Future Forcing: Future-aware Training-free KV Cache Policy for Autoregressive Video Generation

Autoregressive (AR) video generation has emerged as a promising paradigm for long-horizon video synthesis, where each frame is generated conditioned on previously generated tokens. To accelerate inference, the KV cache is used to avoid redundant recomputation across generation steps. Nevertheless, its growth with generation length introduces increasing memory and error accumulation, limiting the scalability of AR models to even longer sequences. Existing KV cache compression methods mitigate this issue by selectively retaining only video tokens deemed important. However, most existing methods assess token importance using short-horizon signals derived from the current or historical generation context, making these methods prone to overlooking tokens that appear unimportant at early steps but later become critical for future frames. In this work, we identify an important property of trained AR video models: although RoPE-modulated queries evolve across autoregressive steps, the underlying canonical pre-RoPE query distribution remains remarkably stable throughout the video generation process. This approximate stationarity implies that future query distributions are estimable from historical statistics, enabling principled future-aware cache decisions without any additional training. Building on this insight, we propose Future Forcing, a training-free future-aware KV cache policy for AR video generation. Specifically, Future Forcing first constructs a future query proxy from historical statistics, then scores KV cache tokens by their importance under this proxy, and finally merges redundant token pairs within the affine subspace induced by the future query. Extensive experiments show that Future Forcing improves long-horizon consistency under limited KV caches, achieving up to 1.49 improvement in subject consistency on VBench-Long for 60s generation over existing AR video KV cache policies.

GTA: Advancing Image-to-3D World Generation via Geometry Then Appearance Video Diffusion

Recent developments in generative models and large-scale datasets have substantially advanced 3D world generation, facilitating a broad range of domains including spatial intelligence, embodied intelligence, and autonomous driving. While achieving remarkable progress, existing approaches to 3D world generation typically prioritize appearance prediction with limited modeling of the underlying geometry, leading to issues such as unreliable scene structure estimation and degraded cross-view consistency. To address these limitations, motivated by the coarse-to-fine nature of human visual perception, we propose GTA, a novel image-to-3D world generation method following a Geometry-Then-Appearance paradigm. Specifically, given a single input image, to improve the structural fidelity of synthesized 3D scenes, GTA adopts a two-stage framework with two dedicated video diffusion models, which first generate coarse geometric structure from novel viewpoints and then synthesize fine-grained appearance conditioned on the predicted geometry. To further enhance cross-view appearance consistency, we introduce a random latent shuffle strategy during the training process, along with a test-time scaling scheme that improves perceptual quality without compromising quantitative performance. Extensive experiments have demonstrated that our proposed method consistently outperforms existing approaches in terms of fidelity, visual quality, and geometric accuracy. Moreover, GTA is shown to be effective as a general enhancement module that further improves the generation quality of existing image-to-3D world pipelines, as well as supporting multiple downstream applications and exhibiting favorable data efficiency during model training, highlighting its versatility and broad applicability. Project page: https://hanxinzhu-lab.github.io/GTA/.

Embody4D: A Generalist 4D World Model for Embodied AI

World models have made significant progress in modeling dynamic environments; however, most embodied world models are still restricted to 2D representations, lacking the comprehensive multi-view information essential for embodied spatial reasoning. Bridging this gap is non-trivial, primarily due to challenges from severe scarcity of paired multi-view data, the difficulty of maintaining spatiotemporal consistency in generated 3D geometries, and the tendency to hallucinate manipulation details. To address these challenges, we propose Embody4D, a dedicated video-to-video world model for embodied scenarios, capable of synthesizing arbitrary novel views from a monocular video. First, to tackle data scarcity, we introduce a 3D-aware compositional synthesis pipeline to curate a heterogeneous dataset compositing cross-embodiment robotic arms with diverse backgrounds, guaranteeing broad generalization. Second, to enforce geometric stability, we devise an adaptive noise injection strategy; by leveraging confidence disparities across image regions, this method selectively regularizes the diffusion process to ensure strict spatiotemporal consistency. Finally, to guarantee manipulation fidelity, we incorporate an interaction-aware attention mechanism that explicitly attends to the robotic interaction regions. Extensive experiments demonstrate that Embody4D achieves state-of-the-art performance, serving as a robust world model that synthesizes high-fidelity, view-consistent videos to empower downstream robotic planning and learning.

Training-Free Sparse Attention for Fast Video Generation via Offline Layer-Wise Sparsity Profiling and Online Bidirectional Co-Clustering

Diffusion Transformers (DiTs) achieve strong video generation quality but suffer from high inference cost due to dense 3D attention, leading to the development of sparse attention technologies to improve efficiency. However, existing training-free sparse attention methods in video generation still face two unresolved limitations: ignoring layer heterogeneity in attention pruning and ignoring query-key coupling in block partitioning, which hinder a better quality-speedup trade-off. In this work, we uncover a critical insight that the attention sparsity of each layer is its intrinsic property, with minor effects across different inputs. Motivated by this, we propose SVOO, a training-free Sparse attention framework for fast Video generation via Offline layer-wise sparsity profiling and Online bidirectional co-clustering. Specifically, SVOO adopts a two-stage paradigm: (i) offline layer-wise sensitivity profiling to derive intrinsic per-layer pruning levels, and (ii) online block-wise sparse attention via a novel bidirectional co-clustering algorithm. Extensive experiments on seven widely used video generation models demonstrate that SVOO achieves a superior quality-speedup trade-off over state-of-the-art methods, delivering up to $1.93\times$ speedup while maintaining a PSNR of up to 29 dB on Wan2.1.

PhysVideo: Physically Plausible Video Generation with Cross-View Geometry Guidance

Recent progress in video generation has led to substantial improvements in visual fidelity, yet ensuring physically consistent motion remains a fundamental challenge. Intuitively, this limitation can be attributed to the fact that real-world object motion unfolds in three-dimensional space, while video observations provide only partial, view-dependent projections of such dynamics. To address these issues, we propose PhysVideo, a two-stage framework that first generates physics-aware orthogonal foreground videos and then synthesizes full videos with background. In the first stage, Phys4View leverages physics-aware attention to capture the influence of physical attributes on motion dynamics, and enhances spatio-temporal consistency by incorporating geometry-enhanced cross-view attention and temporal attention. In the second stage, VideoSyn uses the generated foreground videos as guidance and learns the interactions between foreground dynamics and background context for controllable video synthesis. To support training, we construct PhysMV, a dataset containing 40K scenes, each consisting of four orthogonal viewpoints, resulting in a total of 160K video sequences. Extensive experiments demonstrate that PhysVideo significantly improves physical realism and spatial-temporal coherence over existing video generation methods. Home page: https://anonymous.4open.science/w/Phys4D/.

VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Model

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.

AR4D: Autoregressive 4D Generation from Monocular Videos

Recent advancements in generative models have ignited substantial interest in dynamic 3D content creation (\ie, 4D generation). Existing approaches primarily rely on Score Distillation Sampling (SDS) to infer novel-view videos, typically leading to issues such as limited diversity, spatial-temporal inconsistency and poor prompt alignment, due to the inherent randomness of SDS. To tackle these problems, we propose AR4D, a novel paradigm for SDS-free 4D generation. Specifically, our paradigm consists of three stages. To begin with, for a monocular video that is either generated or captured, we first utilize pre-trained expert models to create a 3D representation of the first frame, which is further fine-tuned to serve as the canonical space. Subsequently, motivated by the fact that videos happen naturally in an autoregressive manner, we propose to generate each frame's 3D representation based on its previous frame's representation, as this autoregressive generation manner can facilitate more accurate geometry and motion estimation. Meanwhile, to prevent overfitting during this process, we introduce a progressive view sampling strategy, utilizing priors from pre-trained large-scale 3D reconstruction models. To avoid appearance drift introduced by autoregressive generation, we further incorporate a refinement stage based on a global deformation field and the geometry of each frame's 3D representation. Extensive experiments have demonstrated that AR4D can achieve state-of-the-art 4D generation without SDS, delivering greater diversity, improved spatial-temporal consistency, and better alignment with input prompts.

GSemSplat: Generalizable Semantic 3D Gaussian Splatting from Uncalibrated Image Pairs

Modeling and understanding the 3D world is crucial for various applications, from augmented reality to robotic navigation. Recent advancements based on 3D Gaussian Splatting have integrated semantic information from multi-view images into Gaussian primitives. However, these methods typically require costly per-scene optimization from dense calibrated images, limiting their practicality. In this paper, we consider the new task of generalizable 3D semantic field modeling from sparse, uncalibrated image pairs. Building upon the Splatt3R architecture, we introduce GSemSplat, a framework that learns open-vocabulary semantic representations linked to 3D Gaussians without the need for per-scene optimization, dense image collections or calibration. To ensure effective and reliable learning of semantic features in 3D space, we employ a dual-feature approach that leverages both region-specific and context-aware semantic features as supervision in the 2D space. This allows us to capitalize on their complementary strengths. Experimental results on the ScanNet++ dataset demonstrate the effectiveness and superiority of our approach compared to the traditional scene-specific method. We hope our work will inspire more research into generalizable 3D understanding.

TIV-Diffusion: Towards Object-Centric Movement for Text-driven Image to Video Generation

Text-driven Image to Video Generation (TI2V) aims to generate controllable video given the first frame and corresponding textual description. The primary challenges of this task lie in two parts: (i) how to identify the target objects and ensure the consistency between the movement trajectory and the textual description. (ii) how to improve the subjective quality of generated videos. To tackle the above challenges, we propose a new diffusion-based TI2V framework, termed TIV-Diffusion, via object-centric textual-visual alignment, intending to achieve precise control and high-quality video generation based on textual-described motion for different objects. Concretely, we enable our TIV-Diffuion model to perceive the textual-described objects and their motion trajectory by incorporating the fused textual and visual knowledge through scale-offset modulation. Moreover, to mitigate the problems of object disappearance and misaligned objects and motion, we introduce an object-centric textual-visual alignment module, which reduces the risk of misaligned objects/motion by decoupling the objects in the reference image and aligning textual features with each object individually. Based on the above innovations, our TIV-Diffusion achieves state-of-the-art high-quality video generation compared with existing TI2V methods.

Compositional 3D-aware Video Generation with LLM Director

Significant progress has been made in text-to-video generation through the use of powerful generative models and large-scale internet data. However, substantial challenges remain in precisely controlling individual concepts within the generated video, such as the motion and appearance of specific characters and the movement of viewpoints. In this work, we propose a novel paradigm that generates each concept in 3D representation separately and then composes them with priors from Large Language Models (LLM) and 2D diffusion models. Specifically, given an input textual prompt, our scheme consists of three stages: 1) We leverage LLM as the director to first decompose the complex query into several sub-prompts that indicate individual concepts within the video~(\textit{e.g.}, scene, objects, motions), then we let LLM to invoke pre-trained expert models to obtain corresponding 3D representations of concepts. 2) To compose these representations, we prompt multi-modal LLM to produce coarse guidance on the scales and coordinates of trajectories for the objects. 3) To make the generated frames adhere to natural image distribution, we further leverage 2D diffusion priors and use Score Distillation Sampling to refine the composition. Extensive experiments demonstrate that our method can generate high-fidelity videos from text with diverse motion and flexible control over each concept. Project page: \url{https://aka.ms/c3v}.