the strength of video diffusion models as a unified action-object state transformer.

End-to-end autonomous driving increasingly leverages self-supervised video pretraining to learn transferable planning representations. However, pretraining video world models for scene understanding has so far brought only limited improvements. This limitation is compounded by the inherent ambiguity of driving: each scene typically provides only a single human trajectory, making it difficult to learn multimodal behaviors. In this work, we propose Drive-JEPA, a framework that integrates Video Joint-Embedding Predictive Architecture (V-JEPA) with multimodal trajectory distillation for end-to-end driving. First, we adapt V-JEPA for end-to-end driving, pretraining a ViT encoder on large-scale driving videos to produce predictive representations aligned with trajectory planning. Second, we introduce a proposal-centric planner that distills diverse simulator-generated trajectories alongside human trajectories, with a momentum-aware selection mechanism to promote stable and safe behavior. When evaluated on NAVSIM, the V-JEPA representation combined with a simple transformer-based decoder outperforms prior methods by 3 PDMS in the perception-free setting. The complete Drive-JEPA framework achieves 93.3 PDMS on v1 and 87.8 EPDMS on v2, setting a new state-of-the-art.

The parameter-efficient adaptation of the image-text pretraining model CLIP for video-text retrieval is a prominent area of research. While CLIP is focused on image-level vision-language matching, video-text retrieval demands comprehensive understanding at the video level. Three key discrepancies emerge in the transfer from image-level to video-level: vision, language, and alignment. However, existing methods mainly focus on vision while neglecting language and alignment. In this paper, we propose Discrepancy Reduction in Vision, Language, and Alignment (DiscoVLA), which simultaneously mitigates all three discrepancies. Specifically, we introduce Image-Video Features Fusion to integrate image-level and video-level features, effectively tackling both vision and language discrepancies. Additionally, we generate pseudo image captions to learn fine-grained image-level alignment. To mitigate alignment discrepancies, we propose Image-to-Video Alignment Distillation, which leverages image-level alignment knowledge to enhance video-level alignment. Extensive experiments demonstrate the superiority of our DiscoVLA. In particular, on MSRVTT with CLIP (ViT-B/16), DiscoVLA outperforms previous methods by 1.5% in R@1, reaching a final score of 50.5% R@1. The code is available at https://github.com/LunarShen/DsicoVLA.

Temporally locating objects with arbitrary class texts is the primary pursuit of open-vocabulary Video Instance Segmentation (VIS). Because of the insufficient vocabulary of video data, previous methods leverage image-text pretraining model for recognizing object instances by separately aligning each frame and class texts, ignoring the correlation between frames. As a result, the separation breaks the instance movement context of videos, causing inferior alignment between video and text. To tackle this issue, we propose to link frame-level instance representations as a Brownian Bridge to model instance dynamics and align bridge-level instance representation to class texts for more precisely open-vocabulary VIS (BriVIS). Specifically, we build our system upon a frozen video segmentor to generate frame-level instance queries, and design Temporal Instance Resampler (TIR) to generate queries with temporal context from frame queries. To mold instance queries to follow Brownian bridge and accomplish alignment with class texts, we design Bridge-Text Alignment (BTA) to learn discriminative bridge-level representations of instances via contrastive objectives. Setting MinVIS as the basic video segmentor, BriVIS surpasses the Open-vocabulary SOTA (OV2Seg) by a clear margin. For example, on the challenging large-vocabulary VIS dataset (BURST), BriVIS achieves 7.43 mAP and exhibits 49.49% improvement compared to OV2Seg (4.97 mAP).

Robot action planning in the real world is challenging as it requires not only understanding the current state of the environment but also predicting how it will evolve in response to actions. Vision-language-action (VLA), which repurpose large-scale vision-language models for robot action generation using action experts, have achieved notable success across a variety of robotic tasks. Nevertheless, their performance remains constrained by the scope of their training data, exhibiting limited generalization to unseen scenarios and vulnerability to diverse contextual perturbations. More recently, world models have been revisited as an alternative to VLAs. These models, referred to as world action models (WAMs), are built upon world models that are trained on large corpora of video data to predict future states. With minor adaptations, their latent representation can be decoded into robot actions. It has been suggested that their explicit dynamic prediction capacity, combined with spatiotemporal priors acquired from web-scale video pretraining, enables WAMs to generalize more effectively than VLAs. In this paper, we conduct a comparative study of prominent state-of-the-art VLA policies and recently released WAMs. We evaluate their performance on the LIBERO-Plus and RoboTwin 2.0-Plus benchmarks under various visual and language perturbations. Our results show that WAMs achieve strong robustness, with LingBot-VA reaching 74.2% success rate on RoboTwin 2.0-Plus and Cosmos-Policy achieving 82.2% on LIBERO-Plus. While VLAs such as π_{0.5} can achieve comparable robustness on certain tasks, they typically require extensive training with diverse robotic datasets and varied learning objectives. Hybrid approaches that partially incorporate video-based dynamic learning exhibit intermediate robustness, highlighting the importance of how video priors are integrated.

Long-form video content constitutes a significant portion of internet traffic, making automated video summarization an essential research problem. However, existing video summarization datasets are notably limited in their size, constraining the effectiveness of state-of-the-art methods for generalization. Our work aims to overcome this limitation by capitalizing on the abundance of long-form videos with dense speech-to-video alignment and the remarkable capabilities of recent large language models (LLMs) in summarizing long text. We introduce an automated and scalable pipeline for generating a large-scale video summarization dataset using LLMs as Oracle summarizers. By leveraging the generated dataset, we analyze the limitations of existing approaches and propose a new video summarization model that effectively addresses them. To facilitate further research in the field, our work also presents a new benchmark dataset that contains 1200 long videos each with high-quality summaries annotated by professionals. Extensive experiments clearly indicate that our proposed approach sets a new state-of-the-art in video summarization across several benchmarks.

Understanding the workflow of surgical procedures in complex operating rooms requires a deep understanding of the interactions between clinicians and their environment. Surgical activity recognition (SAR) is a key computer vision task that detects activities or phases from multi-view camera recordings. Existing SAR models often fail to account for fine-grained clinician movements and multi-view knowledge, or they require calibrated multi-view camera setups and advanced point-cloud processing to obtain better results. In this work, we propose a novel calibration-free multi-view multi-modal pretraining framework called Multiview Pretraining for Video-Pose Surgical Activity Recognition PreViPS, which aligns 2D pose and vision embeddings across camera views. Our model follows CLIP-style dual-encoder architecture: one encoder processes visual features, while the other encodes human pose embeddings. To handle the continuous 2D human pose coordinates, we introduce a tokenized discrete representation to convert the continuous 2D pose coordinates into discrete pose embeddings, thereby enabling efficient integration within the dual-encoder framework. To bridge the gap between these two modalities, we propose several pretraining objectives using cross- and in-modality geometric constraints within the embedding space and incorporating masked pose token prediction strategy to enhance representation learning. Extensive experiments and ablation studies demonstrate improvements over the strong baselines, while data-efficiency experiments on two distinct operating room datasets further highlight the effectiveness of our approach. We highlight the benefits of our approach for surgical activity recognition in both multi-view and single-view settings, showcasing its practical applicability in complex surgical environments. Code will be made available at: https://github.com/CAMMA-public/PreViPS.

Egocentric video-language pretraining is a crucial step in advancing the understanding of hand-object interactions in first-person scenarios. Despite successes on existing testbeds, we find that current EgoVLMs can be easily misled by simple modifications, such as changing the verbs or nouns in interaction descriptions, with models struggling to distinguish between these changes. This raises the question: Do EgoVLMs truly understand hand-object interactions? To address this question, we introduce a benchmark called EgoHOIBench, revealing the performance limitation of current egocentric models when confronted with such challenges. We attribute this performance gap to insufficient fine-grained supervision and the greater difficulty EgoVLMs experience in recognizing verbs compared to nouns. To tackle these issues, we propose a novel asymmetric contrastive objective named EgoNCE++. For the video-to-text objective, we enhance text supervision by generating negative captions using large language models or leveraging pretrained vocabulary for HOI-related word substitutions. For the text-to-video objective, we focus on preserving an object-centric feature space that clusters video representations based on shared nouns. Extensive experiments demonstrate that EgoNCE++ significantly enhances EgoHOI understanding, leading to improved performance across various EgoVLMs in tasks such as multi-instance retrieval, action recognition, and temporal understanding. Our code is available at https://github.com/xuboshen/EgoNCEpp.

Large-scale video-text pretraining achieves strong performance but depends on noisy, synthetic captions with limited semantic coverage, often overlooking implicit world knowledge such as object motion, 3D geometry, and physical cues. In contrast, masked video modeling (MVM) directly exploits spatiotemporal structures but trails text-supervised methods on general tasks. We find this gap arises from overlooked architectural issues: pixel-level reconstruction struggles with convergence and its low-level requirement often conflicts with semantics, while latent prediction often encourages shortcut learning. To address these, we disentangle the traditional encoder-decoder design into an Encoder-Predictor-Decoder (EPD) framework, where the predictor acts as a latent world model, and propose InternVideo-Next, a two-stage pretraining scheme that builds a semantically consistent yet detail-preserving latent space for this world model. First, conventional linear decoder in pixel MVM enforces the predictor output latent to be linearly projected to, thus separable in pixel space, causing the conflict with semantic abstraction. Our Stage 1 proposes a conditional diffusion decoder and injects reliable image-level semantic priors to enhance semantics and convergence, thus bridging pixel-level fidelity with high-level semantic abstraction. Stage 2 further learns world knowledge by predicting frozen Stage 1 targets within this space, mitigating shortcut learning. Trained on public, unlabeled videos, InternVideo-Next achieves state-of-the-art results across benchmarks and provides a scalable path toward general video representation learning.

Surgical video-language pretraining (VLP) faces unique challenges due to the knowledge domain gap and the scarcity of multi-modal data. This study aims to bridge the gap by addressing issues regarding textual information loss in surgical lecture videos and the spatial-temporal challenges of surgical VLP. We propose a hierarchical knowledge augmentation approach and a novel Procedure-Encoded Surgical Knowledge-Augmented Video-Language Pretraining (PeskaVLP) framework to tackle these issues. The knowledge augmentation uses large language models (LLM) for refining and enriching surgical concepts, thus providing comprehensive language supervision and reducing the risk of overfitting. PeskaVLP combines language supervision with visual self-supervision, constructing hard negative samples and employing a Dynamic Time Warping (DTW) based loss function to effectively comprehend the cross-modal procedural alignment. Extensive experiments on multiple public surgical scene understanding and cross-modal retrieval datasets show that our proposed method significantly improves zero-shot transferring performance and offers a generalist visual representation for further advancements in surgical scene understanding.The code is available at https://github.com/CAMMA-public/SurgVLP

Natural language could play an important role in developing generalist surgical models by providing a broad source of supervision from raw texts. This flexible form of supervision can enable the model's transferability across datasets and tasks as natural language can be used to reference learned visual concepts or describe new ones. In this work, we present HecVL, a novel hierarchical video-language pretraining approach for building a generalist surgical model. Specifically, we construct a hierarchical video-text paired dataset by pairing the surgical lecture video with three hierarchical levels of texts: at clip-level, atomic actions using transcribed audio texts; at phase-level, conceptual text summaries; and at video-level, overall abstract text of the surgical procedure. Then, we propose a novel fine-to-coarse contrastive learning framework that learns separate embedding spaces for the three video-text hierarchies using a single model. By disentangling embedding spaces of different hierarchical levels, the learned multi-modal representations encode short-term and long-term surgical concepts in the same model. Thanks to the injected textual semantics, we demonstrate that the HecVL approach can enable zero-shot surgical phase recognition without any human annotation. Furthermore, we show that the same HecVL model for surgical phase recognition can be transferred across different surgical procedures and medical centers. The code is available at https://github.com/CAMMA-public/SurgVLP

A significant aspiration of offline reinforcement learning (RL) is to develop a generalist agent with high capabilities from large and heterogeneous datasets. However, prior approaches that scale offline RL either rely heavily on expert trajectories or struggle to generalize to diverse unseen tasks. Inspired by the excellent generalization of world model in conditional video generation, we explore the potential of image observation-based world model for scaling offline RL and enhancing generalization on novel tasks. In this paper, we introduce JOWA: Jointly-Optimized World-Action model, an offline model-based RL agent pretrained on multiple Atari games with 6 billion tokens data to learn general-purpose representation and decision-making ability. Our method jointly optimizes a world-action model through a shared transformer backbone, which stabilize temporal difference learning with large models during pretraining. Moreover, we propose a provably efficient and parallelizable planning algorithm to compensate for the Q-value estimation error and thus search out better policies. Experimental results indicate that our largest agent, with 150 million parameters, achieves 78.9% human-level performance on pretrained games using only 10% subsampled offline data, outperforming existing state-of-the-art large-scale offline RL baselines by 31.6% on averange. Furthermore, JOWA scales favorably with model capacity and can sample-efficiently transfer to novel games using only 5k offline fine-tuning data (approximately 4 trajectories) per game, demonstrating superior generalization. We will release codes and model weights at https://github.com/CJReinforce/JOWA

Recent Mamba-based architectures for video understanding demonstrate promising computational efficiency and competitive performance, yet struggle with overfitting issues that hinder their scalability. To overcome this challenge, we introduce VideoMAP, a Hybrid Mamba-Transformer framework featuring a novel pre-training approach. VideoMAP uses a 4:1 Mamba-to-Transformer ratio, effectively balancing computational cost and model capacity. This architecture, combined with our proposed frame-wise masked autoregressive pre-training strategy, delivers significant performance gains when scaling to larger models. Additionally, VideoMAP exhibits impressive sample efficiency, significantly outperforming existing methods with less training data. Experiments show that VideoMAP outperforms existing models across various datasets, including Kinetics-400, Something-Something V2, Breakfast, and COIN. Furthermore, we demonstrate the potential of VideoMAP as a visual encoder for multimodal large language models, highlighting its ability to reduce memory usage and enable the processing of longer video sequences. The code is open-source at https://github.com/yunzeliu/MAP

This paper presents a novel approach for pretraining robotic manipulation Vision-Language-Action (VLA) models using a large corpus of unscripted real-life video recordings of human hand activities. Treating human hand as dexterous robot end-effector, we show that "in-the-wild" egocentric human videos without any annotations can be transformed into data formats fully aligned with existing robotic V-L-A training data in terms of task granularity and labels. This is achieved by the development of a fully-automated holistic human activity analysis approach for arbitrary human hand videos. This approach can generate atomic-level hand activity segments and their language descriptions, each accompanied with framewise 3D hand motion and camera motion. We process a large volume of egocentric videos and create a hand-VLA training dataset containing 1M episodes and 26M frames. This training data covers a wide range of objects and concepts, dexterous manipulation tasks, and environment variations in real life, vastly exceeding the coverage of existing robot data. We design a dexterous hand VLA model architecture and pretrain the model on this dataset. The model exhibits strong zero-shot capabilities on completely unseen real-world observations. Additionally, fine-tuning it on a small amount of real robot action data significantly improves task success rates and generalization to novel objects in real robotic experiments. We also demonstrate the appealing scaling behavior of the model's task performance with respect to pretraining data scale. We believe this work lays a solid foundation for scalable VLA pretraining, advancing robots toward truly generalizable embodied intelligence.

Being able to simulate the outcomes of actions in varied environments will revolutionize the development of generalist agents at scale. However, modeling these world dynamics, especially for dexterous robotics tasks, poses significant challenges due to limited data coverage and scarce action labels. As an endeavor towards this end, we introduce DreamDojo, a foundation world model that learns diverse interactions and dexterous controls from 44k hours of egocentric human videos. Our data mixture represents the largest video dataset to date for world model pretraining, spanning a wide range of daily scenarios with diverse objects and skills. To address the scarcity of action labels, we introduce continuous latent actions as unified proxy actions, enhancing interaction knowledge transfer from unlabeled videos. After post-training on small-scale target robot data, DreamDojo demonstrates a strong understanding of physics and precise action controllability. We also devise a distillation pipeline that accelerates DreamDojo to a real-time speed of 10.81 FPS and further improves context consistency. Our work enables several important applications based on generative world models, including live teleoperation, policy evaluation, and model-based planning. Systematic evaluation on multiple challenging out-of-distribution (OOD) benchmarks verifies the significance of our method for simulating open-world, contact-rich tasks, paving the way for general-purpose robot world models.

Video text-based visual question answering (Video TextVQA) is a practical task that aims to answer questions by jointly reasoning textual and visual information in a given video. Inspired by the development of TextVQA in image domain, existing Video TextVQA approaches leverage a language model (e.g. T5) to process text-rich multiple frames and generate answers auto-regressively. Nevertheless, the spatio-temporal relationships among visual entities (including scene text and objects) will be disrupted and models are susceptible to interference from unrelated information, resulting in irrational reasoning and inaccurate answering. To tackle these challenges, we propose the TEA (stands for ``Track thE Answer'') method that better extends the generative TextVQA framework from image to video. TEA recovers the spatio-temporal relationships in a complementary way and incorporates OCR-aware clues to enhance the quality of reasoning questions. Extensive experiments on several public Video TextVQA datasets validate the effectiveness and generalization of our framework. TEA outperforms existing TextVQA methods, video-language pretraining methods and video large language models by great margins.

Recent advances in internet-scale video data pretraining have led to the development of text-to-video generative models that can create high-quality videos across a broad range of visual concepts, synthesize realistic motions and render complex objects. Hence, these generative models have the potential to become general-purpose simulators of the physical world. However, it is unclear how far we are from this goal with the existing text-to-video generative models. To this end, we present VideoPhy, a benchmark designed to assess whether the generated videos follow physical commonsense for real-world activities (e.g. marbles will roll down when placed on a slanted surface). Specifically, we curate diverse prompts that involve interactions between various material types in the physical world (e.g., solid-solid, solid-fluid, fluid-fluid). We then generate videos conditioned on these captions from diverse state-of-the-art text-to-video generative models, including open models (e.g., CogVideoX) and closed models (e.g., Lumiere, Dream Machine). Our human evaluation reveals that the existing models severely lack the ability to generate videos adhering to the given text prompts, while also lack physical commonsense. Specifically, the best performing model, CogVideoX-5B, generates videos that adhere to the caption and physical laws for 39.6% of the instances. VideoPhy thus highlights that the video generative models are far from accurately simulating the physical world. Finally, we propose an auto-evaluator, VideoCon-Physics, to assess the performance reliably for the newly released models.

Self-supervised learning has emerged as a powerful paradigm for label-free model pretraining, particularly in the video domain, where manual annotation is costly and time-intensive. However, existing self-supervised approaches employ diverse experimental setups, making direct comparisons challenging due to the absence of a standardized benchmark. In this work, we establish a unified benchmark that enables fair comparisons across different methods. Additionally, we systematically investigate five critical aspects of self-supervised learning in videos: (1) dataset size, (2) model complexity, (3) data distribution, (4) data noise, and (5) feature representations. To facilitate this study, we evaluate six self-supervised learning methods across six network architectures, conducting extensive experiments on five benchmark datasets and assessing performance on two distinct downstream tasks. Our analysis reveals key insights into the interplay between pretraining strategies, dataset characteristics, pretext tasks, and model architectures. Furthermore, we extend these findings to Video Foundation Models (ViFMs), demonstrating their relevance in large-scale video representation learning. Finally, leveraging these insights, we propose a novel approach that significantly reduces training data requirements while surpassing state-of-the-art methods that rely on 10% more pretraining data. We believe this work will guide future research toward a deeper understanding of self-supervised video representation learning and its broader implications.

Recent work has studied text-to-audio synthesis using large amounts of paired text-audio data. However, audio recordings with high-quality text annotations can be difficult to acquire. In this work, we approach text-to-audio synthesis using unlabeled videos and pretrained language-vision models. We propose to learn the desired text-audio correspondence by leveraging the visual modality as a bridge. We train a conditional diffusion model to generate the audio track of a video, given a video frame encoded by a pretrained contrastive language-image pretraining (CLIP) model. At test time, we first explore performing a zero-shot modality transfer and condition the diffusion model with a CLIP-encoded text query. However, we observe a noticeable performance drop with respect to image queries. To close this gap, we further adopt a pretrained diffusion prior model to generate a CLIP image embedding given a CLIP text embedding. Our results show the effectiveness of the proposed method, and that the pretrained diffusion prior can reduce the modality transfer gap. While we focus on text-to-audio synthesis, the proposed model can also generate audio from image queries, and it shows competitive performance against a state-of-the-art image-to-audio synthesis model in a subjective listening test. This study offers a new direction of approaching text-to-audio synthesis that leverages the naturally-occurring audio-visual correspondence in videos and the power of pretrained language-vision models.

We present CLIP2Video network to transfer the image-language pre-training model to video-text retrieval in an end-to-end manner. Leading approaches in the domain of video-and-language learning try to distill the spatio-temporal video features and multi-modal interaction between videos and languages from a large-scale video-text dataset. Different from them, we leverage pretrained image-language model, simplify it as a two-stage framework with co-learning of image-text and enhancing temporal relations between video frames and video-text respectively, make it able to train on comparatively small datasets. Specifically, based on the spatial semantics captured by Contrastive Language-Image Pretraining (CLIP) model, our model involves a Temporal Difference Block to capture motions at fine temporal video frames, and a Temporal Alignment Block to re-align the tokens of video clips and phrases and enhance the multi-modal correlation. We conduct thorough ablation studies, and achieve state-of-the-art performance on major text-to-video and video-to-text retrieval benchmarks, including new records of retrieval accuracy on MSR-VTT, MSVD and VATEX.

Video generation is a critical pathway toward world models, with efficient long video inference as a key capability. Toward this end, we introduce LongCat-Video, a foundational video generation model with 13.6B parameters, delivering strong performance across multiple video generation tasks. It particularly excels in efficient and high-quality long video generation, representing our first step toward world models. Key features include: Unified architecture for multiple tasks: Built on the Diffusion Transformer (DiT) framework, LongCat-Video supports Text-to-Video, Image-to-Video, and Video-Continuation tasks with a single model; Long video generation: Pretraining on Video-Continuation tasks enables LongCat-Video to maintain high quality and temporal coherence in the generation of minutes-long videos; Efficient inference: LongCat-Video generates 720p, 30fps videos within minutes by employing a coarse-to-fine generation strategy along both the temporal and spatial axes. Block Sparse Attention further enhances efficiency, particularly at high resolutions; Strong performance with multi-reward RLHF: Multi-reward RLHF training enables LongCat-Video to achieve performance on par with the latest closed-source and leading open-source models. Code and model weights are publicly available to accelerate progress in the field.

As humans, we understand events in the visual world contextually, performing multimodal reasoning across time to make inferences about the past, present, and future. We introduce MERLOT, a model that learns multimodal script knowledge by watching millions of YouTube videos with transcribed speech -- in an entirely label-free, self-supervised manner. By pretraining with a mix of both frame-level (spatial) and video-level (temporal) objectives, our model not only learns to match images to temporally corresponding words, but also to contextualize what is happening globally over time. As a result, MERLOT exhibits strong out-of-the-box representations of temporal commonsense, and achieves state-of-the-art performance on 12 different video QA datasets when finetuned. It also transfers well to the world of static images, allowing models to reason about the dynamic context behind visual scenes. On Visual Commonsense Reasoning, MERLOT answers questions correctly with 80.6% accuracy, outperforming state-of-the-art models of similar size by over 3%, even those that make heavy use of auxiliary supervised data (like object bounding boxes). Ablation analyses demonstrate the complementary importance of: 1) training on videos versus static images; 2) scaling the magnitude and diversity of the pretraining video corpus; and 3) using diverse objectives that encourage full-stack multimodal reasoning, from the recognition to cognition level.

Pretraining RL models on offline video datasets is a promising way to improve their training efficiency in online tasks, but challenging due to the inherent mismatch in tasks, dynamics, and behaviors across domains. A recent model, APV, sidesteps the accompanied action records in offline datasets and instead focuses on pretraining a task-irrelevant, action-free world model within the source domains. We present Vid2Act, a model-based RL method that learns to transfer valuable action-conditioned dynamics and potentially useful action demonstrations from offline to online settings. The main idea is to use the world models not only as simulators for behavior learning but also as tools to measure the domain relevance for both dynamics representation transfer and policy transfer. Specifically, we train the world models to generate a set of time-varying task similarities using a domain-selective knowledge distillation loss. These similarities serve two purposes: (i) adaptively transferring the most useful source knowledge to facilitate dynamics learning, and (ii) learning to replay the most relevant source actions to guide the target policy. We demonstrate the advantages of Vid2Act over the action-free visual RL pretraining method in both Meta-World and DeepMind Control Suite.

Localizing unusual activities, such as human errors or surveillance incidents, in videos holds practical significance. However, current video understanding models struggle with localizing these unusual events likely because of their insufficient representation in models' pretraining datasets. To explore foundation models' capability in localizing unusual activity, we introduce UAL-Bench, a comprehensive benchmark for unusual activity localization, featuring three video datasets: UAG-OOPS, UAG-SSBD, UAG-FunQA, and an instruction-tune dataset: OOPS-UAG-Instruct, to improve model capabilities. UAL-Bench evaluates three approaches: Video-Language Models (Vid-LLMs), instruction-tuned Vid-LLMs, and a novel integration of Vision-Language Models and Large Language Models (VLM-LLM). Our results show the VLM-LLM approach excels in localizing short-span unusual events and predicting their onset (start time) more accurately than Vid-LLMs. We also propose a new metric, R@1, TD <= p, to address limitations in existing evaluation methods. Our findings highlight the challenges posed by long-duration videos, particularly in autism diagnosis scenarios, and the need for further advancements in localization techniques. Our work not only provides a benchmark for unusual activity localization but also outlines the key challenges for existing foundation models, suggesting future research directions on this important task.

This paper presents RynnVLA-001, a vision-language-action(VLA) model built upon large-scale video generative pretraining from human demonstrations. We propose a novel two-stage pretraining methodology. The first stage, Ego-Centric Video Generative Pretraining, trains an Image-to-Video model on 12M ego-centric manipulation videos to predict future frames conditioned on an initial frame and a language instruction. The second stage, Human-Centric Trajectory-Aware Modeling, extends this by jointly predicting future keypoint trajectories, thereby effectively bridging visual frame prediction with action prediction. Furthermore, to enhance action representation, we propose ActionVAE, a variational autoencoder that compresses sequences of actions into compact latent embeddings, reducing the complexity of the VLA output space. When finetuned on the same downstream robotics datasets, RynnVLA-001 achieves superior performance over state-of-the-art baselines, demonstrating that the proposed pretraining strategy provides a more effective initialization for VLA models.

Video-language pre-trained models have shown remarkable success in guiding video question-answering (VideoQA) tasks. However, due to the length of video sequences, training large-scale video-based models incurs considerably higher costs than training image-based ones. This motivates us to leverage the knowledge from image-based pretraining, despite the obvious gaps between image and video domains. To bridge these gaps, in this paper, we propose Tem-Adapter, which enables the learning of temporal dynamics and complex semantics by a visual Temporal Aligner and a textual Semantic Aligner. Unlike conventional pretrained knowledge adaptation methods that only concentrate on the downstream task objective, the Temporal Aligner introduces an extra language-guided autoregressive task aimed at facilitating the learning of temporal dependencies, with the objective of predicting future states based on historical clues and language guidance that describes event progression. Besides, to reduce the semantic gap and adapt the textual representation for better event description, we introduce a Semantic Aligner that first designs a template to fuse question and answer pairs as event descriptions and then learns a Transformer decoder with the whole video sequence as guidance for refinement. We evaluate Tem-Adapter and different pre-train transferring methods on two VideoQA benchmarks, and the significant performance improvement demonstrates the effectiveness of our method.

Recent advances in pretraining general foundation models have significantly improved performance across diverse downstream tasks. While autoregressive (AR) generative models like GPT have revolutionized NLP, most visual generative pretraining methods still rely on BERT-style masked modeling, which often disregards the temporal information essential for video analysis. The few existing autoregressive visual pretraining methods suffer from issues such as inaccurate semantic localization and poor generation quality, leading to poor semantics. In this work, we propose NExT-Vid, a novel autoregressive visual generative pretraining framework that utilizes masked next-frame prediction to jointly model images and videos. NExT-Vid introduces a context-isolated autoregressive predictor to decouple semantic representation from target decoding, and a conditioned flow-matching decoder to enhance generation quality and diversity. Through context-isolated flow-matching pretraining, our approach achieves strong representations. Extensive experiments on large-scale pretrained models demonstrate that our proposed method consistently outperforms previous generative pretraining methods for visual representation learning via attentive probing in downstream classification.

Contrastive language-image pretraining has shown great success in learning visual-textual joint representation from web-scale data, demonstrating remarkable "zero-shot" generalization ability for various image tasks. However, how to effectively expand such new language-image pretraining methods to video domains is still an open problem. In this work, we present a simple yet effective approach that adapts the pretrained language-image models to video recognition directly, instead of pretraining a new model from scratch. More concretely, to capture the long-range dependencies of frames along the temporal dimension, we propose a cross-frame attention mechanism that explicitly exchanges information across frames. Such module is lightweight and can be plugged into pretrained language-image models seamlessly. Moreover, we propose a video-specific prompting scheme, which leverages video content information for generating discriminative textual prompts. Extensive experiments demonstrate that our approach is effective and can be generalized to different video recognition scenarios. In particular, under fully-supervised settings, our approach achieves a top-1 accuracy of 87.1% on Kinectics-400, while using 12 times fewer FLOPs compared with Swin-L and ViViT-H. In zero-shot experiments, our approach surpasses the current state-of-the-art methods by +7.6% and +14.9% in terms of top-1 accuracy under two popular protocols. In few-shot scenarios, our approach outperforms previous best methods by +32.1% and +23.1% when the labeled data is extremely limited. Code and models are available at https://aka.ms/X-CLIP

Large-scale image-language pretrained models, e.g., CLIP, have demonstrated remarkable proficiency in acquiring general multi-modal knowledge through web-scale image-text data. Despite the impressive performance of image-language models on various image tasks, how to effectively expand them on general video understanding remains an area of ongoing exploration. In this paper, we investigate the image-to-video transferring from the perspective of the model and the data, unveiling two key obstacles impeding the adaptation of image-language models: non-generalizable temporal modeling and partially misaligned video-text data. To address these challenges, we propose Spatial-Temporal Auxiliary Network with Mutual-guided alignment module (Mug-STAN), a simple yet effective framework extending image-text model to diverse video tasks and video-text data.Specifically, STAN adopts a branch structure with decomposed spatial-temporal modules to enable generalizable temporal modeling, while Mug suppresses misalignment by introducing token-wise feature aggregation of either modality from the other. Extensive experimental results verify Mug-STAN significantly improves adaptation of language-image pretrained models such as CLIP and CoCa at both video-text post-pretraining and finetuning stages. With our solution, state-of-the-art zero-shot and finetuning results on various downstream datasets, including MSR-VTT, DiDeMo, LSMDC, Kinetics-400, Something-Something-2, HMDB-51, UCF- 101, and AVA, are achieved. Moreover, by integrating pretrained Mug-STAN with the emerging multimodal dialogue model, we can realize zero-shot video chatting. Codes are available at https://github.com/farewellthree/STAN

Procedural video representation learning is an active research area where the objective is to learn an agent which can anticipate and forecast the future given the present video input, typically in conjunction with textual annotations. Prior works often rely on large-scale pretraining of visual encoders and prediction models with language supervision. However, the necessity and effectiveness of extending compute intensive pretraining to learn video clip sequences with noisy text supervision have not yet been fully validated by previous works. In this work, we show that a strong off-the-shelf frozen pretrained visual encoder, along with a well designed prediction model, can achieve state-of-the-art (SoTA) performance in forecasting and procedural planning without the need for pretraining the prediction model, nor requiring additional supervision from language or ASR. Instead of learning representations from pixel space, our method utilizes the latent embedding space of publicly available vision encoders. By conditioning on frozen clip-level embeddings from observed steps to predict the actions of unseen steps, our prediction model is able to learn robust representations for forecasting through iterative denoising - leveraging the recent advances in diffusion transformers (Peebles & Xie, 2023). Empirical studies over a total of five procedural learning tasks across four datasets (NIV, CrossTask, COIN and Ego4D-v2) show that our model advances the strong baselines in long-horizon action anticipation (+2.6% in Verb ED@20, +3.1% in Noun ED@20), and significantly improves the SoTA in step forecasting (+5.0%), task classification (+3.8%), and procedure planning tasks (up to +2.28% in success rate, +3.39% in mAcc, and +0.90% in mIoU).

Recent advancements in visual generation technologies have markedly increased the scale and availability of video datasets, which are crucial for training effective video generation models. However, a significant lack of high-quality, human-centric video datasets presents a challenge to progress in this field. To bridge this gap, we introduce OpenHumanVid, a large-scale and high-quality human-centric video dataset characterized by precise and detailed captions that encompass both human appearance and motion states, along with supplementary human motion conditions, including skeleton sequences and speech audio. To validate the efficacy of this dataset and the associated training strategies, we propose an extension of existing classical diffusion transformer architectures and conduct further pretraining of our models on the proposed dataset. Our findings yield two critical insights: First, the incorporation of a large-scale, high-quality dataset substantially enhances evaluation metrics for generated human videos while preserving performance in general video generation tasks. Second, the effective alignment of text with human appearance, human motion, and facial motion is essential for producing high-quality video outputs. Based on these insights and corresponding methodologies, the straightforward extended network trained on the proposed dataset demonstrates an obvious improvement in the generation of human-centric videos. Project page https://fudan-generative-vision.github.io/OpenHumanVid

Building video-language foundation models is costly and difficult due to the redundant nature of video data and the lack of high-quality video-language datasets. In this paper, we propose an efficient framework to harvest video foundation models from image ones. Our method is intuitively simple by randomly dropping input video patches and masking out input text during the post-pretraining procedure. The patch dropping boosts the training efficiency significantly and text masking enforces the learning of cross-modal fusion. We conduct extensive experiments to validate the effectiveness of our method on a wide range of video-language downstream tasks including various zero-shot tasks, video question answering, and video-text retrieval. Despite its simplicity, our method achieves state-of-the-art performances, which are comparable to some heavily pretrained video foundation models. Our method is extremely efficient and can be trained in less than one day on 8 GPUs, requiring only WebVid-10M as pretraining data. We hope our method can serve as a simple yet strong counterpart for prevalent video foundation models, provide useful insights when building them, and make large pretrained models more accessible and sustainable. This is part of the InternVideo project https://github.com/OpenGVLab/InternVideo.

Spatial intelligence requires visual representations that capture both semantic objects and geometric structure in the physical world. To support this, two major pre-training schemes are now widely used as foundation backbones: Vision-Language Models (VLMs), which use language supervision to align visual observations with semantic concepts, and Video Generation Models (VGMs), which learn from temporally evolving visual worlds. However, it still remains unclear which pre-training scheme provides a better representation substrate for spatial intelligence. In this paper, we present the first systematic frozen-feature probing study of VLMs and VGMs across three representative axes of spatial intelligence: semantic tagging, instance grouping, and 3D geometry prediction. Using the lightweight probe, our framework enables a controlled comparison of what information is already encoded in frozen representations from two model families. Experimental results reveal a clear complementarity: VLMs are stronger at semantic tagging and instance grouping, while VGMs provide more accessible signals for dense geometry and camera motion. Moreover, a naive fusion of the two already yields a representation that excels at both geometry and semantics, suggesting a promising direction for building stronger spatial-intelligence backbones by effectively integrating features from both model families. Our code is available at https://github.com/om-ai-lab/Probing-VLM-VGM{https://github.com/om-ai-lab/Probing-VLM-VGM}.

Imitation learning has emerged as a promising approach towards building generalist robots. However, scaling imitation learning for large robot foundation models remains challenging due to its reliance on high-quality expert demonstrations. Meanwhile, large amounts of video data depicting a wide range of environments and diverse behaviors are readily available. This data provides a rich source of information about real-world dynamics and agent-environment interactions. Leveraging this data directly for imitation learning, however, has proven difficult due to the lack of action annotation required for most contemporary methods. In this work, we present Unified World Models (UWM), a framework that allows for leveraging both video and action data for policy learning. Specifically, a UWM integrates an action diffusion process and a video diffusion process within a unified transformer architecture, where independent diffusion timesteps govern each modality. We show that by simply controlling each diffusion timestep, UWM can flexibly represent a policy, a forward dynamics, an inverse dynamics, and a video generator. Through simulated and real-world experiments, we show that: (1) UWM enables effective pretraining on large-scale multitask robot datasets with both dynamics and action predictions, resulting in more generalizable and robust policies than imitation learning, (2) UWM naturally facilitates learning from action-free video data through independent control of modality-specific diffusion timesteps, further improving the performance of finetuned policies. Our results suggest that UWM offers a promising step toward harnessing large, heterogeneous datasets for scalable robot learning, and provides a simple unification between the often disparate paradigms of imitation learning and world modeling. Videos and code are available at https://weirdlabuw.github.io/uwm/.

In this work, we introduce Vid2Seq, a multi-modal single-stage dense event captioning model pretrained on narrated videos which are readily-available at scale. The Vid2Seq architecture augments a language model with special time tokens, allowing it to seamlessly predict event boundaries and textual descriptions in the same output sequence. Such a unified model requires large-scale training data, which is not available in current annotated datasets. We show that it is possible to leverage unlabeled narrated videos for dense video captioning, by reformulating sentence boundaries of transcribed speech as pseudo event boundaries, and using the transcribed speech sentences as pseudo event captions. The resulting Vid2Seq model pretrained on the YT-Temporal-1B dataset improves the state of the art on a variety of dense video captioning benchmarks including YouCook2, ViTT and ActivityNet Captions. Vid2Seq also generalizes well to the tasks of video paragraph captioning and video clip captioning, and to few-shot settings. Our code is publicly available at https://antoyang.github.io/vid2seq.html.

We present PFP, a neural network structure to compress long videos into short contexts, with an explicit pretraining objective to preserve the high-frequency details of single frames at arbitrary temporal positions. The baseline model can compress a 20-second video into a context at about 5k length, where random frames can be retrieved with perceptually preserved appearances. Such pretrained models can be directly fine-tuned as memory encoders for autoregressive video models, enabling long history memory with low context cost and relatively low fidelity loss. We evaluate the framework with ablative settings and discuss the trade-offs of possible neural architecture designs.

Learning specific hands-on skills such as cooking, car maintenance, and home repairs increasingly happens via instructional videos. The user experience with such videos is known to be improved by meta-information such as time-stamped annotations for the main steps involved. Generating such annotations automatically is challenging, and we describe here two relevant contributions. First, we construct and release a new dense video captioning dataset, Video Timeline Tags (ViTT), featuring a variety of instructional videos together with time-stamped annotations. Second, we explore several multimodal sequence-to-sequence pretraining strategies that leverage large unsupervised datasets of videos and caption-like texts. We pretrain and subsequently finetune dense video captioning models using both YouCook2 and ViTT. We show that such models generalize well and are robust over a wide variety of instructional videos.

Contrastive language-image pretraining (CLIP) has demonstrated remarkable success in various image tasks. However, how to extend CLIP with effective temporal modeling is still an open and crucial problem. Existing factorized or joint spatial-temporal modeling trades off between the efficiency and performance. While modeling temporal information within straight through tube is widely adopted in literature, we find that simple frame alignment already provides enough essence without temporal attention. To this end, in this paper, we proposed a novel Implicit Learnable Alignment (ILA) method, which minimizes the temporal modeling effort while achieving incredibly high performance. Specifically, for a frame pair, an interactive point is predicted in each frame, serving as a mutual information rich region. By enhancing the features around the interactive point, two frames are implicitly aligned. The aligned features are then pooled into a single token, which is leveraged in the subsequent spatial self-attention. Our method allows eliminating the costly or insufficient temporal self-attention in video. Extensive experiments on benchmarks demonstrate the superiority and generality of our module. Particularly, the proposed ILA achieves a top-1 accuracy of 88.7% on Kinetics-400 with much fewer FLOPs compared with Swin-L and ViViT-H. Code is released at https://github.com/Francis-Rings/ILA .

We present VideoPoet, a language model capable of synthesizing high-quality video, with matching audio, from a large variety of conditioning signals. VideoPoet employs a decoder-only transformer architecture that processes multimodal inputs -- including images, videos, text, and audio. The training protocol follows that of Large Language Models (LLMs), consisting of two stages: pretraining and task-specific adaptation. During pretraining, VideoPoet incorporates a mixture of multimodal generative objectives within an autoregressive Transformer framework. The pretrained LLM serves as a foundation that can be adapted for a range of video generation tasks. We present empirical results demonstrating the model's state-of-the-art capabilities in zero-shot video generation, specifically highlighting VideoPoet's ability to generate high-fidelity motions. Project page: http://sites.research.google/videopoet/

While foundation models have advanced surgical video analysis, current approaches rely predominantly on pixel-level reconstruction objectives that waste model capacity on low-level visual details - such as smoke, specular reflections, and fluid motion - rather than semantic structures essential for surgical understanding. We present UniSurg, a video-native foundation model that shifts the learning paradigm from pixel-level reconstruction to latent motion prediction. Built on the Video Joint Embedding Predictive Architecture (V-JEPA), UniSurg introduces three key technical innovations tailored to surgical videos: 1) motion-guided latent prediction to prioritize semantically meaningful regions, 2) spatiotemporal affinity self-distillation to enforce relational consistency, and 3) feature diversity regularization to prevent representation collapse in texture-sparse surgical scenes. To enable large-scale pretraining, we curate UniSurg-15M, the largest surgical video dataset to date, comprising 3,658 hours of video from 50 sources across 13 anatomical regions. Extensive experiments across 17 benchmarks demonstrate that UniSurg significantly outperforms state-of-the-art methods on surgical workflow recognition (+14.6% F1 on EgoSurgery, +10.3% on PitVis), action triplet recognition (39.54% mAP-IVT on CholecT50), skill assessment, polyp segmentation, and depth estimation. These results establish UniSurg as a new standard for universal, motion-oriented surgical video understanding.

Recent video and language pretraining frameworks lack the ability to generate sentences. We present Multimodal Video Generative Pretraining (MV-GPT), a new pretraining framework for learning from unlabelled videos which can be effectively used for generative tasks such as multimodal video captioning. Unlike recent video-language pretraining frameworks, our framework trains both a multimodal video encoder and a sentence decoder jointly. To overcome the lack of captions in unlabelled videos, we leverage the future utterance as an additional text source and propose a bidirectional generation objective -- we generate future utterances given the present mulitmodal context, and also the present utterance given future observations. With this objective, we train an encoder-decoder model end-to-end to generate a caption from raw pixels and transcribed speech directly. Our model achieves state-of-the-art performance for multimodal video captioning on four standard benchmarks, as well as for other video understanding tasks such as VideoQA, video retrieval and action classification.

We present a simplified, task-agnostic multi-modal pre-training approach that can accept either video or text input, or both for a variety of end tasks. Existing pre-training are task-specific by adopting either a single cross-modal encoder that requires both modalities, limiting their use for retrieval-style end tasks or more complex multitask learning with two unimodal encoders, limiting early cross-modal fusion. We instead introduce new pretraining masking schemes that better mix across modalities (e.g. by forcing masks for text to predict the closest video embeddings) while also maintaining separability (e.g. unimodal predictions are sometimes required, without using all the input). Experimental results show strong performance across a wider range of tasks than any previous methods, often outperforming task-specific pre-training. Code is made available at https://github.com/pytorch/fairseq/tree/main/examples/MMPT.

Prevailing Vision-Language-Action Models (VLAs) for robotic manipulation are built upon vision-language backbones pretrained on large-scale, but disconnected static web data. As a result, despite improved semantic generalization, the policy must implicitly infer complex physical dynamics and temporal dependencies solely from robot trajectories. This reliance creates an unsustainable data burden, necessitating continuous, large-scale expert data collection to compensate for the lack of innate physical understanding. We contend that while vision-language pretraining effectively captures semantic priors, it remains blind to physical causality. A more effective paradigm leverages video to jointly capture semantics and visual dynamics during pretraining, thereby isolating the remaining task of low-level control. To this end, we introduce mimic-video, a novel Video-Action Model (VAM) that pairs a pretrained Internet-scale video model with a flow matching-based action decoder conditioned on its latent representations. The decoder serves as an Inverse Dynamics Model (IDM), generating low-level robot actions from the latent representation of video-space action plans. Our extensive evaluation shows that our approach achieves state-of-the-art performance on simulated and real-world robotic manipulation tasks, improving sample efficiency by 10x and convergence speed by 2x compared to traditional VLA architectures.

Generalist Vision-Language-Action models are currently hindered by the scarcity of robotic data compared to the abundance of human video demonstrations. Existing Latent Action Models attempt to leverage video data but often suffer from visual entanglement, capturing noise rather than manipulation skills. To address this, we propose Contrastive Latent Action Pretraining (CLAP), a framework that aligns the visual latent space from videos with a proprioceptive latent space from robot trajectories. By employing contrastive learning, CLAP maps video transitions onto a quantized, physically executable codebook. Building on this representation, we introduce a dual-formulation VLA framework offering both CLAP-NTP, an autoregressive model excelling at instruction following and object generalization, and CLAP-RF, a Rectified Flow-based policy designed for high-frequency, precise manipulation. Furthermore, we propose a Knowledge Matching (KM) regularization strategy to mitigate catastrophic forgetting during fine-tuning. Extensive experiments demonstrate that CLAP significantly outperforms strong baselines, enabling the effective transfer of skills from human videos to robotic execution. Project page: https://lin-shan.com/CLAP/.

Vision-language-action (VLA) models have advanced robot manipulation through large-scale pretraining, but real-world deployment remains challenging due to partial observability and delayed feedback. Reinforcement learning addresses this via value functions, which assess task progress and guide policy improvement. However, existing value models built on vision-language models (VLMs) struggle to capture temporal dynamics, undermining reliable value estimation in long-horizon tasks. In this paper, we propose ViVa, a video-generative value model that repurposes a pretrained video generator for value estimation. Taking the current observation and robot proprioception as input, ViVa jointly predicts future proprioception and a scalar value for the current state. By leveraging the spatiotemporal priors of a pretrained video generator, our approach grounds value estimation in anticipated embodiment dynamics, moving beyond static snapshots to intrinsically couple value with foresight. Integrated into RECAP, ViVa delivers substantial improvements on real-world box assembly. Qualitative analysis across all three tasks confirms that ViVa produces more reliable value signals, accurately reflecting task progress. By leveraging spatiotemporal priors from video corpora, ViVa also generalizes to novel objects, highlighting the promise of video-generative models for value estimation.

We evaluate existing foundation models video understanding capabilities using a carefully designed experiment protocol consisting of three hallmark tasks (action recognition, temporal localization, and spatiotemporal localization), eight datasets well received by the community, and four adaptation methods tailoring a foundation model (FM) for a downstream task. Moreover, we propose a scalar VideoGLUE score (VGS) to measure an FMs efficacy and efficiency when adapting to general video understanding tasks. Our main findings are as follows. First, task-specialized models significantly outperform the six FMs studied in this work, in sharp contrast to what FMs have achieved in natural language and image understanding. Second,video-native FMs, whose pretraining data contains the video modality, are generally better than image-native FMs in classifying motion-rich videos, localizing actions in time, and understanding a video of more than one action. Third, the video-native FMs can perform well on video tasks under light adaptations to downstream tasks(e.g., freezing the FM backbones), while image-native FMs win in full end-to-end finetuning. The first two observations reveal the need and tremendous opportunities to conduct research on video-focused FMs, and the last confirms that both tasks and adaptation methods matter when it comes to the evaluation of FMs.

The Video-to-Audio (V2A) model has recently gained attention for its practical application in generating audio directly from silent videos, particularly in video/film production. However, previous methods in V2A have limited generation quality in terms of temporal synchronization and audio-visual relevance. We present Diff-Foley, a synchronized Video-to-Audio synthesis method with a latent diffusion model (LDM) that generates high-quality audio with improved synchronization and audio-visual relevance. We adopt contrastive audio-visual pretraining (CAVP) to learn more temporally and semantically aligned features, then train an LDM with CAVP-aligned visual features on spectrogram latent space. The CAVP-aligned features enable LDM to capture the subtler audio-visual correlation via a cross-attention module. We further significantly improve sample quality with `double guidance'. Diff-Foley achieves state-of-the-art V2A performance on current large scale V2A dataset. Furthermore, we demonstrate Diff-Foley practical applicability and generalization capabilities via downstream finetuning. Project Page: see https://diff-foley.github.io/

Large-scale pretrained transformers have created milestones in text (GPT-3) and text-to-image (DALL-E and CogView) generation. Its application to video generation is still facing many challenges: The potential huge computation cost makes the training from scratch unaffordable; The scarcity and weak relevance of text-video datasets hinder the model understanding complex movement semantics. In this work, we present 9B-parameter transformer CogVideo, trained by inheriting a pretrained text-to-image model, CogView2. We also propose multi-frame-rate hierarchical training strategy to better align text and video clips. As (probably) the first open-source large-scale pretrained text-to-video model, CogVideo outperforms all publicly available models at a large margin in machine and human evaluations.

Current video-to-audio (V2A) methods struggle in complex multi-event scenarios (video scenarios involving multiple sound sources, sound events, or transitions) due to two critical limitations. First, existing methods face challenges in precisely aligning intricate semantic information together with rapid dynamic features. Second, foundational training lacks quantitative preference optimization for semantic-temporal alignment and audio quality. As a result, it fails to enhance integrated generation quality in cluttered multi-event scenes. To address these core limitations, this study proposes a novel V2A framework: MultiSoundGen. It introduces direct preference optimization (DPO) into the V2A domain, leveraging audio-visual pretraining (AVP) to enhance performance in complex multi-event scenarios. Our contributions include two key innovations: the first is SlowFast Contrastive AVP (SF-CAVP), a pioneering AVP model with a unified dual-stream architecture. SF-CAVP explicitly aligns core semantic representations and rapid dynamic features of audio-visual data to handle multi-event complexity; second, we integrate the DPO method into V2A task and propose AVP-Ranked Preference Optimization (AVP-RPO). It uses SF-CAVP as a reward model to quantify and prioritize critical semantic-temporal matches while enhancing audio quality. Experiments demonstrate that MultiSoundGen achieves state-of-the-art (SOTA) performance in multi-event scenarios, delivering comprehensive gains across distribution matching, audio quality, semantic alignment, and temporal synchronization. Demos are available at https://v2aresearch.github.io/MultiSoundGen/.

We introduce Human-like Video Models (HVM-1), large-scale video models pretrained with nearly 5000 hours of curated human-like video data (mostly egocentric, temporally extended, continuous video recordings), using the spatiotemporal masked autoencoder (ST-MAE) algorithm. We release two 633M parameter models trained at spatial resolutions of 224x224 and 448x448 pixels. We evaluate the performance of these models in downstream few-shot video and image recognition tasks and compare them against a model pretrained with 1330 hours of short action-oriented video clips from YouTube (Kinetics-700). HVM-1 models perform competitively against the Kinetics-700 pretrained model in downstream evaluations despite substantial qualitative differences between the spatiotemporal characteristics of the corresponding pretraining datasets. HVM-1 models also learn more accurate and more robust object representations compared to models pretrained with the image-based MAE algorithm on the same data, demonstrating the potential benefits of learning to predict temporal regularities in natural videos for learning better object representations.

With the burgeoning growth of online video platforms and the escalating volume of video content, the demand for proficient video understanding tools has intensified markedly. Given the remarkable capabilities of Large Language Models (LLMs) in language and multimodal tasks, this survey provides a detailed overview of the recent advancements in video understanding harnessing the power of LLMs (Vid-LLMs). The emergent capabilities of Vid-LLMs are surprisingly advanced, particularly their ability for open-ended spatial-temporal reasoning combined with commonsense knowledge, suggesting a promising path for future video understanding. We examine the unique characteristics and capabilities of Vid-LLMs, categorizing the approaches into four main types: LLM-based Video Agents, Vid-LLMs Pretraining, Vid-LLMs Instruction Tuning, and Hybrid Methods. Furthermore, this survey presents a comprehensive study of the tasks, datasets, and evaluation methodologies for Vid-LLMs. Additionally, it explores the expansive applications of Vid-LLMs across various domains, highlighting their remarkable scalability and versatility in real-world video understanding challenges. Finally, it summarizes the limitations of existing Vid-LLMs and outlines directions for future research. For more information, readers are recommended to visit the repository at https://github.com/yunlong10/Awesome-LLMs-for-Video-Understanding.

Vision-Language-Action (VLA) models have emerged as a promising paradigm for robot learning, but their representations are still largely inherited from static image-text pretraining, leaving physical dynamics to be learned from comparatively limited action data. Generative video models, by contrast, encode rich spatiotemporal structure and implicit physics, making them a compelling foundation for robotic manipulation. But their potentials are not fully explored in the literature. To bridge the gap, we introduce DiT4DiT, an end-to-end Video-Action Model that couples a video Diffusion Transformer with an action Diffusion Transformer in a unified cascaded framework. Instead of relying on reconstructed future frames, DiT4DiT extracts intermediate denoising features from the video generation process and uses them as temporally grounded conditions for action prediction. We further propose a dual flow-matching objective with decoupled timesteps and noise scales for video prediction, hidden-state extraction, and action inference, enabling coherent joint training of both modules. Across simulation and real-world benchmarks, DiT4DiT achieves state-of-the-art results, reaching average success rates of 98.6% on LIBERO and 50.8% on RoboCasa GR1 while using substantially less training data. On the Unitree G1 robot, it also delivers superior real-world performance and strong zero-shot generalization. Importantly, DiT4DiT improves sample efficiency by over 10x and speeds up convergence by up to 7x, demonstrating that video generation can serve as an effective scaling proxy for robot policy learning. We release code and models at https://dit4dit.github.io/.

Addressing the dual challenges of local redundancy and global dependencies in video understanding, this work innovatively adapts the Mamba to the video domain. The proposed VideoMamba overcomes the limitations of existing 3D convolution neural networks and video transformers. Its linear-complexity operator enables efficient long-term modeling, which is crucial for high-resolution long video understanding. Extensive evaluations reveal VideoMamba's four core abilities: (1) Scalability in the visual domain without extensive dataset pretraining, thanks to a novel self-distillation technique; (2) Sensitivity for recognizing short-term actions even with fine-grained motion differences; (3) Superiority in long-term video understanding, showcasing significant advancements over traditional feature-based models; and (4) Compatibility with other modalities, demonstrating robustness in multi-modal contexts. Through these distinct advantages, VideoMamba sets a new benchmark for video understanding, offering a scalable and efficient solution for comprehensive video understanding. All the code and models are available at https://github.com/OpenGVLab/VideoMamba.

People perceive the world with multiple senses (e.g., through hearing sounds, reading words and seeing objects). However, most existing AI systems only process an individual modality. This paper presents an approach that excels at handling multiple modalities of information with a single model. In our "{SkillNet}" model, different parts of the parameters are specialized for processing different modalities. Unlike traditional dense models that always activate all the model parameters, our model sparsely activates parts of the parameters whose skills are relevant to the task. Such model design enables SkillNet to learn skills in a more interpretable way. We develop our model for five modalities including text, image, sound, video and code. Results show that, SkillNet performs comparably to five modality-specific fine-tuned models. Moreover, our model supports self-supervised pretraining with the same sparsely activated way, resulting in better initialized parameters for different modalities. We find that pretraining significantly improves the performance of SkillNet on five modalities, on par with or even better than baselines with modality-specific pretraining. On the task of Chinese text-to-image retrieval, our final system achieves higher accuracy than existing leading systems including Wukong{ViT-B} and Wenlan 2.0 while using less number of activated parameters.

World models integrate raw data from various modalities, such as images and language to simulate comprehensive interactions in the world, thereby displaying crucial roles in fields like mixed reality and robotics. Yet, applying the world model for accurate video prediction is quite challenging due to the complex and dynamic intentions of the various scenes in practice. In this paper, inspired by the human rethinking process, we decompose the complex video prediction into four meta-tasks that enable the world model to handle this issue in a more fine-grained manner. Alongside these tasks, we introduce a new benchmark named Embodied Video Anticipation Benchmark (EVA-Bench) to provide a well-rounded evaluation. EVA-Bench focused on evaluating the video prediction ability of human and robot actions, presenting significant challenges for both the language model and the generation model. Targeting embodied video prediction, we propose the Embodied Video Anticipator (EVA), a unified framework aiming at video understanding and generation. EVA integrates a video generation model with a visual language model, effectively combining reasoning capabilities with high-quality generation. Moreover, to enhance the generalization of our framework, we tailor-designed a multi-stage pretraining paradigm that adaptatively ensembles LoRA to produce high-fidelity results. Extensive experiments on EVA-Bench highlight the potential of EVA to significantly improve performance in embodied scenes, paving the way for large-scale pre-trained models in real-world prediction tasks.

The visual world offers a critical axis for advancing foundation models beyond language. Despite growing interest in this direction, the design space for native multimodal models remains opaque. We provide empirical clarity through controlled, from-scratch pretraining experiments, isolating the factors that govern multimodal pretraining without interference from language pretraining. We adopt the Transfusion framework, using next-token prediction for language and diffusion for vision, to train on diverse data including text, video, image-text pairs, and even action-conditioned video. Our experiments yield four key insights: (i) Representation Autoencoder (RAE) provides an optimal unified visual representation by excelling at both visual understanding and generation; (ii) visual and language data are complementary and yield synergy for downstream capabilities; (iii) unified multimodal pretraining leads naturally to world modeling, with capabilities emerging from general training; and (iv) Mixture-of-Experts (MoE) enables efficient and effective multimodal scaling while naturally inducing modality specialization. Through IsoFLOP analysis, we compute scaling laws for both modalities and uncover a scaling asymmetry: vision is significantly more data-hungry than language. We demonstrate that the MoE architecture harmonizes this scaling asymmetry by providing the high model capacity required by language while accommodating the data-intensive nature of vision, paving the way for truly unified multimodal models.

The foundation models have recently shown excellent performance on a variety of downstream tasks in computer vision. However, most existing vision foundation models simply focus on image-level pretraining and adpation, which are limited for dynamic and complex video-level understanding tasks. To fill the gap, we present general video foundation models, InternVideo, by taking advantage of both generative and discriminative self-supervised video learning. Specifically, InternVideo efficiently explores masked video modeling and video-language contrastive learning as the pretraining objectives, and selectively coordinates video representations of these two complementary frameworks in a learnable manner to boost various video applications. Without bells and whistles, InternVideo achieves state-of-the-art performance on 39 video datasets from extensive tasks including video action recognition/detection, video-language alignment, and open-world video applications. Especially, our methods can obtain 91.1% and 77.2% top-1 accuracy on the challenging Kinetics-400 and Something-Something V2 benchmarks, respectively. All of these results effectively show the generality of our InternVideo for video understanding. The code will be released at https://github.com/OpenGVLab/InternVideo .

We introduce OmChat, a model designed to excel in handling long contexts and video understanding tasks. OmChat's new architecture standardizes how different visual inputs are processed, making it more efficient and adaptable. It uses a dynamic vision encoding process to effectively handle images of various resolutions, capturing fine details across a range of image qualities. OmChat utilizes an active progressive multimodal pretraining strategy, which gradually increases the model's capacity for long contexts and enhances its overall abilities. By selecting high-quality data during training, OmChat learns from the most relevant and informative data points. With support for a context length of up to 512K, OmChat demonstrates promising performance in tasks involving multiple images and videos, outperforming most open-source models in these benchmarks. Additionally, OmChat proposes a prompting strategy for unifying complex multimodal inputs including single image text, multi-image text and videos, and achieving competitive performance on single-image benchmarks. To further evaluate the model's capabilities, we proposed a benchmark dataset named Temporal Visual Needle in a Haystack. This dataset assesses OmChat's ability to comprehend temporal visual details within long videos. Our analysis highlights several key factors contributing to OmChat's success: support for any-aspect high image resolution, the active progressive pretraining strategy, and high-quality supervised fine-tuning datasets. This report provides a detailed overview of OmChat's capabilities and the strategies that enhance its performance in visual understanding.

In this paper, we propose ``SimVLG'', a streamlined framework for the pre-training of computationally intensive vision-language generative models, leveraging frozen pre-trained large language models (LLMs). The prevailing paradigm in vision-language pre-training (VLP) typically involves a two-stage optimization process: an initial resource-intensive phase dedicated to general-purpose vision-language representation learning, aimed at extracting and consolidating pertinent visual features, followed by a subsequent phase focusing on end-to-end alignment between visual and linguistic modalities. Our one-stage, single-loss framework circumvents the aforementioned computationally demanding first stage of training by gradually merging similar visual tokens during training. This gradual merging process effectively compacts the visual information while preserving the richness of semantic content, leading to fast convergence without sacrificing performance. Our experiments show that our approach can speed up the training of vision-language models by a factor times 5 without noticeable impact on the overall performance. Additionally, we show that our models can achieve comparable performance to current vision-language models with only 1/10 of the data. Finally, we demonstrate how our image-text models can be easily adapted to video-language generative tasks through a novel soft attentive temporal token merging modules.

Video is an increasingly prominent and information-dense medium, yet it poses substantial challenges for language models. A typical video consists of a sequence of shorter segments, or shots, that collectively form a coherent narrative. Each shot is analogous to a word in a sentence where multiple data streams of information (such as visual and auditory data) must be processed simultaneously. Comprehension of the entire video requires not only understanding the visual-audio information of each shot but also requires that the model links the ideas between each shot to generate a larger, all-encompassing story. Despite significant progress in the field, current works often overlook videos' more granular shot-by-shot semantic information. In this project, we propose a family of efficient large language vision models (LLVMs) to boost video summarization and captioning called Shotluck Holmes. By leveraging better pretraining and data collection strategies, we extend the abilities of existing small LLVMs from being able to understand a picture to being able to understand a sequence of frames. Specifically, we show that Shotluck Holmes achieves better performance than state-of-the-art results on the Shot2Story video captioning and summary task with significantly smaller and more computationally efficient models.

World models simulate future states of the world in response to different actions. They facilitate interactive content creation and provides a foundation for grounded, long-horizon reasoning. Current foundation models do not fully meet the capabilities of general world models: large language models (LLMs) are constrained by their reliance on language modality and their limited understanding of the physical world, while video models lack interactive action control over the world simulations. This paper makes a step towards building a general world model by introducing Pandora, a hybrid autoregressive-diffusion model that simulates world states by generating videos and allows real-time control with free-text actions. Pandora achieves domain generality, video consistency, and controllability through large-scale pretraining and instruction tuning. Crucially, Pandora bypasses the cost of training-from-scratch by integrating a pretrained LLM (7B) and a pretrained video model, requiring only additional lightweight finetuning. We illustrate extensive outputs by Pandora across diverse domains (indoor/outdoor, natural/urban, human/robot, 2D/3D, etc.). The results indicate great potential of building stronger general world models with larger-scale training.

The pre-trained image-text models, like CLIP, have demonstrated the strong power of vision-language representation learned from a large scale of web-collected image-text data. In light of the well-learned visual features, some existing works transfer image representation to video domain and achieve good results. However, how to utilize image-language pre-trained model (e.g., CLIP) for video-language pre-training (post-pretraining) is still under explored. In this paper, we investigate two questions: 1) what are the factors hindering post-pretraining CLIP to further improve the performance on video-language tasks? and 2) how to mitigate the impact of these factors? Through a series of comparative experiments and analyses, we find that the data scale and domain gap between language sources have great impacts. Motivated by these, we propose a Omnisource Cross-modal Learning method equipped with a Video Proxy mechanism on the basis of CLIP, namely CLIP-ViP. Extensive results show that our approach improves the performance of CLIP on video-text retrieval by a large margin. Our model also achieves SOTA results on a variety of datasets, including MSR-VTT, DiDeMo, LSMDC, and ActivityNet. We will release our code and pre-trained CLIP-ViP models at https://github.com/microsoft/XPretrain/tree/main/CLIP-ViP.

In this paper, we propose VideoLLaMA3, a more advanced multimodal foundation model for image and video understanding. The core design philosophy of VideoLLaMA3 is vision-centric. The meaning of "vision-centric" is two-fold: the vision-centric training paradigm and vision-centric framework design. The key insight of our vision-centric training paradigm is that high-quality image-text data is crucial for both image and video understanding. Instead of preparing massive video-text datasets, we focus on constructing large-scale and high-quality image-text datasets. VideoLLaMA3 has four training stages: 1) vision-centric alignment stage, which warms up the vision encoder and projector; 2) vision-language pretraining stage, which jointly tunes the vision encoder, projector, and LLM with large-scale image-text data covering multiple types (including scene images, documents, charts) as well as text-only data. 3) multi-task fine-tuning stage, which incorporates image-text SFT data for downstream tasks and video-text data to establish a foundation for video understanding. 4) video-centric fine-tuning, which further improves the model's capability in video understanding. As for the framework design, to better capture fine-grained details in images, the pretrained vision encoder is adapted to encode images of varying sizes into vision tokens with corresponding numbers, rather than a fixed number of tokens. For video inputs, we reduce the number of vision tokens according to their similarity so that the representation of videos will be more precise and compact. Benefit from vision-centric designs, VideoLLaMA3 achieves compelling performances in both image and video understanding benchmarks.

In this work, we discuss evaluating video foundation models in a fair and robust manner. Unlike language or image foundation models, many video foundation models are evaluated with differing parameters (such as sampling rate, number of frames, pretraining steps, etc.), making fair and robust comparisons challenging. Therefore, we present a carefully designed evaluation framework for measuring two core capabilities of video comprehension: appearance and motion understanding. Our findings reveal that existing video foundation models, whether text-supervised like UMT or InternVideo2, or self-supervised like V-JEPA, exhibit limitations in at least one of these capabilities. As an alternative, we introduce TWLV-I, a new video foundation model that constructs robust visual representations for both motion- and appearance-based videos. Based on the average top-1 accuracy of linear probing on five action recognition benchmarks, pretrained only on publicly accessible datasets, our model shows a 4.6%p improvement compared to V-JEPA (ViT-L) and a 7.7%p improvement compared to UMT (ViT-L). Even when compared to much larger models, our model demonstrates a 7.2%p improvement compared to DFN (ViT-H), a 2.7%p improvement compared to V-JEPA~(ViT-H) and a 2.8%p improvement compared to InternVideo2 (ViT-g). We provide embedding vectors obtained by TWLV-I from videos of several commonly used video benchmarks, along with evaluation source code that can directly utilize these embeddings. The code is available on "https://github.com/twelvelabs-io/video-embeddings-evaluation-framework".

Recent years have witnessed a big convergence of language, vision, and multi-modal pretraining. In this work, we present mPLUG-2, a new unified paradigm with modularized design for multi-modal pretraining, which can benefit from modality collaboration while addressing the problem of modality entanglement. In contrast to predominant paradigms of solely relying on sequence-to-sequence generation or encoder-based instance discrimination, mPLUG-2 introduces a multi-module composition network by sharing common universal modules for modality collaboration and disentangling different modality modules to deal with modality entanglement. It is flexible to select different modules for different understanding and generation tasks across all modalities including text, image, and video. Empirical study shows that mPLUG-2 achieves state-of-the-art or competitive results on a broad range of over 30 downstream tasks, spanning multi-modal tasks of image-text and video-text understanding and generation, and uni-modal tasks of text-only, image-only, and video-only understanding. Notably, mPLUG-2 shows new state-of-the-art results of 48.0 top-1 accuracy and 80.3 CIDEr on the challenging MSRVTT video QA and video caption tasks with a far smaller model size and data scale. It also demonstrates strong zero-shot transferability on vision-language and video-language tasks. Code and models will be released in https://github.com/alibaba/AliceMind.

We introduce VideoComp, a benchmark and learning framework for advancing video-text compositionality understanding, aimed at improving vision-language models (VLMs) in fine-grained temporal alignment. Unlike existing benchmarks focused on static image-text compositionality or isolated single-event videos, our benchmark targets alignment in continuous multi-event videos. Leveraging video-text datasets with temporally localized event captions (e.g. ActivityNet-Captions, YouCook2), we construct two compositional benchmarks, ActivityNet-Comp and YouCook2-Comp. We create challenging negative samples with subtle temporal disruptions such as reordering, action word replacement, partial captioning, and combined disruptions. These benchmarks comprehensively test models' compositional sensitivity across extended, cohesive video-text sequences. To improve model performance, we propose a hierarchical pairwise preference loss that strengthens alignment with temporally accurate pairs and gradually penalizes increasingly disrupted ones, encouraging fine-grained compositional learning. To mitigate the limited availability of densely annotated video data, we introduce a pretraining strategy that concatenates short video-caption pairs to simulate multi-event sequences. We evaluate video-text foundational models and large multimodal models (LMMs) on our benchmark, identifying both strengths and areas for improvement in compositionality. Overall, our work provides a comprehensive framework for evaluating and enhancing model capabilities in achieving fine-grained, temporally coherent video-text alignment.

In recent years, large-scale generative models for visual content (e.g., images, videos, and 3D objects/scenes) have made remarkable progress. However, training large-scale video generation models remains particularly challenging and resource-intensive due to cross-modal text-video alignment, the long sequences involved, and the complex spatiotemporal dependencies. To address these challenges, we present a training framework that optimizes four pillars: (i) data processing, (ii) model architecture, (iii) training strategy, and (iv) infrastructure for large-scale video generation models. These optimizations delivered significant efficiency gains and performance improvements across all stages of data preprocessing, video compression, parameter scaling, curriculum-based pretraining, and alignment-focused post-training. Our resulting model, MUG-V 10B, matches recent state-of-the-art video generators overall and, on e-commerce-oriented video generation tasks, surpasses leading open-source baselines in human evaluations. More importantly, we open-source the complete stack, including model weights, Megatron-Core-based large-scale training code, and inference pipelines for video generation and enhancement. To our knowledge, this is the first public release of large-scale video generation training code that exploits Megatron-Core to achieve high training efficiency and near-linear multi-node scaling, details are available in https://github.com/Shopee-MUG/MUG-V{our webpage}.

Masked video modeling (MVM) has emerged as a simple and scalable self-supervised pretraining paradigm, but only encodes motion information implicitly, limiting the encoding of temporal dynamics in the learned representations. As a result, such models struggle on motion-centric tasks that require fine-grained motion awareness. To address this, we propose TrackMAE, a simple masked video modeling paradigm that explicitly uses motion information as a reconstruction signal. In TrackMAE, we use an off-the-shelf point tracker to sparsely track points in the input videos, generating motion trajectories. Furthermore, we exploit the extracted trajectories to improve random tube masking with a motion-aware masking strategy. We enhance video representations learned in both pixel and feature semantic reconstruction spaces by providing a complementary supervision signal in the form of motion targets. We evaluate on six datasets across diverse downstream settings and find that TrackMAE consistently outperforms state-of-the-art video self-supervised learning baselines, learning more discriminative and generalizable representations. Code available at https://github.com/rvandeghen/TrackMAE

We present Emu, a Transformer-based multimodal foundation model, which can seamlessly generate images and texts in multimodal context. This omnivore model can take in any single-modality or multimodal data input indiscriminately (e.g., interleaved image, text and video) through a one-model-for-all autoregressive training process. First, visual signals are encoded into embeddings, and together with text tokens form an interleaved input sequence. Emu is then end-to-end trained with a unified objective of classifying the next text token or regressing the next visual embedding in the multimodal sequence. This versatile multimodality empowers the exploration of diverse pretraining data sources at scale, such as videos with interleaved frames and text, webpages with interleaved images and text, as well as web-scale image-text pairs and video-text pairs. Emu can serve as a generalist multimodal interface for both image-to-text and text-to-image tasks, and supports in-context image and text generation. Across a broad range of zero-shot/few-shot tasks including image captioning, visual question answering, video question answering and text-to-image generation, Emu demonstrates superb performance compared to state-of-the-art large multimodal models. Extended capabilities such as multimodal assistants via instruction tuning are also demonstrated with impressive performance.

Pretraining Vision-Language-Action (VLA) policies on internet-scale video is appealing, yet current latent-action objectives often learn the wrong thing: they remain anchored to pixel variation rather than action-relevant state transitions, making them vulnerable to appearance bias, nuisance motion, and information leakage. We introduce VLA-JEPA, a JEPA-style pretraining framework that sidesteps these pitfalls by design. The key idea is leakage-free state prediction: a target encoder produces latent representations from future frames, while the student pathway sees only the current observation -- future information is used solely as supervision targets, never as input. By predicting in latent space rather than pixel space, VLA-JEPA learns dynamics abstractions that are robust to camera motion and irrelevant background changes. This yields a simple two-stage recipe -- JEPA pretraining followed by action-head fine-tuning -- without the multi-stage complexity of prior latent-action pipelines. Experiments on LIBERO, LIBERO-Plus, SimplerEnv and real-world manipulation tasks show that VLA-JEPA achieves consistent gains in generalization and robustness over existing methods.

Video Large Language Models (Video LLMs) have achieved significant success by leveraging a two-stage paradigm: pretraining on large-scale video-text data for vision-language alignment, followed by supervised fine-tuning (SFT) for task-specific capabilities. However, existing approaches struggle with temporal reasoning due to weak temporal correspondence in the data and reliance on the next-token prediction paradigm during training. To address these limitations, we propose TEMPLE (TEMporal Preference Learning), a systematic framework that enhances Video LLMs' temporal reasoning capabilities through Direct Preference Optimization (DPO). To facilitate this, we introduce an automated preference data generation pipeline that systematically constructs preference pairs by selecting videos that are rich in temporal information, designing video-specific perturbation strategies, and finally evaluating model responses on clean and perturbed video inputs. Our temporal alignment features two key innovations: curriculum learning which that progressively increases perturbation difficulty to improve model robustness and adaptability; and "Pre-SFT Alignment'', applying preference optimization before instruction tuning to prioritize fine-grained temporal comprehension. Extensive experiments demonstrate that our approach consistently improves Video LLM performance across multiple benchmarks with a relatively small set of self-generated DPO data. We further analyze the transferability of DPO data across architectures and the role of difficulty scheduling in optimization. Our findings highlight our TEMPLE as a scalable and efficient complement to SFT-based methods, paving the way for developing reliable Video LLMs. Code is available at https://github.com/lscpku/TEMPLE.

Despite the rapid progress of instruction-based image editing, its extension to video remains underexplored, primarily due to the prohibitive cost and complexity of constructing large-scale paired video editing datasets. To address this challenge, we introduce a low-cost pretraining strategy for instruction-based video editing that leverages in-context learning from unpaired video clips. We show that pretraining a foundation video generation model with this strategy endows it with general editing capabilities, such as adding, replacing, or deleting operations, according to input editing instructions. The pretrained model can then be efficiently refined with a small amount of high-quality paired editing data. Built upon HunyuanVideoT2V, our framework first pretrains on approximately 1M real video clips to learn basic editing concepts, and subsequently fine-tunes on fewer than 150k curated editing pairs to extend more editing tasks and improve the editing quality. Comparative experiments show that our method surpasses existing instruction-based video editing approaches in both instruction alignment and visual fidelity, achieving a 12\% improvement in editing instruction following and a 15\% improvement in editing quality.

Learning new robot tasks on new platforms and in new scenes from only a handful of demonstrations remains challenging. While videos of other embodiments - humans and different robots - are abundant, differences in embodiment, camera, and environment hinder their direct use. We address the small-data problem by introducing a unifying, symbolic representation - a compact 3D "trace-space" of scene-level trajectories - that enables learning from cross-embodiment, cross-environment, and cross-task videos. We present TraceGen, a world model that predicts future motion in trace-space rather than pixel space, abstracting away appearance while retaining the geometric structure needed for manipulation. To train TraceGen at scale, we develop TraceForge, a data pipeline that transforms heterogeneous human and robot videos into consistent 3D traces, yielding a corpus of 123K videos and 1.8M observation-trace-language triplets. Pretraining on this corpus produces a transferable 3D motion prior that adapts efficiently: with just five target robot videos, TraceGen attains 80% success across four tasks while offering 50-600x faster inference than state-of-the-art video-based world models. In the more challenging case where only five uncalibrated human demonstration videos captured on a handheld phone are available, it still reaches 67.5% success on a real robot, highlighting TraceGen's ability to adapt across embodiments without relying on object detectors or heavy pixel-space generation.

Compared with visual signals, Inertial Measurement Units (IMUs) placed on human limbs can capture accurate motion signals while being robust to lighting variation and occlusion. While these characteristics are intuitively valuable to help egocentric action recognition, the potential of IMUs remains under-explored. In this work, we present a novel method for action recognition that integrates motion data from body-worn IMUs with egocentric video. Due to the scarcity of labeled multimodal data, we design an MAE-based self-supervised pretraining method, obtaining strong multi-modal representations via modeling the natural correlation between visual and motion signals. To model the complex relation of multiple IMU devices placed across the body, we exploit the collaborative dynamics in multiple IMU devices and propose to embed the relative motion features of human joints into a graph structure. Experiments show our method can achieve state-of-the-art performance on multiple public datasets. The effectiveness of our MAE-based pretraining and graph-based IMU modeling are further validated by experiments in more challenging scenarios, including partially missing IMU devices and video quality corruption, promoting more flexible usages in the real world.

Vision-language-action (VLA) models can enable broad open world generalization, but require large and diverse datasets. It is appealing to consider whether some of this data can come from human videos, which cover diverse real-world situations and are easy to obtain. However, it is difficult to train VLAs with human videos alone, and establishing a mapping between humans and robots requires manual engineering and presents a major research challenge. Drawing inspiration from advances in large language models, where the ability to learn from diverse supervision emerges with scale, we ask whether a similar phenomenon holds for VLAs that incorporate human video data. We introduce a simple co-training recipe, and find that human-to-robot transfer emerges once the VLA is pre-trained on sufficient scenes, tasks, and embodiments. Our analysis suggests that this emergent capability arises because diverse pretraining produces embodiment-agnostic representations for human and robot data. We validate these findings through a series of experiments probing human to robot skill transfer and find that with sufficiently diverse robot pre-training our method can nearly double the performance on generalization settings seen only in human data.

Echocardiography is a vital non-invasive modality for cardiac assessment, with left ventricular ejection fraction (LVEF) serving as a key indicator of heart function. Existing LVEF estimation methods depend on large-scale annotated video datasets, which are costly and limit adaptability across various clinical settings. Recent vision-language models for echocardiography, such as EchoCLIP, apply image-to-text pretraining but fail to capture crucial temporal dynamics and localized cardiac structures essential for accurate diagnosis. To address these challenges, we propose CardiacCLIP, a video-based framework that enhances LVEF prediction through attention-based frame aggregation and multi-resolution input scaling. Specifically, we introduce MFL (Multi Frame Learning), a novel attention-based mechanism for selectively fusing informative frames, and EchoZoom, a multi-scale feature extraction strategy that refines spatial representations of cardiac structures. As a novel adaptation of CLIP models for few-shot echocardiogram video analysis, our approach significantly improves diagnostic accuracy, reducing MAE by 2.07 on the EchoNet-Dynamic dataset under 1-shot setting. The code is available at https://github.com/xmed-lab/CardiacCLIP.

Can we turn a video prediction model into a robot policy? Videos, including those of humans or teleoperated robots, capture rich physical interactions. However, most of them lack labeled actions, which limits their use in robot learning. We present Video Prediction for Robot Actions (ViPRA), a simple pretraining-finetuning framework that learns continuous robot control from these actionless videos. Instead of directly predicting actions, we train a video-language model to predict both future visual observations and motion-centric latent actions, which serve as intermediate representations of scene dynamics. We train these latent actions using perceptual losses and optical flow consistency to ensure they reflect physically grounded behavior. For downstream control, we introduce a chunked flow matching decoder that maps latent actions to robot-specific continuous action sequences, using only 100 to 200 teleoperated demonstrations. This approach avoids expensive action annotation, supports generalization across embodiments, and enables smooth, high-frequency continuous control upto 22 Hz via chunked action decoding. Unlike prior latent action works that treat pretraining as autoregressive policy learning, explicitly models both what changes and how. Our method outperforms strong baselines, with a 16% gain on the SIMPLER benchmark and a 13% improvement across real world manipulation tasks. We will release models and code at https://vipra-project.github.io

While proprietary systems such as Seedance-2.0 have achieved remarkable success in omni-capable video generation, open-source alternatives significantly lag behind. Most academic models remain heavily fragmented, and the few existing efforts toward unified video generation still struggle to seamlessly integrate diverse tasks within a single framework. To bridge this gap, we propose OmniWeaving, an omni-level video generation model featuring powerful multimodal composition and reasoning-informed capabilities. By leveraging a massive-scale pretraining dataset that encompasses diverse compositional and reasoning-augmented scenarios, OmniWeaving learns to temporally bind interleaved text, multi-image, and video inputs while acting as an intelligent agent to infer complex user intentions for sophisticated video creation. Furthermore, we introduce IntelligentVBench, the first comprehensive benchmark designed to rigorously assess next-level intelligent unified video generation. Extensive experiments demonstrate that OmniWeaving achieves SoTA performance among open-source unified models. The codes and model will be made publicly available soon. Project Page: https://omniweaving.github.io.

While large-scale video diffusion models have demonstrated impressive capabilities in generating high-resolution and semantically rich content, a significant gap remains between their pretraining performance and real-world deployment requirements due to critical issues such as prompt sensitivity, temporal inconsistency, and prohibitive inference costs. To bridge this gap, we propose a comprehensive post-training framework that systematically aligns pretrained models with user intentions through four synergistic stages: we first employ Supervised Fine-Tuning (SFT) to transform the base model into a stable instruction-following policy, followed by a Reinforcement Learning from Human Feedback (RLHF) stage that utilizes a novel Group Relative Policy Optimization (GRPO) method tailored for video diffusion to enhance perceptual quality and temporal coherence; subsequently, we integrate Prompt Enhancement via a specialized language model to refine user inputs, and finally address system efficiency through Inference Optimization. Together, these components provide a systematic approach to improving visual quality, temporal coherence, and instruction following, while preserving the controllability learned during pretraining. The result is a practical blueprint for building scalable post-training pipelines that are stable, adaptable, and effective in real-world deployment. Extensive experiments demonstrate that this unified pipeline effectively mitigates common artifacts and significantly improves controllability and visual aesthetics while adhering to strict sampling cost constraints.

Video generation is rapidly evolving towards unified audio-video generation. In this paper, we present ALIVE, a generation model that adapts a pretrained Text-to-Video (T2V) model to Sora-style audio-video generation and animation. In particular, the model unlocks the Text-to-Video&Audio (T2VA) and Reference-to-Video&Audio (animation) capabilities compared to the T2V foundation models. To support the audio-visual synchronization and reference animation, we augment the popular MMDiT architecture with a joint audio-video branch which includes TA-CrossAttn for temporally-aligned cross-modal fusion and UniTemp-RoPE for precise audio-visual alignment. Meanwhile, a comprehensive data pipeline consisting of audio-video captioning, quality control, etc., is carefully designed to collect high-quality finetuning data. Additionally, we introduce a new benchmark to perform a comprehensive model test and comparison. After continue pretraining and finetuning on million-level high-quality data, ALIVE demonstrates outstanding performance, consistently outperforming open-source models and matching or surpassing state-of-the-art commercial solutions. With detailed recipes and benchmarks, we hope ALIVE helps the community develop audio-video generation models more efficiently. Official page: https://github.com/FoundationVision/Alive.

WALL-WM is a World Action Model that shifts video-action learning from chunk-centric optimization to event-grounded Vision-Language-Action pretraining, using semantically coherent action events as the atomic unit of learning. Existing WAMs commonly initialize from multimodal or video foundation models and then optimize fixed-length action chunks conditioned directly on the current observation and instruction. Although convenient, this chunk-centric formulation creates a fundamental granularity mismatch. Language describes semantic goals and events, vision evolves through continuous scene dynamics, and actions operate at control-level timescales; forcing all three into the same fixed-length prediction window turns VLA training into short-horizon correlation fitting. WALL-WM addresses this mismatch by organizing both supervision and data around semantic events. Specifically, it pairs event-grounded VLA pretraining with a data ecosystem built from event-level captions and cluster-balanced sampling, enabling scalable learning over diverse behaviors, scenes, and task structures. From the same event-pretrained backbone, WALL-WM supports two complementary inference modes. The event mode consumes next-event descriptions and enables variable-length execution chunks, while the unified mode uses a VLM with Staircase Decoding to condition conventional fixed-length chunk inference while preserving a gradient-continuous VLA path. Together with Muon-optimizer-based large-scale pretraining infrastructure, WALL-WM provides a practical scale-up recipe for general-purpose WAMs. Experiments show that WALL-WM generalizes broadly across language, scenes, and tasks, achieving state-of-the-art performance in large-scale real-world generalization evaluation.

Recent advances in video generation have been driven by diffusion models and autoregressive frameworks, yet critical challenges persist in harmonizing prompt adherence, visual quality, motion dynamics, and duration: compromises in motion dynamics to enhance temporal visual quality, constrained video duration (5-10 seconds) to prioritize resolution, and inadequate shot-aware generation stemming from general-purpose MLLMs' inability to interpret cinematic grammar, such as shot composition, actor expressions, and camera motions. These intertwined limitations hinder realistic long-form synthesis and professional film-style generation. To address these limitations, we propose SkyReels-V2, an Infinite-length Film Generative Model, that synergizes Multi-modal Large Language Model (MLLM), Multi-stage Pretraining, Reinforcement Learning, and Diffusion Forcing Framework. Firstly, we design a comprehensive structural representation of video that combines the general descriptions by the Multi-modal LLM and the detailed shot language by sub-expert models. Aided with human annotation, we then train a unified Video Captioner, named SkyCaptioner-V1, to efficiently label the video data. Secondly, we establish progressive-resolution pretraining for the fundamental video generation, followed by a four-stage post-training enhancement: Initial concept-balanced Supervised Fine-Tuning (SFT) improves baseline quality; Motion-specific Reinforcement Learning (RL) training with human-annotated and synthetic distortion data addresses dynamic artifacts; Our diffusion forcing framework with non-decreasing noise schedules enables long-video synthesis in an efficient search space; Final high-quality SFT refines visual fidelity. All the code and models are available at https://github.com/SkyworkAI/SkyReels-V2.

Recent advancements in trajectory-guided video generation have achieved notable progress. However, existing models still face challenges in generating object motions with potentially changing 6D poses under wide-range rotations, due to limited 3D understanding. To address this problem, we introduce PoseTraj, a pose-aware video dragging model for generating 3D-aligned motion from 2D trajectories. Our method adopts a novel two-stage pose-aware pretraining framework, improving 3D understanding across diverse trajectories. Specifically, we propose a large-scale synthetic dataset PoseTraj-10K, containing 10k videos of objects following rotational trajectories, and enhance the model perception of object pose changes by incorporating 3D bounding boxes as intermediate supervision signals. Following this, we fine-tune the trajectory-controlling module on real-world videos, applying an additional camera-disentanglement module to further refine motion accuracy. Experiments on various benchmark datasets demonstrate that our method not only excels in 3D pose-aligned dragging for rotational trajectories but also outperforms existing baselines in trajectory accuracy and video quality.

A major challenge in text-video and text-audio retrieval is the lack of large-scale training data. This is unlike image-captioning, where datasets are in the order of millions of samples. To close this gap we propose a new video mining pipeline which involves transferring captions from image captioning datasets to video clips with no additional manual effort. Using this pipeline, we create a new large-scale, weakly labelled audio-video captioning dataset consisting of millions of paired clips and captions. We show that training a multimodal transformed based model on this data achieves competitive performance on video retrieval and video captioning, matching or even outperforming HowTo100M pretraining with 20x fewer clips. We also show that our mined clips are suitable for text-audio pretraining, and achieve state of the art results for the task of audio retrieval.

Multimodal learning on video and text data has been receiving growing attention from many researchers in various research tasks, including text-to-video retrieval, video-to-text retrieval, and video captioning. Although many algorithms have been proposed for those challenging tasks, most of them are developed on English language datasets. Despite Indonesian being one of the most spoken languages in the world, the research progress on the multimodal video-text with Indonesian sentences is still under-explored, likely due to the absence of the public benchmark dataset. To address this issue, we construct the first public Indonesian video-text dataset by translating English sentences from the MSVD dataset to Indonesian sentences. Using our dataset, we then train neural network models which were developed for the English video-text dataset on three tasks, i.e., text-to-video retrieval, video-to-text retrieval, and video captioning. The recent neural network-based approaches to video-text tasks often utilized a feature extractor that is primarily pretrained on an English vision-language dataset. Since the availability of the pretraining resources with Indonesian sentences is relatively limited, the applicability of those approaches to our dataset is still questionable. To overcome the lack of pretraining resources, we apply cross-lingual transfer learning by utilizing the feature extractors pretrained on the English dataset, and we then fine-tune the models on our Indonesian dataset. Our experimental results show that this approach can help to improve the performance for the three tasks on all metrics. Finally, we discuss potential future works using our dataset, inspiring further research in the Indonesian multimodal video-text tasks. We believe that our dataset and our experimental results could provide valuable contributions to the community. Our dataset is available on GitHub.

Unifying multimodal understanding and generation has shown impressive capabilities in cutting-edge proprietary systems. In this work, we introduce BAGEL, an open0source foundational model that natively supports multimodal understanding and generation. BAGEL is a unified, decoder0only model pretrained on trillions of tokens curated from large0scale interleaved text, image, video, and web data. When scaled with such diverse multimodal interleaved data, BAGEL exhibits emerging capabilities in complex multimodal reasoning. As a result, it significantly outperforms open-source unified models in both multimodal generation and understanding across standard benchmarks, while exhibiting advanced multimodal reasoning abilities such as free-form image manipulation, future frame prediction, 3D manipulation, and world navigation. In the hope of facilitating further opportunities for multimodal research, we share the key findings, pretraining details, data creation protocal, and release our code and checkpoints to the community. The project page is at https://bagel-ai.org/

Despite the promise of synthesizing high-fidelity videos, Diffusion Transformers (DiTs) with 3D full attention suffer from expensive inference due to the complexity of attention computation and numerous sampling steps. For example, the popular Open-Sora-Plan model consumes more than 9 minutes for generating a single video of 29 frames. This paper addresses the inefficiency issue from two aspects: 1) Prune the 3D full attention based on the redundancy within video data; We identify a prevalent tile-style repetitive pattern in the 3D attention maps for video data, and advocate a new family of sparse 3D attention that holds a linear complexity w.r.t. the number of video frames. 2) Shorten the sampling process by adopting existing multi-step consistency distillation; We split the entire sampling trajectory into several segments and perform consistency distillation within each one to activate few-step generation capacities. We further devise a three-stage training pipeline to conjoin the low-complexity attention and few-step generation capacities. Notably, with 0.1% pretraining data, we turn the Open-Sora-Plan-1.2 model into an efficient one that is 7.4x -7.8x faster for 29 and 93 frames 720p video generation with a marginal performance trade-off in VBench. In addition, we demonstrate that our approach is amenable to distributed inference, achieving an additional 3.91x speedup when running on 4 GPUs with sequence parallelism.

Vision-language-action (VLA) models have shown great potential in building generalist robots, but still face a dilemma-misalignment of 2D image forecasting and 3D action prediction. Besides, such a vision-action entangled training manner limits model learning from large-scale, action-free web video data. To address these issues, we propose DeFI, a novel framework that Decouples visual Forward and Inverse dynamics pretraining to exploit respective data sources, wherein video generation and action prediction are disentangled. We introduce the General Forward Dynamics Model (GFDM), pretrained on diverse human and robot videos for future prediction, and the General Inverse Dynamics Model (GIDM), trained via self-supervised learning to infer latent actions from unlabeled video transitions. These models are then integrated into a unified architecture for end-to-end finetuning on downstream tasks. In this manner, GFDM and GIDM first shine separately and then cooperate for mutual benefit. Extensive experiments on CALVIN ABC-D and SimplerEnv demonstrate state-of-the-art performance, with DeFI achieving an average task length of 4.51 for CALVIN, 51.2% success rate on SimplerEnv-Fractal benchmark and 81.3% success rate in real-world deployment, significantly outperforming prior methods.

This report presents a unified technical system addressing the two core capabilities of world models for autonomous driving: world representation and world generation. For world representation, we propose WorldRec, a feed-forward reconstruction architecture driven by sparse scene queries. WorldRec initializes structured queries in 3D space, leveraging them to aggregate cross-view, cross-temporal features, thereby naturally enforcing spatial consistency across frames and yielding compact yet high-fidelity 3D Gaussian scene representations. For world generation, we propose WorldGen, a two-stage training framework of bidirectional pretraining followed by causal fine-tuning through three progressive stages (Teacher Forcing, ODE distillation, and DMD), enabling high-quality online causal video generation in as few as 4 denoising steps. Building on both modules, we further introduce the JWM, which deeply integrates WorldRec and WorldGen to achieve synergistic gains in generation stability, cross-frame consistency, and visual fidelity, providing a solid foundation for closed-loop simulation, data synthesis, and end-to-end training in autonomous driving.

Learning discriminative spatiotemporal representation is the key problem of video understanding. Recently, Vision Transformers (ViTs) have shown their power in learning long-term video dependency with self-attention. Unfortunately, they exhibit limitations in tackling local video redundancy, due to the blind global comparison among tokens. UniFormer has successfully alleviated this issue, by unifying convolution and self-attention as a relation aggregator in the transformer format. However, this model has to require a tiresome and complicated image-pretraining phrase, before being finetuned on videos. This blocks its wide usage in practice. On the contrary, open-sourced ViTs are readily available and well-pretrained with rich image supervision. Based on these observations, we propose a generic paradigm to build a powerful family of video networks, by arming the pretrained ViTs with efficient UniFormer designs. We call this family UniFormerV2, since it inherits the concise style of the UniFormer block. But it contains brand-new local and global relation aggregators, which allow for preferable accuracy-computation balance by seamlessly integrating advantages from both ViTs and UniFormer. Without any bells and whistles, our UniFormerV2 gets the state-of-the-art recognition performance on 8 popular video benchmarks, including scene-related Kinetics-400/600/700 and Moments in Time, temporal-related Something-Something V1/V2, untrimmed ActivityNet and HACS. In particular, it is the first model to achieve 90% top-1 accuracy on Kinetics-400, to our best knowledge. Code will be available at https://github.com/OpenGVLab/UniFormerV2.

We investigate the emergence of intuitive physics understanding in general-purpose deep neural network models trained to predict masked regions in natural videos. Leveraging the violation-of-expectation framework, we find that video prediction models trained to predict outcomes in a learned representation space demonstrate an understanding of various intuitive physics properties, such as object permanence and shape consistency. In contrast, video prediction in pixel space and multimodal large language models, which reason through text, achieve performance closer to chance. Our comparisons of these architectures reveal that jointly learning an abstract representation space while predicting missing parts of sensory input, akin to predictive coding, is sufficient to acquire an understanding of intuitive physics, and that even models trained on one week of unique video achieve above chance performance. This challenges the idea that core knowledge -- a set of innate systems to help understand the world -- needs to be hardwired to develop an understanding of intuitive physics.

We introduce Matrix-Game, an interactive world foundation model for controllable game world generation. Matrix-Game is trained using a two-stage pipeline that first performs large-scale unlabeled pretraining for environment understanding, followed by action-labeled training for interactive video generation. To support this, we curate Matrix-Game-MC, a comprehensive Minecraft dataset comprising over 2,700 hours of unlabeled gameplay video clips and over 1,000 hours of high-quality labeled clips with fine-grained keyboard and mouse action annotations. Our model adopts a controllable image-to-world generation paradigm, conditioned on a reference image, motion context, and user actions. With over 17 billion parameters, Matrix-Game enables precise control over character actions and camera movements, while maintaining high visual quality and temporal coherence. To evaluate performance, we develop GameWorld Score, a unified benchmark measuring visual quality, temporal quality, action controllability, and physical rule understanding for Minecraft world generation. Extensive experiments show that Matrix-Game consistently outperforms prior open-source Minecraft world models (including Oasis and MineWorld) across all metrics, with particularly strong gains in controllability and physical consistency. Double-blind human evaluations further confirm the superiority of Matrix-Game, highlighting its ability to generate perceptually realistic and precisely controllable videos across diverse game scenarios. To facilitate future research on interactive image-to-world generation, we will open-source the Matrix-Game model weights and the GameWorld Score benchmark at https://github.com/SkyworkAI/Matrix-Game.

In various video-language learning tasks, the challenge of achieving cross-modality alignment with multi-grained data persists. We propose a method to tackle this challenge from two crucial perspectives: data and modeling. Given the absence of a multi-grained video-text pretraining dataset, we introduce a Granularity EXpansion (GEX) method with Integration and Compression operations to expand the granularity of a single-grained dataset. To better model multi-grained data, we introduce an Iterative Approximation Module (IAM), which embeds multi-grained videos and texts into a unified, low-dimensional semantic space while preserving essential information for cross-modal alignment. Furthermore, GEXIA is highly scalable with no restrictions on the number of video-text granularities for alignment. We evaluate our work on three categories of video tasks across seven benchmark datasets, showcasing state-of-the-art or comparable performance. Remarkably, our model excels in tasks involving long-form video understanding, even though the pretraining dataset only contains short video clips.

Due to the limited scale and quality of video-text training corpus, most vision-language foundation models employ image-text datasets for pretraining and primarily focus on modeling visually semantic representations while disregarding temporal semantic representations and correlations. To address this issue, we propose COSA, a COncatenated SAmple pretrained vision-language foundation model. COSA jointly models visual contents and event-level temporal cues using only image-text corpora. We achieve this by sequentially concatenating multiple image-text pairs as inputs for pretraining. This transformation effectively converts existing image-text corpora into a pseudo long-form video-paragraph corpus, enabling richer scene transformations and explicit event-description correspondence. Extensive experiments demonstrate that COSA consistently improves performance across a broad range of downstream tasks, including long-form/short-form video-text tasks and image-text tasks such as retrieval, captioning, and question answering. Notably, COSA achieves state-of-the-art results on various competitive benchmarks. Code and model are released at https://github.com/TXH-mercury/COSA.

Existing text-to-video diffusion models rely solely on text-only encoders for their pretraining. This limitation stems from the absence of large-scale multimodal prompt video datasets, resulting in a lack of visual grounding and restricting their versatility and application in multimodal integration. To address this, we construct a large-scale multimodal prompt dataset by employing retrieval methods to pair in-context examples with the given text prompts and then utilize a two-stage training strategy to enable diverse video generation tasks within the same model. In the first stage, we propose a multimodal conditional video generation framework for pretraining on these augmented datasets, establishing a foundational model for grounded video generation. Secondly, we finetune the model from the first stage on three video generation tasks, incorporating multi-modal instructions. This process further refines the model's ability to handle diverse inputs and tasks, ensuring seamless integration of multi-modal information. After this two-stage train-ing process, VIMI demonstrates multimodal understanding capabilities, producing contextually rich and personalized videos grounded in the provided inputs, as shown in Figure 1. Compared to previous visual grounded video generation methods, VIMI can synthesize consistent and temporally coherent videos with large motion while retaining the semantic control. Lastly, VIMI also achieves state-of-the-art text-to-video generation results on UCF101 benchmark.

Dense video captioning, a task of localizing meaningful moments and generating relevant captions for videos, often requires a large, expensive corpus of annotated video segments paired with text. In an effort to minimize the annotation cost, we propose ZeroTA, a novel method for dense video captioning in a zero-shot manner. Our method does not require any videos or annotations for training; instead, it localizes and describes events within each input video at test time by optimizing solely on the input. This is accomplished by introducing a soft moment mask that represents a temporal segment in the video and jointly optimizing it with the prefix parameters of a language model. This joint optimization aligns a frozen language generation model (i.e., GPT-2) with a frozen vision-language contrastive model (i.e., CLIP) by maximizing the matching score between the generated text and a moment within the video. We also introduce a pairwise temporal IoU loss to let a set of soft moment masks capture multiple distinct events within the video. Our method effectively discovers diverse significant events within the video, with the resulting captions appropriately describing these events. The empirical results demonstrate that ZeroTA surpasses zero-shot baselines and even outperforms the state-of-the-art few-shot method on the widely-used benchmark ActivityNet Captions. Moreover, our method shows greater robustness compared to supervised methods when evaluated in out-of-domain scenarios. This research provides insight into the potential of aligning widely-used models, such as language generation models and vision-language models, to unlock a new capability: understanding temporal aspects of videos.

Existing benchmarks for Vision-Language Models (VLMs) primarily evaluate spatio-temporal understanding on simple single-action videos, closed attribute sets and restricted entity types, failing to capture the freeform, multi-action interactions between diverse entities which characterize real-world video understanding. Furthermore, the lack of a systematic framework for analyzing model failures across complementary spatio-temporal axes hinders comprehensive evaluation. To address these gaps, we introduce VISTA, a Video Interaction Spatio-Temporal Analysis benchmark designed for open-set, multi-entity and multi-action spatio-temporal understanding in VLMs. VISTA decomposes videos into interpretable entities, their associated actions, and relational dynamics, enabling multi-axis diagnostics and unified assessment of relational, spatial, and temporal understanding. Our benchmark integrates multiple datasets into a single interaction-aware taxonomy and comprises ~12K curated video-query pairs spanning diverse scenes and complexities. We systematically evaluate 11 state-of-the-art VLMs on VISTA, and break down aggregate performance across our taxonomy to reveal shortcomings and pronounced spatio-temporal biases obscured by traditional metrics. By providing detailed, taxonomy-driven diagnostics on a challenging dataset, VISTA offers a nuanced framework to guide advances in model design, pretraining strategies, and evaluation protocols. Overall, VISTA is the first, large-scale, interaction-aware diagnostic benchmark for spatio-temporal understanding in VLMs.

This paper presents JavisGPT, the first unified multimodal large language model (MLLM) for Joint Audio-Video (JAV) comprehension and generation. JavisGPT adopts a concise encoder-LLM-decoder architecture, featuring a SyncFusion module for spatio-temporal audio-video fusion and synchrony-aware learnable queries to bridge a pretrained JAV-DiT generator. This design enables temporally coherent video-audio understanding and generation from multimodal instructions. We design an effective three-stage training pipeline consisting of multimodal pretraining, audio-video fine-tuning, and large-scale instruction-tuning, to progressively build multimodal comprehension and generation from existing vision-language models. To support this, we further construct JavisInst-Omni, a high-quality instruction dataset with over 200K GPT-4o-curated audio-video-text dialogues that span diverse and multi-level comprehension and generation scenarios. Extensive experiments on JAV comprehension and generation benchmarks show that JavisGPT outperforms existing MLLMs, particularly in complex and temporally synchronized settings.

Understanding multimodal signals in egocentric vision, such as RGB video, depth, camera poses, and gaze, is essential for applications in augmented reality, robotics, and human-computer interaction, enabling systems to better interpret the camera wearer's actions, intentions, and surrounding environment. However, building large-scale egocentric multimodal and multitask models presents unique challenges. Egocentric data are inherently heterogeneous, with large variations in modality coverage across devices and settings. Generating pseudo-labels for missing modalities, such as gaze or head-mounted camera trajectories, is often infeasible, making standard supervised learning approaches difficult to scale. Furthermore, dynamic camera motion and the complex temporal and spatial structure of first-person video pose additional challenges for the direct application of existing multimodal foundation models. To address these challenges, we introduce a set of efficient temporal tokenizers and propose EgoM2P, a masked modeling framework that learns from temporally-aware multimodal tokens to train a large, general-purpose model for egocentric 4D understanding. This unified design supports multitasking across diverse egocentric perception and synthesis tasks, including gaze prediction, egocentric camera tracking, and monocular depth estimation from egocentric video, and also serves as a generative model for conditional egocentric video synthesis. Across these tasks, EgoM2P matches or outperforms specialist models while being an order of magnitude faster. We will fully open-source EgoM2P to support the community and advance egocentric vision research. Project page: https://egom2p.github.io/.

Recently, image-based Large Multimodal Models (LMMs) have made significant progress in video question-answering (VideoQA) using a frame-wise approach by leveraging large-scale pretraining in a zero-shot manner. Nevertheless, these models need to be capable of finding relevant information, extracting it, and answering the question simultaneously. Currently, existing methods perform all of these steps in a single pass without being able to adapt if insufficient or incorrect information is collected. To overcome this, we introduce a modular multi-LMM agent framework based on several agents with different roles, instructed by a Planner agent that updates its instructions using shared feedback from the other agents. Specifically, we propose TraveLER, a method that can create a plan to "Traverse" through the video, ask questions about individual frames to "Locate" and store key information, and then "Evaluate" if there is enough information to answer the question. Finally, if there is not enough information, our method is able to "Replan" based on its collected knowledge. Through extensive experiments, we find that the proposed TraveLER approach improves performance on several VideoQA benchmarks without the need to fine-tune on specific datasets. Our code is available at https://github.com/traveler-framework/TraveLER.

Training computer-use agents requires massive amounts of GUI interaction data, but manually annotating action trajectories at scale is prohibitively expensive. We present VideoAgentTrek, a scalable pipeline that automatically mines training data from publicly available screen-recorded videos at web scale, eliminating the need for manual annotation. Our approach addresses a key challenge: raw videos contain implicit demonstrations but lack explicit action labels. To solve this, we develop Video2Action, an inverse dynamics module (IDM) with two components: (1) a video grounding model that detects and localizes GUI actions with precise temporal boundaries and context, and (2) an action-content recognizer that extracts structured parameters like click coordinates and typed text with high fidelity. Applied to 39,000 YouTube tutorial videos, our pipeline generates 1.52 million interaction steps automatically. We leverage this data through continued pretraining followed by supervised fine-tuning. On OSWorld-Verified, our approach improves task success rates from 9.3% (SFT-only baseline) to 15.8%, a 70% relative improvement. On AgentNetBench, step accuracy increases from 64.1% to 69.3%. Our results demonstrate that passive internet videos can be transformed into high-quality supervision for computer-use agents, providing a scalable alternative to expensive manual annotation.

Soccer understanding has recently garnered growing research interest due to its domain-specific complexity and unique challenges. Unlike prior works that typically rely on isolated, task-specific expert models, this work aims to propose a unified model to handle diverse soccer visual understanding tasks, ranging from fine-grained perception (e.g., athlete detection) to semantic reasoning (e.g., event classification). Specifically, our contributions are threefold: (i) we present SoccerMaster, the first soccer-specific vision foundation model that unifies diverse understanding tasks within a single framework via supervised multi-task pretraining; (ii) we develop an automated data curation pipeline to generate scalable spatial annotations, and integrate them with various existing soccer video datasets to construct SoccerFactory, a comprehensive pretraining data resource; and (iii) we conduct extensive evaluations demonstrating that SoccerMaster consistently outperforms task-specific expert models across diverse downstream tasks, highlighting its breadth and superiority. The data, code, and model will be publicly available.

Video diffusion models (DMs) have enabled high-quality video synthesis. However, their computation costs scale quadratically with sequence length because self-attention has quadratic complexity. While linear attention lowers the cost, fully replacing quadratic attention requires expensive pretraining due to the limited expressiveness of linear attention and the complexity of spatiotemporal modeling in video generation. In this paper, we present LinVideo, an efficient data-free post-training framework that replaces a target number of self-attention modules with linear attention while preserving the original model's performance. First, we observe a significant disparity in the replaceability of different layers. Instead of manual or heuristic choices, we frame layer selection as a binary classification problem and propose selective transfer, which automatically and progressively converts layers to linear attention with minimal performance impact. Additionally, to overcome the ineffectiveness and inefficiency of existing objectives for this transfer process, we introduce an anytime distribution matching (ADM) objective that aligns the distributions of samples across any timestep along the sampling trajectory. This objective is efficient and recovers model performance. Extensive experiments show that our method achieves a 1.25-2.00x speedup while preserving generation quality, and our 4-step distilled model further delivers a 15.92x latency reduction with minimal visual quality drop.

To perform automatic family audio analysis, past studies have collected recordings using phone, video, or audio-only recording devices like LENA, investigated supervised learning methods, and used or fine-tuned general-purpose embeddings learned from large pretrained models. In this study, we advance the audio component of a new infant wearable multi-modal device called LittleBeats (LB) by learning family audio representation via wav2vec 2.0 (W2V2) pertaining. We show given a limited number of labeled LB home recordings, W2V2 pretrained using 1k-hour of unlabeled home recordings outperforms oracle W2V2 pretrained on 52k-hour unlabeled audio in terms of parent/infant speaker diarization (SD) and vocalization classifications (VC) at home. Extra relevant external unlabeled and labeled data further benefit W2V2 pretraining and fine-tuning. With SpecAug and environmental speech corruptions, we obtain 12% relative gain on SD and moderate boost on VC. Code and model weights are available.

We present Pix2Gif, a motion-guided diffusion model for image-to-GIF (video) generation. We tackle this problem differently by formulating the task as an image translation problem steered by text and motion magnitude prompts, as shown in teaser fig. To ensure that the model adheres to motion guidance, we propose a new motion-guided warping module to spatially transform the features of the source image conditioned on the two types of prompts. Furthermore, we introduce a perceptual loss to ensure the transformed feature map remains within the same space as the target image, ensuring content consistency and coherence. In preparation for the model training, we meticulously curated data by extracting coherent image frames from the TGIF video-caption dataset, which provides rich information about the temporal changes of subjects. After pretraining, we apply our model in a zero-shot manner to a number of video datasets. Extensive qualitative and quantitative experiments demonstrate the effectiveness of our model -- it not only captures the semantic prompt from text but also the spatial ones from motion guidance. We train all our models using a single node of 16xV100 GPUs. Code, dataset and models are made public at: https://hiteshk03.github.io/Pix2Gif/.

In this paper, we propose CharacterShot, a controllable and consistent 4D character animation framework that enables any individual designer to create dynamic 3D characters (i.e., 4D character animation) from a single reference character image and a 2D pose sequence. We begin by pretraining a powerful 2D character animation model based on a cutting-edge DiT-based image-to-video model, which allows for any 2D pose sequnce as controllable signal. We then lift the animation model from 2D to 3D through introducing dual-attention module together with camera prior to generate multi-view videos with spatial-temporal and spatial-view consistency. Finally, we employ a novel neighbor-constrained 4D gaussian splatting optimization on these multi-view videos, resulting in continuous and stable 4D character representations. Moreover, to improve character-centric performance, we construct a large-scale dataset Character4D, containing 13,115 unique characters with diverse appearances and motions, rendered from multiple viewpoints. Extensive experiments on our newly constructed benchmark, CharacterBench, demonstrate that our approach outperforms current state-of-the-art methods. Code, models, and datasets will be publicly available at https://github.com/Jeoyal/CharacterShot.

Traditional visual object tracking (VOT) methods typically rely on task-specific supervised training, limiting their generalization to unseen objects and challenging scenarios with distractors, occlusion, and nonlinear motion. Recent vision foundation models, exemplified by SAM 2, learn strong video understanding priors from large-scale pretraining and offer a promising foundation for building more robust and generalizable trackers. However, directly applying SAM 2 to VOT remains suboptimal, as it does not explicitly model target motion dynamics or enforce geometric and semantic consistency across frames, both of which are essential for reliable tracking. To address this issue, we propose SAMOSA, a new tracking framework that adapts SAM 2 to complex VOT scenarios by explicitly leveraging motion, geometry, and semantic cues. Specifically, we introduce a lightweight nonlinear motion predictor to model target dynamics and guide mask selection as well as memory filtering. We further exploit semantic cues to detect target shifts and recover from tracking failures, while geometric cues are incorporated as structural constraints to improve tracking stability. In this way, SAMOSA bridges the gap between the implicit video understanding prior of SAM 2 and explicit tracking-oriented modeling. Extensive experiments show that SAMOSA consistently outperforms state-of-the-art SAM 2--based approaches on general benchmarks, demonstrates stronger generalization than supervised VOT methods, and achieves substantial gains on anti-UAV datasets, which typify complex nonlinear motion scenarios. Our code is available at https://github.com/DurYi/SAMOSA.