Recent advances in video generation can produce realistic, minute-long single-shot videos with scalable diffusion transformers. However, real-world narrative videos require multi-shot scenes with visual and dynamic consistency across shots. In this work, we introduce Long Context Tuning (LCT), a training paradigm that expands the context window of pre-trained single-shot video diffusion models to learn scene-level consistency directly from data. Our method expands full attention mechanisms from individual shots to encompass all shots within a scene, incorporating interleaved 3D position embedding and an asynchronous noise strategy, enabling both joint and auto-regressive shot generation without additional parameters. Models with bidirectional attention after LCT can further be fine-tuned with context-causal attention, facilitating auto-regressive generation with efficient KV-cache. Experiments demonstrate single-shot models after LCT can produce coherent multi-shot scenes and exhibit emerging capabilities, including compositional generation and interactive shot extension, paving the way for more practical visual content creation. See https://guoyww.github.io/projects/long-context-video/ for more details.
Recent advancements in portrait video generation have been noteworthy. However, existing methods rely heavily on human priors and pre-trained generation models, which may introduce unrealistic motion and lead to inefficient inference. To address these challenges, we propose Semantic Latent Motion (SeMo), a compact and expressive motion representation. Leveraging this representation, our approach achieve both high-quality visual results and efficient inference. SeMo follows an effective three-step framework: Abstraction, Reasoning, and Generation. First, in the Abstraction step, we use a carefully designed Mask Motion Encoder to compress the subject's motion state into a compact and abstract latent motion (1D token). Second, in the Reasoning step, long-term modeling and efficient reasoning are performed in this latent space to generate motion sequences. Finally, in the Generation step, the motion dynamics serve as conditional information to guide the generation model in synthesizing realistic transitions from reference frames to target frames. Thanks to the compact and descriptive nature of Semantic Latent Motion, our method enables real-time video generation with highly realistic motion. User studies demonstrate that our approach surpasses state-of-the-art models with an 81% win rate in realism. Extensive experiments further highlight its strong compression capability, reconstruction quality, and generative potential. Moreover, its fully self-supervised nature suggests promising applications in broader video generation tasks.
Video generation has advanced rapidly, improving evaluation methods, yet assessing video's motion remains a major challenge. Specifically, there are two key issues: 1) current motion metrics do not fully align with human perceptions; 2) the existing motion prompts are limited. Based on these findings, we introduce VMBench--a comprehensive Video Motion Benchmark that has perception-aligned motion metrics and features the most diverse types of motion. VMBench has several appealing properties: 1) Perception-Driven Motion Evaluation Metrics, we identify five dimensions based on human perception in motion video assessment and develop fine-grained evaluation metrics, providing deeper insights into models' strengths and weaknesses in motion quality. 2) Meta-Guided Motion Prompt Generation, a structured method that extracts meta-information, generates diverse motion prompts with LLMs, and refines them through human-AI validation, resulting in a multi-level prompt library covering six key dynamic scene dimensions. 3) Human-Aligned Validation Mechanism, we provide human preference annotations to validate our benchmarks, with our metrics achieving an average 35.3% improvement in Spearman's correlation over baseline methods. This is the first time that the quality of motion in videos has been evaluated from the perspective of human perception alignment. Additionally, we will soon release VMBench at https://github.com/GD-AIGC/VMBench, setting a new standard for evaluating and advancing motion generation models.
Video generation has witnessed remarkable progress with the advent of deep generative models, particularly diffusion models. While existing methods excel in generating high-quality videos from text prompts or single images, personalized multi-subject video generation remains a largely unexplored challenge. This task involves synthesizing videos that incorporate multiple distinct subjects, each defined by separate reference images, while ensuring temporal and spatial consistency. Current approaches primarily rely on mapping subject images to keywords in text prompts, which introduces ambiguity and limits their ability to model subject relationships effectively. In this paper, we propose CINEMA, a novel framework for coherent multi-subject video generation by leveraging Multimodal Large Language Model (MLLM). Our approach eliminates the need for explicit correspondences between subject images and text entities, mitigating ambiguity and reducing annotation effort. By leveraging MLLM to interpret subject relationships, our method facilitates scalability, enabling the use of large and diverse datasets for training. Furthermore, our framework can be conditioned on varying numbers of subjects, offering greater flexibility in personalized content creation. Through extensive evaluations, we demonstrate that our approach significantly improves subject consistency, and overall video coherence, paving the way for advanced applications in storytelling, interactive media, and personalized video generation.
Long video generation remains a challenging and compelling topic in computer vision. Diffusion based models, among the various approaches to video generation, have achieved state of the art quality with their iterative denoising procedures. However, the intrinsic complexity of the video domain renders the training of such diffusion models exceedingly expensive in terms of both data curation and computational resources. Moreover, these models typically operate on a fixed noise tensor that represents the video, resulting in predetermined spatial and temporal dimensions. Although several high quality open-source pretrained video diffusion models, jointly trained on images and videos of varying lengths and resolutions, are available, it is generally not recommended to specify a video length at inference that was not included in the training set. Consequently, these models are not readily adaptable to the direct generation of longer videos by merely increasing the specified video length. In addition to feasibility challenges, long-video generation also encounters quality issues. The domain of long videos is inherently more complex than that of short videos: extended durations introduce greater variability and necessitate long-range temporal consistency, thereby increasing the overall difficulty of the task. We propose VideoMerge, a training-free method that can be seamlessly adapted to merge short videos generated by pretrained text-to-video diffusion model. Our approach preserves the model's original expressiveness and consistency while allowing for extended duration and dynamic variation as specified by the user. By leveraging the strengths of pretrained models, our method addresses challenges related to smoothness, consistency, and dynamic content through orthogonal strategies that operate collaboratively to achieve superior quality.
Automatically generating training supervision for embodied tasks is crucial, as manual designing is tedious and not scalable. While prior works use large language models (LLMs) or vision-language models (VLMs) to generate rewards, these approaches are largely limited to simple tasks with well-defined rewards, such as pick-and-place. This limitation arises because LLMs struggle to interpret complex scenes compressed into text or code due to their restricted input modality, while VLM-based rewards, though better at visual perception, remain limited by their less expressive output modality. To address these challenges, we leverage the imagination capability of general-purpose video generation models. Given an initial simulation frame and a textual task description, the video generation model produces a video demonstrating task completion with correct semantics. We then extract rich supervisory signals from the generated video, including 6D object pose sequences, 2D segmentations, and estimated depth, to facilitate task learning in simulation. Our approach significantly improves supervision quality for complex embodied tasks, enabling large-scale training in simulators.
Vision Language Models (VLMs) have demonstrated significant potential in various downstream tasks, including Image/Video Generation, Visual Question Answering, Multimodal Chatbots, and Video Understanding. However, these models often struggle with basic image transformations. This paper investigates the image-level understanding of VLMs, specifically CLIP by OpenAI and SigLIP by Google. Our findings reveal that these models lack comprehension of multiple image-level augmentations. To facilitate this study, we created an augmented version of the Flickr8k dataset, pairing each image with a detailed description of the applied transformation. We further explore how this deficiency impacts downstream tasks, particularly in image editing, and evaluate the performance of state-of-the-art Image2Image models on simple transformations.
We present I2V3D, a novel framework for animating static images into dynamic videos with precise 3D control, leveraging the strengths of both 3D geometry guidance and advanced generative models. Our approach combines the precision of a computer graphics pipeline, enabling accurate control over elements such as camera movement, object rotation, and character animation, with the visual fidelity of generative AI to produce high-quality videos from coarsely rendered inputs. To support animations with any initial start point and extended sequences, we adopt a two-stage generation process guided by 3D geometry: 1) 3D-Guided Keyframe Generation, where a customized image diffusion model refines rendered keyframes to ensure consistency and quality, and 2) 3D-Guided Video Interpolation, a training-free approach that generates smooth, high-quality video frames between keyframes using bidirectional guidance. Experimental results highlight the effectiveness of our framework in producing controllable, high-quality animations from single input images by harmonizing 3D geometry with generative models. The code for our framework will be publicly released.
Text-based diffusion models have made significant breakthroughs in generating high-quality images and videos from textual descriptions. However, the lengthy sampling time of the denoising process remains a significant bottleneck in practical applications. Previous methods either ignore the statistical relationships between adjacent steps or rely on attention or feature similarity between them, which often only works with specific network structures. To address this issue, we discover a new statistical relationship in the transition operator between adjacent steps, focusing on the relationship of the outputs from the network. This relationship does not impose any requirements on the network structure. Based on this observation, we propose a novel training-free acceleration method called LTC-Accel, which uses the identified relationship to estimate the current transition operator based on adjacent steps. Due to no specific assumptions regarding the network structure, LTC-Accel is applicable to almost all diffusion-based methods and orthogonal to almost all existing acceleration techniques, making it easy to combine with them. Experimental results demonstrate that LTC-Accel significantly speeds up sampling in text-to-image and text-to-video synthesis while maintaining competitive sample quality. Specifically, LTC-Accel achieves a speedup of 1.67-fold in Stable Diffusion v2 and a speedup of 1.55-fold in video generation models. When combined with distillation models, LTC-Accel achieves a remarkable 10-fold speedup in video generation, allowing real-time generation of more than 16FPS.
Video generation models have achieved remarkable progress in the past year. The quality of AI video continues to improve, but at the cost of larger model size, increased data quantity, and greater demand for training compute. In this report, we present Open-Sora 2.0, a commercial-level video generation model trained for only $200k. With this model, we demonstrate that the cost of training a top-performing video generation model is highly controllable. We detail all techniques that contribute to this efficiency breakthrough, including data curation, model architecture, training strategy, and system optimization. According to human evaluation results and VBench scores, Open-Sora 2.0 is comparable to global leading video generation models including the open-source HunyuanVideo and the closed-source Runway Gen-3 Alpha. By making Open-Sora 2.0 fully open-source, we aim to democratize access to advanced video generation technology, fostering broader innovation and creativity in content creation. All resources are publicly available at: https://github.com/hpcaitech/Open-Sora.
Large-scale pre-trained video generation models excel in content creation but are not reliable as physically accurate world simulators out of the box. This work studies the process of post-training these models for accurate world modeling through the lens of the simple, yet fundamental, physics task of modeling object freefall. We show state-of-the-art video generation models struggle with this basic task, despite their visually impressive outputs. To remedy this problem, we find that fine-tuning on a relatively small amount of simulated videos is effective in inducing the dropping behavior in the model, and we can further improve results through a novel reward modeling procedure we introduce. Our study also reveals key limitations of post-training in generalization and distribution modeling. Additionally, we release a benchmark for this task that may serve as a useful diagnostic tool for tracking physical accuracy in large-scale video generative model development.
Joint audio-video (AV) generation is still a significant challenge in generative AI, primarily due to three critical requirements: quality of the generated samples, seamless multimodal synchronization and temporal coherence, with audio tracks that match the visual data and vice versa, and limitless video duration. In this paper, we present $^R$-FLAV, a novel transformer-based architecture that addresses all the key challenges of AV generation. We explore three distinct cross modality interaction modules, with our lightweight temporal fusion module emerging as the most effective and computationally efficient approach for aligning audio and visual modalities. Our experimental results demonstrate that $^R$-FLAV outperforms existing state-of-the-art models in multimodal AV generation tasks. Our code and checkpoints are available at https://github.com/ErgastiAlex/R-FLAV.
Recent advancements in video generation have spurred the development of video editing techniques, which can be divided into inversion-based and end-to-end methods. However, current video editing methods still suffer from several challenges. Inversion-based methods, though training-free and flexible, are time-consuming during inference, struggle with fine-grained editing instructions, and produce artifacts and jitter. On the other hand, end-to-end methods, which rely on edited video pairs for training, offer faster inference speeds but often produce poor editing results due to a lack of high-quality training video pairs. In this paper, to close the gap in end-to-end methods, we introduce Se\~norita-2M, a high-quality video editing dataset. Se\~norita-2M consists of approximately 2 millions of video editing pairs. It is built by crafting four high-quality, specialized video editing models, each crafted and trained by our team to achieve state-of-the-art editing results. We also propose a filtering pipeline to eliminate poorly edited video pairs. Furthermore, we explore common video editing architectures to identify the most effective structure based on current pre-trained generative model. Extensive experiments show that our dataset can help to yield remarkably high-quality video editing results. More details are available at https://senorita-2m-dataset.github.io.
We present a unified network for simultaneously generating videos and their corresponding entity segmentation and depth maps from text prompts. We utilize colormap to represent entity masks and depth maps, tightly integrating dense prediction with RGB video generation. Introducing dense prediction information improves video generation's consistency and motion smoothness without increasing computational costs. Incorporating learnable task embeddings brings multiple dense prediction tasks into a single model, enhancing flexibility and further boosting performance. We further propose a large-scale dense prediction video dataset~\datasetname, addressing the issue that existing datasets do not concurrently contain captions, videos, segmentation, or depth maps. Comprehensive experiments demonstrate the high efficiency of our method, surpassing the state-of-the-art in terms of video quality, consistency, and motion smoothness.
Advanced end-to-end autonomous driving systems predict other vehicles' motions and plan ego vehicle's trajectory. The world model that can foresee the outcome of the trajectory has been used to evaluate the end-to-end autonomous driving system. However, existing world models predominantly emphasize the trajectory of the ego vehicle and leave other vehicles uncontrollable. This limitation hinders their ability to realistically simulate the interaction between the ego vehicle and the driving scenario. In addition, it remains a challenge to match multiple trajectories with each vehicle in the video to control the video generation. To address above issues, a driving \textbf{W}orld \textbf{M}odel named EOT-WM is proposed in this paper, unifying \textbf{E}go-\textbf{O}ther vehicle \textbf{T}rajectories in videos. Specifically, we first project ego and other vehicle trajectories in the BEV space into the image coordinate to match each trajectory with its corresponding vehicle in the video. Then, trajectory videos are encoded by the Spatial-Temporal Variational Auto Encoder to align with driving video latents spatially and temporally in the unified visual space. A trajectory-injected diffusion Transformer is further designed to denoise the noisy video latents for video generation with the guidance of ego-other vehicle trajectories. In addition, we propose a metric based on control latent similarity to evaluate the controllability of trajectories. Extensive experiments are conducted on the nuScenes dataset, and the proposed model outperforms the state-of-the-art method by 30\% in FID and 55\% in FVD. The model can also predict unseen driving scenes with self-produced trajectories.
We introduce Reangle-A-Video, a unified framework for generating synchronized multi-view videos from a single input video. Unlike mainstream approaches that train multi-view video diffusion models on large-scale 4D datasets, our method reframes the multi-view video generation task as video-to-videos translation, leveraging publicly available image and video diffusion priors. In essence, Reangle-A-Video operates in two stages. (1) Multi-View Motion Learning: An image-to-video diffusion transformer is synchronously fine-tuned in a self-supervised manner to distill view-invariant motion from a set of warped videos. (2) Multi-View Consistent Image-to-Images Translation: The first frame of the input video is warped and inpainted into various camera perspectives under an inference-time cross-view consistency guidance using DUSt3R, generating multi-view consistent starting images. Extensive experiments on static view transport and dynamic camera control show that Reangle-A-Video surpasses existing methods, establishing a new solution for multi-view video generation. We will publicly release our code and data. Project page: https://hyeonho99.github.io/reangle-a-video/
Simple as it seems, moving an object to another location within an image is, in fact, a challenging image-editing task that requires re-harmonizing the lighting, adjusting the pose based on perspective, accurately filling occluded regions, and ensuring coherent synchronization of shadows and reflections while maintaining the object identity. In this paper, we present ObjectMover, a generative model that can perform object movement in highly challenging scenes. Our key insight is that we model this task as a sequence-to-sequence problem and fine-tune a video generation model to leverage its knowledge of consistent object generation across video frames. We show that with this approach, our model is able to adjust to complex real-world scenarios, handling extreme lighting harmonization and object effect movement. As large-scale data for object movement are unavailable, we construct a data generation pipeline using a modern game engine to synthesize high-quality data pairs. We further propose a multi-task learning strategy that enables training on real-world video data to improve the model generalization. Through extensive experiments, we demonstrate that ObjectMover achieves outstanding results and adapts well to real-world scenarios.
Relational video customization refers to the creation of personalized videos that depict user-specified relations between two subjects, a crucial task for comprehending real-world visual content. While existing methods can personalize subject appearances and motions, they still struggle with complex relational video customization, where precise relational modeling and high generalization across subject categories are essential. The primary challenge arises from the intricate spatial arrangements, layout variations, and nuanced temporal dynamics inherent in relations; consequently, current models tend to overemphasize irrelevant visual details rather than capturing meaningful interactions. To address these challenges, we propose DreamRelation, a novel approach that personalizes relations through a small set of exemplar videos, leveraging two key components: Relational Decoupling Learning and Relational Dynamics Enhancement. First, in Relational Decoupling Learning, we disentangle relations from subject appearances using relation LoRA triplet and hybrid mask training strategy, ensuring better generalization across diverse relationships. Furthermore, we determine the optimal design of relation LoRA triplet by analyzing the distinct roles of the query, key, and value features within MM-DiT's attention mechanism, making DreamRelation the first relational video generation framework with explainable components. Second, in Relational Dynamics Enhancement, we introduce space-time relational contrastive loss, which prioritizes relational dynamics while minimizing the reliance on detailed subject appearances. Extensive experiments demonstrate that DreamRelation outperforms state-of-the-art methods in relational video customization. Code and models will be made publicly available.
Diffusion Transformer has demonstrated powerful capability and scalability in generating high-quality images and videos. Further pursuing the unification of generation and editing tasks has yielded significant progress in the domain of image content creation. However, due to the intrinsic demands for consistency across both temporal and spatial dynamics, achieving a unified approach for video synthesis remains challenging. We introduce VACE, which enables users to perform Video tasks within an All-in-one framework for Creation and Editing. These tasks include reference-to-video generation, video-to-video editing, and masked video-to-video editing. Specifically, we effectively integrate the requirements of various tasks by organizing video task inputs, such as editing, reference, and masking, into a unified interface referred to as the Video Condition Unit (VCU). Furthermore, by utilizing a Context Adapter structure, we inject different task concepts into the model using formalized representations of temporal and spatial dimensions, allowing it to handle arbitrary video synthesis tasks flexibly. Extensive experiments demonstrate that the unified model of VACE achieves performance on par with task-specific models across various subtasks. Simultaneously, it enables diverse applications through versatile task combinations. Project page: https://ali-vilab.github.io/VACE-Page/.
The task of video generation requires synthesizing visually realistic and temporally coherent video frames. Existing methods primarily use asynchronous auto-regressive models or synchronous diffusion models to address this challenge. However, asynchronous auto-regressive models often suffer from inconsistencies between training and inference, leading to issues such as error accumulation, while synchronous diffusion models are limited by their reliance on rigid sequence length. To address these issues, we introduce Auto-Regressive Diffusion (AR-Diffusion), a novel model that combines the strengths of auto-regressive and diffusion models for flexible, asynchronous video generation. Specifically, our approach leverages diffusion to gradually corrupt video frames in both training and inference, reducing the discrepancy between these phases. Inspired by auto-regressive generation, we incorporate a non-decreasing constraint on the corruption timesteps of individual frames, ensuring that earlier frames remain clearer than subsequent ones. This setup, together with temporal causal attention, enables flexible generation of videos with varying lengths while preserving temporal coherence. In addition, we design two specialized timestep schedulers: the FoPP scheduler for balanced timestep sampling during training, and the AD scheduler for flexible timestep differences during inference, supporting both synchronous and asynchronous generation. Extensive experiments demonstrate the superiority of our proposed method, which achieves competitive and state-of-the-art results across four challenging benchmarks.