Recent advances in video diffusion models have driven rapid progress in video editing techniques. However, video object removal, a critical subtask of video editing, remains challenging due to issues such as hallucinated objects and visual artifacts. Furthermore, existing methods often rely on computationally expensive sampling procedures and classifier-free guidance (CFG), resulting in slow inference. To address these limitations, we propose MiniMax-Remover, a novel two-stage video object removal approach. Motivated by the observation that text condition is not best suited for this task, we simplify the pretrained video generation model by removing textual input and cross-attention layers, resulting in a more lightweight and efficient model architecture in the first stage. In the second stage, we distilled our remover on successful videos produced by the stage-1 model and curated by human annotators, using a minimax optimization strategy to further improve editing quality and inference speed. Specifically, the inner maximization identifies adversarial input noise ("bad noise") that makes failure removals, while the outer minimization step trains the model to generate high-quality removal results even under such challenging conditions. As a result, our method achieves a state-of-the-art video object removal results with as few as 6 sampling steps and doesn't rely on CFG, significantly improving inference efficiency. Extensive experiments demonstrate the effectiveness and superiority of MiniMax-Remover compared to existing methods. Codes and Videos are available at: https://minimax-remover.github.io.
This paper presents DreamDance, a novel character art animation framework capable of producing stable, consistent character and scene motion conditioned on precise camera trajectories. To achieve this, we re-formulate the animation task as two inpainting-based steps: Camera-aware Scene Inpainting and Pose-aware Video Inpainting. The first step leverages a pre-trained image inpainting model to generate multi-view scene images from the reference art and optimizes a stable large-scale Gaussian field, which enables coarse background video rendering with camera trajectories. However, the rendered video is rough and only conveys scene motion. To resolve this, the second step trains a pose-aware video inpainting model that injects the dynamic character into the scene video while enhancing background quality. Specifically, this model is a DiT-based video generation model with a gating strategy that adaptively integrates the character's appearance and pose information into the base background video. Through extensive experiments, we demonstrate the effectiveness and generalizability of DreamDance, producing high-quality and consistent character animations with remarkable camera dynamics.
Recently, methods leveraging diffusion model priors to assist monocular geometric estimation (e.g., depth and normal) have gained significant attention due to their strong generalization ability. However, most existing works focus on estimating geometric properties within the camera coordinate system of individual video frames, neglecting the inherent ability of diffusion models to determine inter-frame correspondence. In this work, we demonstrate that, through appropriate design and fine-tuning, the intrinsic consistency of video generation models can be effectively harnessed for consistent geometric estimation. Specifically, we 1) select geometric attributes in the global coordinate system that share the same correspondence with video frames as the prediction targets, 2) introduce a novel and efficient conditioning method by reusing positional encodings, and 3) enhance performance through joint training on multiple geometric attributes that share the same correspondence. Our results achieve superior performance in predicting global geometric attributes in videos and can be directly applied to reconstruction tasks. Even when trained solely on static video data, our approach exhibits the potential to generalize to dynamic video scenes.
Text-conditioned diffusion models have emerged as powerful tools for high-quality video generation. However, enabling Interactive Video Generation (IVG), where users control motion elements such as object trajectory, remains challenging. Recent training-free approaches introduce attention masking to guide trajectory, but this often degrades perceptual quality. We identify two key failure modes in these methods, both of which we interpret as domain shift problems, and propose solutions inspired by domain adaptation. First, we attribute the perceptual degradation to internal covariate shift induced by attention masking, as pretrained models are not trained to handle masked attention. To address this, we propose mask normalization, a pre-normalization layer designed to mitigate this shift via distribution matching. Second, we address initialization gap, where the randomly sampled initial noise does not align with IVG conditioning, by introducing a temporal intrinsic diffusion prior that enforces spatio-temporal consistency at each denoising step. Extensive qualitative and quantitative evaluations demonstrate that mask normalization and temporal intrinsic denoising improve both perceptual quality and trajectory control over the existing state-of-the-art IVG techniques.
Diffusion models (DMs) have demonstrated remarkable performance in high-fidelity image and video generation. Because high-quality generations with DMs typically require a large number of function evaluations (NFEs), resulting in slow sampling, there has been extensive research successfully reducing the NFE to a small range (<10) while maintaining acceptable image quality. However, many practical applications, such as those involving Stable Diffusion 3.5, FLUX, and SANA, commonly operate in the mid-NFE regime (20-50 NFE) to achieve superior results, and, despite the practical relevance, research on the effective sampling within this mid-NFE regime remains underexplored. In this work, we propose a novel, training-free, and structure-independent DM ODE solver called the Stabilized Taylor Orthogonal Runge--Kutta (STORK) method, based on a class of stiff ODE solvers with a Taylor expansion adaptation. Unlike prior work such as DPM-Solver, which is dependent on the semi-linear structure of the DM ODE, STORK is applicable to any DM sampling, including noise-based and flow matching-based models. Within the 20-50 NFE range, STORK achieves improved generation quality, as measured by FID scores, across unconditional pixel-level generation and conditional latent-space generation tasks using models like Stable Diffusion 3.5 and SANA. Code is available at https://github.com/ZT220501/STORK.
Audio-driven human animation methods, such as talking head and talking body generation, have made remarkable progress in generating synchronized facial movements and appealing visual quality videos. However, existing methods primarily focus on single human animation and struggle with multi-stream audio inputs, facing incorrect binding problems between audio and persons. Additionally, they exhibit limitations in instruction-following capabilities. To solve this problem, in this paper, we propose a novel task: Multi-Person Conversational Video Generation, and introduce a new framework, MultiTalk, to address the challenges during multi-person generation. Specifically, for audio injection, we investigate several schemes and propose the Label Rotary Position Embedding (L-RoPE) method to resolve the audio and person binding problem. Furthermore, during training, we observe that partial parameter training and multi-task training are crucial for preserving the instruction-following ability of the base model. MultiTalk achieves superior performance compared to other methods on several datasets, including talking head, talking body, and multi-person datasets, demonstrating the powerful generation capabilities of our approach.
Recent advances in text-to-video generation, particularly with autoregressive models, have enabled the synthesis of high-quality videos depicting individual scenes. However, extending these models to generate long, cross-scene videos remains a significant challenge. As the context length grows during autoregressive decoding, computational costs rise sharply, and the model's ability to maintain consistency and adhere to evolving textual prompts deteriorates. We introduce InfLVG, an inference-time framework that enables coherent long video generation without requiring additional long-form video data. InfLVG leverages a learnable context selection policy, optimized via Group Relative Policy Optimization (GRPO), to dynamically identify and retain the most semantically relevant context throughout the generation process. Instead of accumulating the entire generation history, the policy ranks and selects the top-$K$ most contextually relevant tokens, allowing the model to maintain a fixed computational budget while preserving content consistency and prompt alignment. To optimize the policy, we design a hybrid reward function that jointly captures semantic alignment, cross-scene consistency, and artifact reduction. To benchmark performance, we introduce the Cross-scene Video Benchmark (CsVBench) along with an Event Prompt Set (EPS) that simulates complex multi-scene transitions involving shared subjects and varied actions/backgrounds. Experimental results show that InfLVG can extend video length by up to 9$\times$, achieving strong consistency and semantic fidelity across scenes. Our code is available at https://github.com/MAPLE-AIGC/InfLVG.
Generating engaging, accurate short-form videos from scientific papers is challenging due to content complexity and the gap between expert authors and readers. Existing end-to-end methods often suffer from factual inaccuracies and visual artifacts, limiting their utility for scientific dissemination. To address these issues, we propose SciTalk, a novel multi-LLM agentic framework, grounding videos in various sources, such as text, figures, visual styles, and avatars. Inspired by content creators' workflows, SciTalk uses specialized agents for content summarization, visual scene planning, and text and layout editing, and incorporates an iterative feedback mechanism where video agents simulate user roles to give feedback on generated videos from previous iterations and refine generation prompts. Experimental evaluations show that SciTalk outperforms simple prompting methods in generating scientifically accurate and engaging content over the refined loop of video generation. Although preliminary results are still not yet matching human creators' quality, our framework provides valuable insights into the challenges and benefits of feedback-driven video generation. Our code, data, and generated videos will be publicly available.
Video generation has made substantial strides with the emergence of deep generative models, especially diffusion-based approaches. However, video generation based on multiple reference subjects still faces significant challenges in maintaining multi-subject consistency and ensuring high generation quality. In this paper, we propose MAGREF, a unified framework for any-reference video generation that introduces masked guidance to enable coherent multi-subject video synthesis conditioned on diverse reference images and a textual prompt. Specifically, we propose (1) a region-aware dynamic masking mechanism that enables a single model to flexibly handle various subject inference, including humans, objects, and backgrounds, without architectural changes, and (2) a pixel-wise channel concatenation mechanism that operates on the channel dimension to better preserve appearance features. Our model delivers state-of-the-art video generation quality, generalizing from single-subject training to complex multi-subject scenarios with coherent synthesis and precise control over individual subjects, outperforming existing open-source and commercial baselines. To facilitate evaluation, we also introduce a comprehensive multi-subject video benchmark. Extensive experiments demonstrate the effectiveness of our approach, paving the way for scalable, controllable, and high-fidelity multi-subject video synthesis. Code and model can be found at: https://github.com/MAGREF-Video/MAGREF
MLLMs have been widely studied for video question answering recently. However, most existing assessments focus on natural videos, overlooking synthetic videos, such as AI-generated content (AIGC). Meanwhile, some works in video generation rely on MLLMs to evaluate the quality of generated videos, but the capabilities of MLLMs on interpreting AIGC videos remain largely underexplored. To address this, we propose a new benchmark, VF-Eval, which introduces four tasks-coherence validation, error awareness, error type detection, and reasoning evaluation-to comprehensively evaluate the abilities of MLLMs on AIGC videos. We evaluate 13 frontier MLLMs on VF-Eval and find that even the best-performing model, GPT-4.1, struggles to achieve consistently good performance across all tasks. This highlights the challenging nature of our benchmark. Additionally, to investigate the practical applications of VF-Eval in improving video generation, we conduct an experiment, RePrompt, demonstrating that aligning MLLMs more closely with human feedback can benefit video generation.
Recent advancements in text-to-video (T2V) diffusion models have enabled high-fidelity and realistic video synthesis. However, current T2V models often struggle to generate physically plausible content due to their limited inherent ability to accurately understand physics. We found that while the representations within T2V models possess some capacity for physics understanding, they lag significantly behind those from recent video self-supervised learning methods. To this end, we propose a novel framework called VideoREPA, which distills physics understanding capability from video understanding foundation models into T2V models by aligning token-level relations. This closes the physics understanding gap and enable more physics-plausible generation. Specifically, we introduce the Token Relation Distillation (TRD) loss, leveraging spatio-temporal alignment to provide soft guidance suitable for finetuning powerful pre-trained T2V models, a critical departure from prior representation alignment (REPA) methods. To our knowledge, VideoREPA is the first REPA method designed for finetuning T2V models and specifically for injecting physical knowledge. Empirical evaluations show that VideoREPA substantially enhances the physics commonsense of baseline method, CogVideoX, achieving significant improvement on relevant benchmarks and demonstrating a strong capacity for generating videos consistent with intuitive physics. More video results are available at https://videorepa.github.io/.
Video captions play a crucial role in text-to-video generation tasks, as their quality directly influences the semantic coherence and visual fidelity of the generated videos. Although large vision-language models (VLMs) have demonstrated significant potential in caption generation, existing benchmarks inadequately address fine-grained evaluation, particularly in capturing spatial-temporal details critical for video generation. To address this gap, we introduce the Fine-grained Video Caption Evaluation Benchmark (VCapsBench), the first large-scale fine-grained benchmark comprising 5,677 (5K+) videos and 109,796 (100K+) question-answer pairs. These QA-pairs are systematically annotated across 21 fine-grained dimensions (e.g., camera movement, and shot type) that are empirically proven critical for text-to-video generation. We further introduce three metrics (Accuracy (AR), Inconsistency Rate (IR), Coverage Rate (CR)), and an automated evaluation pipeline leveraging large language model (LLM) to verify caption quality via contrastive QA-pairs analysis. By providing actionable insights for caption optimization, our benchmark can advance the development of robust text-to-video models. The dataset and codes are available at website: https://github.com/GXYM/VCapsBench.
Diffusion Transformer (DiT)-based video diffusion models generate high-quality videos at scale but incur prohibitive processing latency and memory costs for long videos. To address this, we propose a novel distributed inference strategy, termed DualParal. The core idea is that, instead of generating an entire video on a single GPU, we parallelize both temporal frames and model layers across GPUs. However, a naive implementation of this division faces a key limitation: since diffusion models require synchronized noise levels across frames, this implementation leads to the serialization of original parallelisms. We leverage a block-wise denoising scheme to handle this. Namely, we process a sequence of frame blocks through the pipeline with progressively decreasing noise levels. Each GPU handles a specific block and layer subset while passing previous results to the next GPU, enabling asynchronous computation and communication. To further optimize performance, we incorporate two key enhancements. Firstly, a feature cache is implemented on each GPU to store and reuse features from the prior block as context, minimizing inter-GPU communication and redundant computation. Secondly, we employ a coordinated noise initialization strategy, ensuring globally consistent temporal dynamics by sharing initial noise patterns across GPUs without extra resource costs. Together, these enable fast, artifact-free, and infinitely long video generation. Applied to the latest diffusion transformer video generator, our method efficiently produces 1,025-frame videos with up to 6.54$\times$ lower latency and 1.48$\times$ lower memory cost on 8$\times$RTX 4090 GPUs.
Imitation Learning has become a fundamental approach in robotic manipulation. However, collecting large-scale real-world robot demonstrations is prohibitively expensive. Simulators offer a cost-effective alternative, but the sim-to-real gap make it extremely challenging to scale. Therefore, we introduce RoboTransfer, a diffusion-based video generation framework for robotic data synthesis. Unlike previous methods, RoboTransfer integrates multi-view geometry with explicit control over scene components, such as background and object attributes. By incorporating cross-view feature interactions and global depth/normal conditions, RoboTransfer ensures geometry consistency across views. This framework allows fine-grained control, including background edits and object swaps. Experiments demonstrate that RoboTransfer is capable of generating multi-view videos with enhanced geometric consistency and visual fidelity. In addition, policies trained on the data generated by RoboTransfer achieve a 33.3% relative improvement in the success rate in the DIFF-OBJ setting and a substantial 251% relative improvement in the more challenging DIFF-ALL scenario. Explore more demos on our project page: https://horizonrobotics.github.io/robot_lab/robotransfer
Co-Speech Gesture Video Generation aims to generate vivid speech videos from audio-driven still images, which is challenging due to the diversity of different parts of the body in terms of amplitude of motion, audio relevance, and detailed features. Relying solely on audio as the control signal often fails to capture large gesture movements in video, leading to more pronounced artifacts and distortions. Existing approaches typically address this issue by introducing additional a priori information, but this can limit the practical application of the task. Specifically, we propose a Motion Mask-Guided Two-Stage Network (MMGT) that uses audio, as well as motion masks and motion features generated from the audio signal to jointly drive the generation of synchronized speech gesture videos. In the first stage, the Spatial Mask-Guided Audio Pose Generation (SMGA) Network generates high-quality pose videos and motion masks from audio, effectively capturing large movements in key regions such as the face and gestures. In the second stage, we integrate the Motion Masked Hierarchical Audio Attention (MM-HAA) into the Stabilized Diffusion Video Generation model, overcoming limitations in fine-grained motion generation and region-specific detail control found in traditional methods. This guarantees high-quality, detailed upper-body video generation with accurate texture and motion details. Evaluations show improved video quality, lip-sync, and gesture. The model and code are available at https://github.com/SIA-IDE/MMGT.
Recent advances in diffusion-based text-to-video (T2V) models have demonstrated remarkable progress, but these models still face challenges in generating videos with multiple objects. Most models struggle with accurately capturing complex object interactions, often treating some objects as static background elements and limiting their movement. In addition, they often fail to generate multiple distinct objects as specified in the prompt, resulting in incorrect generations or mixed features across objects. In this paper, we present a novel training-free approach for multi-object video generation that leverages the open world knowledge of diffusion models and large language models (LLMs). We use an LLM as the ``director'' of object trajectories, and apply the trajectories through noise re-initialization to achieve precise control of realistic movements. We further refine the generation process by manipulating the attention mechanism to better capture object-specific features and motion patterns, and prevent cross-object feature interference. Extensive experiments validate the effectiveness of our training free approach in significantly enhancing the multi-object generation capabilities of existing video diffusion models, resulting in 42% absolute improvement in motion dynamics and object generation accuracy, while also maintaining high fidelity and motion smoothness.
We propose a unified framework for motion control in video generation that seamlessly integrates camera movement, object-level translation, and fine-grained local motion using trajectory-based inputs. In contrast to prior methods that address these motion types through separate modules or task-specific designs, our approach offers a cohesive solution by projecting user-defined trajectories into the latent space of pre-trained image-to-video generation models via a lightweight motion injector. Users can specify keypoints and their motion paths to control localized deformations, entire object motion, virtual camera dynamics, or combinations of these. The injected trajectory signals guide the generative process to produce temporally consistent and semantically aligned motion sequences. Our framework demonstrates superior performance across multiple video motion control tasks, including stylized motion effects (e.g., motion brushes), dynamic viewpoint changes, and precise local motion manipulation. Experiments show that our method provides significantly better controllability and visual quality compared to prior approaches and commercial solutions, while remaining broadly compatible with various state-of-the-art video generation backbones. Project page: https://anytraj.github.io/.
Estimating accurate and temporally consistent 3D human geometry from videos is a challenging problem in computer vision. Existing methods, primarily optimized for single images, often suffer from temporal inconsistencies and fail to capture fine-grained dynamic details. To address these limitations, we present GeoMan, a novel architecture designed to produce accurate and temporally consistent depth and normal estimations from monocular human videos. GeoMan addresses two key challenges: the scarcity of high-quality 4D training data and the need for metric depth estimation to accurately model human size. To overcome the first challenge, GeoMan employs an image-based model to estimate depth and normals for the first frame of a video, which then conditions a video diffusion model, reframing video geometry estimation task as an image-to-video generation problem. This design offloads the heavy lifting of geometric estimation to the image model and simplifies the video model's role to focus on intricate details while using priors learned from large-scale video datasets. Consequently, GeoMan improves temporal consistency and generalizability while requiring minimal 4D training data. To address the challenge of accurate human size estimation, we introduce a root-relative depth representation that retains critical human-scale details and is easier to be estimated from monocular inputs, overcoming the limitations of traditional affine-invariant and metric depth representations. GeoMan achieves state-of-the-art performance in both qualitative and quantitative evaluations, demonstrating its effectiveness in overcoming longstanding challenges in 3D human geometry estimation from videos.
Recent advances in diffusion models have significantly improved conditional video generation, particularly in the pose-guided human image animation task. Although existing methods are capable of generating high-fidelity and time-consistent animation sequences in regular motions and static scenes, there are still obvious limitations when facing complex human body motions (Hypermotion) that contain highly dynamic, non-standard motions, and the lack of a high-quality benchmark for evaluation of complex human motion animations. To address this challenge, we introduce the \textbf{Open-HyperMotionX Dataset} and \textbf{HyperMotionX Bench}, which provide high-quality human pose annotations and curated video clips for evaluating and improving pose-guided human image animation models under complex human motion conditions. Furthermore, we propose a simple yet powerful DiT-based video generation baseline and design spatial low-frequency enhanced RoPE, a novel module that selectively enhances low-frequency spatial feature modeling by introducing learnable frequency scaling. Our method significantly improves structural stability and appearance consistency in highly dynamic human motion sequences. Extensive experiments demonstrate the effectiveness of our dataset and proposed approach in advancing the generation quality of complex human motion image animations. Code and dataset will be made publicly available.
Token merging can effectively accelerate various vision systems by processing groups of similar tokens only once and sharing the results across them. However, existing token grouping methods are often ad hoc and random, disregarding the actual content of the samples. We show that preserving high-information tokens during merging - those essential for semantic fidelity and structural details - significantly improves sample quality, producing finer details and more coherent, realistic generations. Despite being simple and intuitive, this approach remains underexplored. To do so, we propose an importance-based token merging method that prioritizes the most critical tokens in computational resource allocation, leveraging readily available importance scores, such as those from classifier-free guidance in diffusion models. Experiments show that our approach significantly outperforms baseline methods across multiple applications, including text-to-image synthesis, multi-view image generation, and video generation with various model architectures such as Stable Diffusion, Zero123++, AnimateDiff, or PixArt-$\alpha$.