视频大模型

Video world models are moving toward preserving an observed world under controllable camera and object motion while allowing its environmental state to change. Yet these controls remain isolated, and weather generation typically relies on a source video or reconstructed scene that already specifies future structure. We study a first-frame-anchored source-to-state setting, where the model starts from a single image and follows explicit camera and object controls and an optional weather instruction, then generates a video that either preserves the source world or transfers it to a target weather state. To address these challenges, we first build HoloStateData, a state video dataset that turns diverse videos into unified control samples for camera, object, and weather supervision. Second, we introduce Holo-World, a unified controllable video world model that jointly controls scene from a single image. Its Unified Scene Adapter factorizes world preservation and weather transfer into distinct parameter subspaces, using rendered background, geometry buffers, and object controls to maintain controlled scene structure while modeling weather-dependent appearance and particle effects. Additionally, Scene-Weather Decomposed CFG guides scene and weather residuals separately, strengthening target weather effects without over-amplifying the full condition. Quantitative and qualitative experiments demonstrate that Holo-World maintains precise camera and object control with consistent scene structure while transferring scenes into diverse target weather state, outperforming video-to-video weather editing baselines on weather-state generation. Our project page is available at \url{https://xiangchenyin.github.io/Holo-World/}.

Sketching is inherently sequential: strokes are drawn progressively to explore and refine ideas. Yet most generative approaches treat sketches as static images, ignoring the temporal process underlying creative exploration. Modeling this sequential structure remains challenging: prior methods either rely on large-scale human-drawn datasets with limited diversity, or use large language models (LLMs) to produce drawing instructions, often at the cost of visual fidelity. We present VideoSketcher, a method for generating high-quality sketching processes by adapting pretrained text-to-video diffusion models to the sparse, continuous nature of sketch formation. Our key insight is that LLMs and video diffusion models offer complementary strengths: LLMs act as semantic planners that decompose concepts into step-by-step instructions, while video diffusion models serve as powerful "renderers" that translate them into temporally coherent sketch sequences. We introduce a two-stage fine-tuning strategy that decouples temporal structure from visual appearance: stroke ordering is learned from synthetic shape compositions, while style is distilled from as few as seven hand-drawn examples. Despite minimal supervision, our method can generate diverse, high-quality sequential sketches that faithfully follow specified drawing orders. Our framework naturally extends to brush style control and autoregressive generation, supporting artistic applications.

Evaluating video generation with clean, pixel-based reward models disconnects evaluation from the noisy diffusion process and incurs massive VAE decoding costs. In this paper, we challenge this paradigm by asking a fundamental question: Can a powerful video generator inherently discriminate preferences directly from noisy latents? To answer this, we introduce \textbf{PRISM} (\textbf{P}reference \textbf{R}epresentation in \textbf{I}ntermediate \textbf{S}tates of Diffusion \textbf{M}odels). PRISM employs a lightweight Query-based Aggregation head with a frozen video diffusion backbone to decode preference signals from noisy latents. Surprisingly, PRISM not only achieves SOTA preference accuracy but also unlocks strong noise-robustness, which enables early-stage Best-of-$N$ sampling. This allows for filtering suboptimal candidates at the very beginning of denoising, drastically reducing computation while boosting video quality. We also reveal a strong positive correlation between a backbone's generative performance and its inherent evaluative power, enabling self-improving video backbones.

Cinematic compositing aims to integrate green-screen characters into novel environments while maintaining physical and photometric realism. Previous methods often fail to capture the complex bidirectional interactions between characters and their surroundings, which we characterize as Character-to-Environment (C2E) physical interaction and Environment-to-Character (E2C) lighting harmonization. To address this, we propose an end-to-end video diffusion framework that jointly models C2E and E2C interactions, specifically handling the challenges of interactive props. Our approach introduces a tri-mask-guided architecture with RGB-D joint denoising to ensure physically consistent interactions among the character, props, and environment. We further develop an efficient prior-driven data curation pipeline to construct high-quality relighting pairs without expensive rendering. Finally, a reference-conditioned mechanism enables controllable environment synthesis and precise prop replacement. Extensive experiments demonstrate that our framework significantly outperforms existing methods in cinematic-quality dynamic video compositing.

Traditional animation production relies heavily on manual drawing and iterative refinement, particularly for key-pose design, in-betweening, and character coloring. While existing animation and video generation methods have made notable progress, they typically depend on RGB boundary frames, dense frame-wise conditions, or complete sketch sequences, limiting their applicability under low-cost input conditions. We present SketchKeyAnime, a video diffusion framework for generating structurally controllable, appearance-consistent, and temporally coherent animations from sparse key-sketch inputs. Given a single reference RGB image and a few temporally indexed key sketches, SketchKeyAnime introduces a dual-branch conditioning mechanism to encode local geometric constraints alongside semantic-temporal context. It leverages Sketch Cross Attention to fuse reference image and sketch conditions with learnable gating, and incorporates an Adaptive Weighted Loss to strengthen supervision on key-sketch frames and line-art regions. Experimental results on the Aesthetic subset of Sakuga-42M show that our approach consistently outperforms representative animation interpolation and sketch-guided generation baselines. Compared to the best-performing baseline, SketchKeyAnime reduces EDMD by 31.9\% and FVD by 9.5\%, demonstrating superior sketch fidelity and temporal coherence, while achieving the best overall performance across most quantitative metrics. These results validate the proposed framework and highlight its potential for low-cost, highly controllable animation creation.

Diffusion models have achieved remarkable success in image and video generation and editing. While recent studies have extended these efforts toward motion editing, simultaneously transforming both motion and location-despite its practical importance-remains largely unexplored. To better understand robust motion-location editing, we first analyze the fundamental factors that degrade its quality. Based on this analysis, we propose TeleMorpher, one of the first one-shot frameworks to the best of our knowledge, for simultaneous motion-location editing. Our approach leverages motion priors, a target motion-centric video generated from an off-the-shelf model as motion-editing guidance, and the ground truth motion to enable more controllable and precise motion-location editing. Via this, our framework works as follows: (1) we first disentangle the protagonist and the background via pre-trained segmentation and inpainting models. (2) Then, we introduce a training-free pose warping that edits the protagonist's motion with the motion prior as the guidance. (3) The result of warped motion video is directly injected into a baseline motion editor during inference, mitigating the difference between source and target motions while preserving the appearance of the source video. (4) To enhance the reliability of quantitative evaluations, we propose two new LPIPS-based metrics that measure the background consistency before and after the motion editing and the fidelity of motion editing performance via measuring the difference between the extracted protagonist's skeletons from source and target videos. Experiments with in-the-wild videos and the TaiChi dataset demonstrate that TeleMorpher achieves superior performance across both quantitative and qualitative measurements (real-human evaluation), underscoring its effectiveness.

Precise 3D spatial orchestration in text-to-video generation remains a significant challenge, particularly for multi-object scenes where semantic layout and temporal dynamics are often entangled. While existing depth-conditioned models achieve good structural fidelity, they necessitate dense, frame-accurate guidance that is labor-intensive to author for dynamic events involving deformable objects. We present LooseControlVideo, a framework that enables intuitive and expressive control by using sparse, oriented 3D boxes as a "blocking" proxy. This allows users to author high-level layout and trajectory while leveraging a video generative model to generate realistic occlusions, dynamics and interactions. We achieve this by fine-tuning a Wan 2.2 backbone on a video dataset annotated with DNOCS, a novel encoding for 3D size, orientation and depth-ordered occlusions. Furthermore, our method allows for localized refinement, such as adjusting a jump trajectory or adding an interaction, with minimal disruption to the global scene context. Extensive evaluations on the nuScenes, HO-3D, and BEHAVE benchmarks demonstrate that LooseControlVideo significantly outperforms existing 2D-box and flow-based baselines. Our findings indicate a 1.2x to 3x improvement in Trajectory Error; 2x improvement in Rigid Motion Consistency; and a 1.5x to 2x increase in Occlusion Accuracy over current state-of-the-art layout-conditioned models, demonstrating that oriented 3D primitives provide good geometric prior for complex, multi-agent video authoring.

Interactive video world models generate video chunk by chunk in response to user-controlled camera movements, enabling applications such as real-time game simulation, virtual scene navigation, and embodied AI training. However, scaling to long interactive trajectories is prohibitively expensive due to growing context memory, quadratic attention complexity, and repeated denoising steps. We present Light Interaction, a training-free inference acceleration framework for interactive video world models. Our key insight is that interaction naturally enables trajectory-dependent adaptive computation: retrieved spatial memory can be discarded during novel exploration, temporal context can be adjusted according to local latent dynamics, and early-step model outputs can be reused when the camera revisits familiar regions. Based on this insight, Light Interaction combines adaptive context management, denoising cache acceleration, and hardware-software co-designed 3D block sparse attention with fused Triton kernels. Evaluated on HY-WorldPlay and Matrix-Game-3.0, Light Interaction achieves up to 2.59x speedup without model retraining while maintaining competitive visual quality.

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

Streaming VideoLLMs must continuously process incoming video while maintaining low query latency, making both video-ingestion throughput and query-time responsiveness critical for real-time deployment. Existing methods largely focus on accelerating individual modules, such as visual encoding, token pruning, or KV-cache compression, but provide limited insight into whether the resulting system can sustain real-time streaming performance. We formulate streaming VideoLLM inference as a coordinated pipeline spanning visual preprocessing, visual encoding, token dropping, and LLM prefilling/decoding. Building on this formulation, we propose ViCoStream (Video Coordinated Streaming), a stage-wise coordinated streaming framework that combines chunk-wise execution, CUDA-stream overlap, visual token control, bounded visual attention, and query-side retrieval to bound per-chunk computation and memory costs. We further provide a systematic study of bottleneck migration, revealing how chunk size, token retention, attention locality, and retrieval scope shape the throughput-accuracy trade-off. Experiments with Qwen2.5-VL-3B/7B-Instruct across multiple streaming benchmarks show that ViCoStream achieves 134 FPS video throughput and less than 50 ms TTFT on a single A100 GPU while maintaining accuracy close to full-history baselines.

Recent progress in vision-language models has enabled the processing of increasingly long video sequences, but the ability to handle extended token streams does not translate to understanding of narrative structure in long videos. Existing long video benchmarks focus on needle-in-a-haystack retrieval rather than evaluating how low-level actions form events, how events interact across time, and how narratives progress, for example, whether a model can connect an early setback, such as a job loss to a later relationship breakup, despite long gaps, intervening scenes, or flashbacks that reframe what occurred. We introduce NEST (Narrative Event Structures in Time for Long Video Understanding), a dataset of 1005 full-length movies (avg. 98 minutes), each annotated with 102 multimodal narrative events grounded in visual content, dialogue, and audio. NEST captures multimodal narrative events with structured annotations grounded in visual content, dialogue, and audio, and links them through relations that reflect narrative structure, including temporal ordering, hierarchical composition, and long-range dependencies. We introduce baselines for event trigger detection (ETD), event localization (EL), event argument extraction (EAE), and event relation extraction (ERE). The benchmark is highly challenging for grounded event discovery, with ETD below 8%, EL under 6%, and EAE below 11%. In contrast, ERE is more tractable once events are given, reaching 35.45% F1 zero-shot and 44.42% F1 after fine-tuning.

Learning everyday skills, like cooking a dish, relies increasingly on instructional media such as online videos. This opens the door to the use of video (and multimodal) large language models (LLMs) as task guidance assistants. A crucial capability for the real-world success of a prospective task guidance assistant is it's ability to intervene proactively as soon as a mistake is apparent in order to guide the user. To evaluate this crucial capability, we introduce Ego-MC-Bench (Mistake Corrections), a benchmark for evaluating reactive, step-by-step task guidance in realistic cooking scenarios. Extensive experiments show that Ego-MC-Bench is highly challenging for state-of-the-art video LLMs. We argue that a key reason is the limited availability of training data for fine-tuning models on this task. Although there exists a wide range of cooking video datasets, existing datasets lack examples of mistakes along with appropriately timed interventions. To help address this data limitation, we also introduce Ego-CoMist, a counterfactual synthetic dataset created by transforming non -interactive cooking videos into supervised training examples showing proactive interventions. We show that fine-tuning on Ego-CoMist yields performance gains especially for smaller and more efficient video LLMs that are well suited for delivering assistance on edge devices.

Long Video Question Answering (LVQA) requires identifying sparse, query-relevant evidence within hours-long untrimmed videos. Existing approaches either process videos densely with large vision-language models (VLMs), incurring prohibitive computational cost, or rely on sparse caption-based reasoning, which often misses temporally localized and motion-centric evidence. We introduce TimeProVe, a cost-efficient hybrid framework for temporally grounded reasoning in long videos. TimeProVe first employs lightweight modules to generate action-grounded answer--evidence hypotheses and subsequently invokes an expensive VLM only for targeted verification. The core of our framework lies in the Action-based Candidate Evidence (ACE) module, which converts temporally localized actions into query-conditioned candidate answers and supporting evidence windows through lightweight LLM reasoning. We further introduce OpenTSUBench (OTB), an open-ended benchmark designed to evaluate temporally grounded reasoning in real-world Activities of Daily Living (ADL) scenarios. Experiments show that TimeProVe outperforms the strongest baseline on OTB by 7.3%, while reducing VLM calls by 75% and inference cost by 93%. Furthermore, without explicit temporal grounding training, TimeProVe achieves competitive performance on Charades-STA, and reaches state-of-the-art results when enhanced with grounding VLMs.

This paper presents Vortex, the multimodal video retrieval system developed by our team, FocusOnFun, for the Ho Chi Minh City AI Challenge 2025, designed to advance intelligent multimedia search and temporal reasoning. The system integrates adaptive keyframe extraction, multimodal metadata generation from vision-language and speech models, and a hybrid retrieval strategy that fuses CLIP and SigLIP2 embeddings through Reciprocal Rank Fusion to balance global and fine-grained semantics. To enhance interactivity, Vortex incorporates Rocchio-based relevance feedback and a multi-stage temporal search mechanism for sequential event alignment. Built on Milvus and Elasticsearch, the architecture enables scalable indexing and efficient retrieval. Evaluated in the official competition, our FocusOnFun team's system achieved a score of 79.6/88 (90.5\%) in the Preliminary Round and was further evaluated in the Final Round, achieving an `Excellent' overall performance with `Outstanding' results in the question-answering (QA) task. This demonstrating the complementary strengths of CLIP and SigLIP2 and confirming the effectiveness of the hybrid retrieval approach. The system establishes a robust foundation for future research in intelligent, context-aware, and interactive video retrieval.

Egocentric video understanding is inherently limited by the narrow perspective of wearable cameras: a single viewpoint, a single modality, a single model cannot capture the full richness of human action. We argue that a truly expressive egocentric representation must subsume complementary knowledge across viewpoints, modalities, and foundation model representations, yet remain deployable from egocentric video alone. To this end, we introduce a hierarchical multi-teacher distillation framework that produces UNIEGO, a unified egocentric encoder trained with nine teachers spanning ego-exo viewpoints, RGB, depth, and skeleton modalities, and four foundation models. Rather than distilling directly from heterogeneous teachers whose incompatible architectures and feature geometries induce conflicting gradients, our framework interposes a layer of representation-specific Proxy models that translate diverse teacher knowledge into a homogeneous egocentric space. A second distillation stage, Selective Proxy Distillation (SPD), then adaptively selects, for each training sample, the subset of proxies that are both correct and confident, distilling exclusively from reliable supervision and suppressing erroneous signals. SPD is further stabilized by initializing UNIEGO as a learned convex combination of proxy parameters, placing the unified model in a well-conditioned region of the loss landscape before distillation begins. UNIEGO achieves state-of-the-art performance across three egocentric video understanding tasks - action recognition, video retrieval, and action segmentation on three challenging ego-exo benchmarks, outperforming naive multi-teacher distillation baselines and demonstrating that structured, proxy-mediated knowledge transfer yields richer and more discriminative egocentric representations.

World models are increasingly regarded as a decisive step toward artificial general intelligence, yet modeling the physical world demands more than rendering convincing frames on demand: it requires an internal world state that keeps evolving over time, decoupled from observation, so that objects endure and events run to their conclusions whether or not a camera is watching, much as the moon holds to its orbit when no one is looking. This requirement is a blind spot of existing benchmarks, which reward surface properties such as fidelity, motion, and camera controllability while never asking whether a generated world keeps evolving once it is unobserved. We introduce \textbf{WRBench}, the first systematic diagnostic benchmark that treats camera motion as an intervention on observability and resolves evaluation into a human-calibrated chain that asks whether the camera executes the requested interaction, whether the scene stays continuous and identifiable while in view, and whether a returning target remains consistent with the event that was set in motion. Across 9{,}600 videos from 23 models spanning four control paradigms, one finding proves stubborn: current systems maintain the observed world as a tracking shot, resuming a returning target in the state at which it was abandoned rather than advancing the event while it went unseen. Because this failure recurs across control paradigms, model families, and increments of scale, robust world-state evolution does not follow from cleaner imagery, tighter control, richer geometric priors, or sheer parameter count We therefore argue that the stability of the physical state kernel and the consistency of worldlines under viewpoint intervention should become first-class objectives of world-model design, so that a world model captures how the world will unfold rather than how the next frame appears.