Video Diffusion Models (VDMs) have emerged as powerful generative tools, capable of synthesizing high-quality spatiotemporal content. Yet, their potential goes far beyond mere video generation. We argue that the training dynamics of VDMs, driven by the need to model coherent sequences, naturally pushes them to internalize structured representations and an implicit understanding of the visual world. To probe the extent of this internal knowledge, we introduce a few-shot fine-tuning framework that repurposes VDMs for new tasks using only a handful of examples. Our method transforms each task into a visual transition, enabling the training of LoRA weights on short input-output sequences without altering the generative interface of a frozen VDM. Despite minimal supervision, the model exhibits strong generalization across diverse tasks, from low-level vision (for example, segmentation and pose estimation) to high-level reasoning (for example, on ARC-AGI). These results reframe VDMs as more than generative engines. They are adaptable visual learners with the potential to serve as the backbone for future foundation models in vision.
Acquiring real-world robotic simulation data through teleoperation is notoriously time-consuming and labor-intensive. Recently, action-driven generative models have gained widespread adoption in robot learning and simulation, as they eliminate safety concerns and reduce maintenance efforts. However, the action sequences used in these methods often result in limited control precision and poor generalization due to their globally coarse alignment. To address these limitations, we propose ORV, an Occupancy-centric Robot Video generation framework, which utilizes 4D semantic occupancy sequences as a fine-grained representation to provide more accurate semantic and geometric guidance for video generation. By leveraging occupancy-based representations, ORV enables seamless translation of simulation data into photorealistic robot videos, while ensuring high temporal consistency and precise controllability. Furthermore, our framework supports the simultaneous generation of multi-view videos of robot gripping operations - an important capability for downstream robotic learning tasks. Extensive experimental results demonstrate that ORV consistently outperforms existing baseline methods across various datasets and sub-tasks. Demo, Code and Model: https://orangesodahub.github.io/ORV
The emergence of Diffusion Transformers (DiT) has brought significant advancements to video generation, especially in text-to-video and image-to-video tasks. Although video generation is widely applied in various fields, most existing models are limited to single scenarios and cannot perform diverse video generation and editing through dynamic content manipulation. We propose OmniV2V, a video model capable of generating and editing videos across different scenarios based on various operations, including: object movement, object addition, mask-guided video edit, try-on, inpainting, outpainting, human animation, and controllable character video synthesis. We explore a unified dynamic content manipulation injection module, which effectively integrates the requirements of the above tasks. In addition, we design a visual-text instruction module based on LLaVA, enabling the model to effectively understand the correspondence between visual content and instructions. Furthermore, we build a comprehensive multi-task data processing system. Since there is data overlap among various tasks, this system can efficiently provide data augmentation. Using this system, we construct a multi-type, multi-scenario OmniV2V dataset and its corresponding OmniV2V-Test benchmark. Extensive experiments show that OmniV2V works as well as, and sometimes better than, the best existing open-source and commercial models for many video generation and editing tasks.
Driving world models are used to simulate futures by video generation based on the condition of the current state and actions. However, current models often suffer serious error accumulations when predicting the long-term future, which limits the practical application. Recent studies utilize the Diffusion Transformer (DiT) as the backbone of driving world models to improve learning flexibility. However, these models are always trained on short video clips (high fps and short duration), and multiple roll-out generations struggle to produce consistent and reasonable long videos due to the training-inference gap. To this end, we propose several solutions to build a simple yet effective long-term driving world model. First, we hierarchically decouple world model learning into large motion learning and bidirectional continuous motion learning. Then, considering the continuity of driving scenes, we propose a simple distillation method where fine-grained video flows are self-supervised signals for coarse-grained flows. The distillation is designed to improve the coherence of infinite video generation. The coarse-grained and fine-grained modules are coordinated to generate long-term and temporally coherent videos. In the public benchmark NuScenes, compared with the state-of-the-art front-view model, our model improves FVD by $27\%$ and reduces inference time by $85\%$ for the video task of generating 110+ frames. More videos (including 90s duration) are available at https://Wang-Xiaodong1899.github.io/longdwm/.
World models serve as essential building blocks toward Artificial General Intelligence (AGI), enabling intelligent agents to predict future states and plan actions by simulating complex physical interactions. However, existing interactive models primarily predict visual observations, thereby neglecting crucial hidden states like geometric structures and spatial coherence. This leads to rapid error accumulation and temporal inconsistency. To address these limitations, we introduce DeepVerse, a novel 4D interactive world model explicitly incorporating geometric predictions from previous timesteps into current predictions conditioned on actions. Experiments demonstrate that by incorporating explicit geometric constraints, DeepVerse captures richer spatio-temporal relationships and underlying physical dynamics. This capability significantly reduces drift and enhances temporal consistency, enabling the model to reliably generate extended future sequences and achieve substantial improvements in prediction accuracy, visual realism, and scene rationality. Furthermore, our method provides an effective solution for geometry-aware memory retrieval, effectively preserving long-term spatial consistency. We validate the effectiveness of DeepVerse across diverse scenarios, establishing its capacity for high-fidelity, long-horizon predictions grounded in geometry-aware dynamics.
Foundational world models must be both interactive and preserve spatiotemporal coherence for effective future planning with action choices. However, present models for long video generation have limited inherent world modeling capabilities due to two main challenges: compounding errors and insufficient memory mechanisms. We enhance image-to-video models with interactive capabilities through additional action conditioning and autoregressive framework, and reveal that compounding error is inherently irreducible in autoregressive video generation, while insufficient memory mechanism leads to incoherence of world models. We propose video retrieval augmented generation (VRAG) with explicit global state conditioning, which significantly reduces long-term compounding errors and increases spatiotemporal consistency of world models. In contrast, naive autoregressive generation with extended context windows and retrieval-augmented generation prove less effective for video generation, primarily due to the limited in-context learning capabilities of current video models. Our work illuminates the fundamental challenges in video world models and establishes a comprehensive benchmark for improving video generation models with internal world modeling capabilities.
Recent video diffusion models have demonstrated their great capability in generating visually-pleasing results, while synthesizing the correct physical effects in generated videos remains challenging. The complexity of real-world motions, interactions, and dynamics introduce great difficulties when learning physics from data. In this work, we propose DiffPhy, a generic framework that enables physically-correct and photo-realistic video generation by fine-tuning a pre-trained video diffusion model. Our method leverages large language models (LLMs) to explicitly reason a comprehensive physical context from the text prompt and use it to guide the generation. To incorporate physical context into the diffusion model, we leverage a Multimodal large language model (MLLM) as a supervisory signal and introduce a set of novel training objectives that jointly enforce physical correctness and semantic consistency with the input text. We also establish a high-quality physical video dataset containing diverse phyiscal actions and events to facilitate effective finetuning. Extensive experiments on public benchmarks demonstrate that DiffPhy is able to produce state-of-the-art results across diverse physics-related scenarios. Our project page is available at https://bwgzk-keke.github.io/DiffPhy/
Recent advances in video generation demand increasingly efficient training recipes to mitigate escalating computational costs. In this report, we present ContentV, an 8B-parameter text-to-video model that achieves state-of-the-art performance (85.14 on VBench) after training on 256 x 64GB Neural Processing Units (NPUs) for merely four weeks. ContentV generates diverse, high-quality videos across multiple resolutions and durations from text prompts, enabled by three key innovations: (1) A minimalist architecture that maximizes reuse of pre-trained image generation models for video generation; (2) A systematic multi-stage training strategy leveraging flow matching for enhanced efficiency; and (3) A cost-effective reinforcement learning with human feedback framework that improves generation quality without requiring additional human annotations. All the code and models are available at: https://contentv.github.io.
Video diffusion transformers (vDiTs) have made impressive progress in text-to-video generation, but their high computational demands present major challenges for practical deployment. While existing acceleration methods reduce workload at various granularities, they often rely on heuristics, limiting their applicability. We introduce ASTRAEA, an automatic framework that searches for near-optimal configurations for vDiT-based video generation. At its core, ASTRAEA proposes a lightweight token selection mechanism and a memory-efficient, GPU-parallel sparse attention strategy, enabling linear reductions in execution time with minimal impact on generation quality. To determine optimal token reduction for different timesteps, we further design a search framework that leverages a classic evolutionary algorithm to automatically determine the distribution of the token budget effectively. Together, ASTRAEA achieves up to 2.4x inference speedup on a single GPU with great scalability (up to 13.2x speedup on 8 GPUs) while retaining better video quality compared to the state-of-the-art methods (<0.5% loss on the VBench score compared to the baseline vDiT models).
Synthesizing high-quality dynamic medical videos remains a significant challenge due to the need for modeling both spatial consistency and temporal dynamics. Existing Transformer-based approaches face critical limitations, including insufficient channel interactions, high computational complexity from self-attention, and coarse denoising guidance from timestep embeddings when handling varying noise levels. In this work, we propose FEAT, a full-dimensional efficient attention Transformer, which addresses these issues through three key innovations: (1) a unified paradigm with sequential spatial-temporal-channel attention mechanisms to capture global dependencies across all dimensions, (2) a linear-complexity design for attention mechanisms in each dimension, utilizing weighted key-value attention and global channel attention, and (3) a residual value guidance module that provides fine-grained pixel-level guidance to adapt to different noise levels. We evaluate FEAT on standard benchmarks and downstream tasks, demonstrating that FEAT-S, with only 23\% of the parameters of the state-of-the-art model Endora, achieves comparable or even superior performance. Furthermore, FEAT-L surpasses all comparison methods across multiple datasets, showcasing both superior effectiveness and scalability. Code is available at https://github.com/Yaziwel/FEAT.
Fine-grained and efficient controllability on video diffusion transformers has raised increasing desires for the applicability. Recently, In-context Conditioning emerged as a powerful paradigm for unified conditional video generation, which enables diverse controls by concatenating varying context conditioning signals with noisy video latents into a long unified token sequence and jointly processing them via full-attention, e.g., FullDiT. Despite their effectiveness, these methods face quadratic computation overhead as task complexity increases, hindering practical deployment. In this paper, we study the efficiency bottleneck neglected in original in-context conditioning video generation framework. We begin with systematic analysis to identify two key sources of the computation inefficiencies: the inherent redundancy within context condition tokens and the computational redundancy in context-latent interactions throughout the diffusion process. Based on these insights, we propose FullDiT2, an efficient in-context conditioning framework for general controllability in both video generation and editing tasks, which innovates from two key perspectives. Firstly, to address the token redundancy, FullDiT2 leverages a dynamic token selection mechanism to adaptively identify important context tokens, reducing the sequence length for unified full-attention. Additionally, a selective context caching mechanism is devised to minimize redundant interactions between condition tokens and video latents. Extensive experiments on six diverse conditional video editing and generation tasks demonstrate that FullDiT2 achieves significant computation reduction and 2-3 times speedup in averaged time cost per diffusion step, with minimal degradation or even higher performance in video generation quality. The project page is at \href{https://fulldit2.github.io/}{https://fulldit2.github.io/}.
Transformer-based models like ViViT and TimeSformer have advanced video understanding by effectively modeling spatiotemporal dependencies. Recent video generation models, such as Sora and Vidu, further highlight the power of transformers in long-range feature extraction and holistic spatiotemporal modeling. However, directly applying these models to real-world video super-resolution (VSR) is challenging, as VSR demands pixel-level precision, which can be compromised by tokenization and sequential attention mechanisms. While recent transformer-based VSR models attempt to address these issues using smaller patches and local attention, they still face limitations such as restricted receptive fields and dependence on optical flow-based alignment, which can introduce inaccuracies in real-world settings. To overcome these issues, we propose Dual Axial Spatial$\times$Temporal Transformer for Real-World Video Super-Resolution (DualX-VSR), which introduces a novel dual axial spatial$\times$temporal attention mechanism that integrates spatial and temporal information along orthogonal directions. DualX-VSR eliminates the need for motion compensation, offering a simplified structure that provides a cohesive representation of spatiotemporal information. As a result, DualX-VSR achieves high fidelity and superior performance in real-world VSR task.
Diffusion generative models have become the standard for producing high-quality, coherent video content, yet their slow inference speeds and high computational demands hinder practical deployment. Although both quantization and sparsity can independently accelerate inference while maintaining generation quality, naively combining these techniques in existing training-free approaches leads to significant performance degradation due to the lack of joint optimization. We introduce FPSAttention, a novel training-aware co-design of FP8 quantization and sparsity for video generation, with a focus on the 3D bi-directional attention mechanism. Our approach features three key innovations: 1) A unified 3D tile-wise granularity that simultaneously supports both quantization and sparsity; 2) A denoising step-aware strategy that adapts to the noise schedule, addressing the strong correlation between quantization/sparsity errors and denoising steps; 3) A native, hardware-friendly kernel that leverages FlashAttention and is implemented with optimized Hopper architecture features for highly efficient execution. Trained on Wan2.1's 1.3B and 14B models and evaluated on the VBench benchmark, FPSAttention achieves a 7.09x kernel speedup for attention operations and a 4.96x end-to-end speedup for video generation compared to the BF16 baseline at 720p resolution-without sacrificing generation quality.
We introduce Follow-Your-Creation, a novel 4D video creation framework capable of both generating and editing 4D content from a single monocular video input. By leveraging a powerful video inpainting foundation model as a generative prior, we reformulate 4D video creation as a video inpainting task, enabling the model to fill in missing content caused by camera trajectory changes or user edits. To facilitate this, we generate composite masked inpainting video data to effectively fine-tune the model for 4D video generation. Given an input video and its associated camera trajectory, we first perform depth-based point cloud rendering to obtain invisibility masks that indicate the regions that should be completed. Simultaneously, editing masks are introduced to specify user-defined modifications, and these are combined with the invisibility masks to create a composite masks dataset. During training, we randomly sample different types of masks to construct diverse and challenging inpainting scenarios, enhancing the model's generalization and robustness in various 4D editing and generation tasks. To handle temporal consistency under large camera motion, we design a self-iterative tuning strategy that gradually increases the viewing angles during training, where the model is used to generate the next-stage training data after each fine-tuning iteration. Moreover, we introduce a temporal packaging module during inference to enhance generation quality. Our method effectively leverages the prior knowledge of the base model without degrading its original performance, enabling the generation of 4D videos with consistent multi-view coherence. In addition, our approach supports prompt-based content editing, demonstrating strong flexibility and significantly outperforming state-of-the-art methods in both quality and versatility.
Robotics has broad applications from automating house chores to taking care of patients. However, evaluating robot control policies is challenging, as real-world testing is expensive, while handcrafted simulations often fail to accurately reflect real-world conditions, resulting in poor correlation between simulated evaluation and real-world outcomes. In this work, we investigate World-model-based Policy Evaluation (WPE). We first train an action-conditioned video generation model as a proxy to real-world environments. To enable efficient rollouts of hundreds of interactive steps while mitigating error accumulation in the world model, we propose an inference scheme which we call Blockwise-Autoregressive Diffusion Transformer with adjustable context and decoding horizon lengths. To ensure that the world model indeed follows action input, we propose metrics based on the agreement between the ground truth video and generated video conditioned on the same sequence of actions to evaluate the world model. We then use the world model for policy evaluation by performing Monte Carlo rollouts in the world model while employing a vision-language model (VLM) as a reward function. Interestingly, we found that WPE tends to underestimate the policy values for in-distribution actions and overestimate policy values for out-of-distribution actions. Nevertheless, WPE preserves the relative rankings of different policies. In emulating real robot executions, WPE achieves high fidelity in mimicing robot arm movements as in real videos, while emulating highly realistic object interaction remains challenging. Despite this limitation, we show that a world model can serve as a starting point for evaluating robot policies before real-world deployment.
Video diffusion techniques have advanced significantly in recent years; however, they struggle to generate realistic imagery of car crashes due to the scarcity of accident events in most driving datasets. Improving traffic safety requires realistic and controllable accident simulations. To tackle the problem, we propose Ctrl-Crash, a controllable car crash video generation model that conditions on signals such as bounding boxes, crash types, and an initial image frame. Our approach enables counterfactual scenario generation where minor variations in input can lead to dramatically different crash outcomes. To support fine-grained control at inference time, we leverage classifier-free guidance with independently tunable scales for each conditioning signal. Ctrl-Crash achieves state-of-the-art performance across quantitative video quality metrics (e.g., FVD and JEDi) and qualitative measurements based on a human-evaluation of physical realism and video quality compared to prior diffusion-based methods.
Although diffusion-based models can generate high-quality and high-resolution video sequences from textual or image inputs, they lack explicit integration of geometric cues when controlling scene lighting and visual appearance across frames. To address this limitation, we propose IllumiCraft, an end-to-end diffusion framework accepting three complementary inputs: (1) high-dynamic-range (HDR) video maps for detailed lighting control; (2) synthetically relit frames with randomized illumination changes (optionally paired with a static background reference image) to provide appearance cues; and (3) 3D point tracks that capture precise 3D geometry information. By integrating the lighting, appearance, and geometry cues within a unified diffusion architecture, IllumiCraft generates temporally coherent videos aligned with user-defined prompts. It supports background-conditioned and text-conditioned video relighting and provides better fidelity than existing controllable video generation methods. Project Page: https://yuanze-lin.me/IllumiCraft_page
In this paper, we present TalkingMachines -- an efficient framework that transforms pretrained video generation models into real-time, audio-driven character animators. TalkingMachines enables natural conversational experiences by integrating an audio large language model (LLM) with our video generation foundation model. Our primary contributions include: (1) We adapt a pretrained SOTA image-to-video DiT into an audio-driven avatar generation model of 18 billion parameters; (2) We enable infinite video streaming without error accumulation through asymmetric knowledge distillation from a bidirectional teacher model into a sparse causal, autoregressive student model; (3) We design a high-throughput, low-latency inference pipeline incorporating several key engineering optimizations such as: (a) disaggregation of the DiT and VAE decoder across separate devices, (b) efficient overlap of inter-device communication and computation using CUDA streams, (c) elimination of redundant recomputations to maximize frame-generation throughput. Please see demo videos here - https://aaxwaz.github.io/TalkingMachines/
The rapid development of Artificial Intelligence Generated Content (AIGC) has led to significant progress in video generation but also raises serious concerns about intellectual property protection and reliable content tracing. Watermarking is a widely adopted solution to this issue, but existing methods for video generation mainly follow a post-generation paradigm, which introduces additional computational overhead and often fails to effectively balance the trade-off between video quality and watermark extraction. To address these issues, we propose Video Signature (VIDSIG), an in-generation watermarking method for latent video diffusion models, which enables implicit and adaptive watermark integration during generation. Specifically, we achieve this by partially fine-tuning the latent decoder, where Perturbation-Aware Suppression (PAS) pre-identifies and freezes perceptually sensitive layers to preserve visual quality. Beyond spatial fidelity, we further enhance temporal consistency by introducing a lightweight Temporal Alignment module that guides the decoder to generate coherent frame sequences during fine-tuning. Experimental results show that VIDSIG achieves the best overall performance in watermark extraction, visual quality, and generation efficiency. It also demonstrates strong robustness against both spatial and temporal tampering, highlighting its practicality in real-world scenarios.
Hand-Object Interaction (HOI) generation has significant application potential. However, current 3D HOI motion generation approaches heavily rely on predefined 3D object models and lab-captured motion data, limiting generalization capabilities. Meanwhile, HOI video generation methods prioritize pixel-level visual fidelity, often sacrificing physical plausibility. Recognizing that visual appearance and motion patterns share fundamental physical laws in the real world, we propose a novel framework that combines visual priors and dynamic constraints within a synchronized diffusion process to generate the HOI video and motion simultaneously. To integrate the heterogeneous semantics, appearance, and motion features, our method implements tri-modal adaptive modulation for feature aligning, coupled with 3D full-attention for modeling inter- and intra-modal dependencies. Furthermore, we introduce a vision-aware 3D interaction diffusion model that generates explicit 3D interaction sequences directly from the synchronized diffusion outputs, then feeds them back to establish a closed-loop feedback cycle. This architecture eliminates dependencies on predefined object models or explicit pose guidance while significantly enhancing video-motion consistency. Experimental results demonstrate our method's superiority over state-of-the-art approaches in generating high-fidelity, dynamically plausible HOI sequences, with notable generalization capabilities in unseen real-world scenarios. Project page at \href{https://github.com/Droliven}{https://github.com/Droliven}.