Audio is inherently temporal and closely synchronized with the visual world, making it a naturally aligned and expressive control signal for controllable video generation (e.g., movies). Beyond control, directly translating audio into video is essential for understanding and visualizing rich audio narratives (e.g., Podcasts or historical recordings). However, existing approaches fall short in generating high-quality videos with precise audio-visual synchronization, especially across diverse and complex audio types. In this work, we introduce MTV, a versatile framework for audio-sync video generation. MTV explicitly separates audios into speech, effects, and music tracks, enabling disentangled control over lip motion, event timing, and visual mood, respectively -- resulting in fine-grained and semantically aligned video generation. To support the framework, we additionally present DEMIX, a dataset comprising high-quality cinematic videos and demixed audio tracks. DEMIX is structured into five overlapped subsets, enabling scalable multi-stage training for diverse generation scenarios. Extensive experiments demonstrate that MTV achieves state-of-the-art performance across six standard metrics spanning video quality, text-video consistency, and audio-video alignment. Project page: https://hjzheng.net/projects/MTV/.
The parameter-efficient adaptation of the image-text pretraining model CLIP for video-text retrieval is a prominent area of research. While CLIP is focused on image-level vision-language matching, video-text retrieval demands comprehensive understanding at the video level. Three key discrepancies emerge in the transfer from image-level to video-level: vision, language, and alignment. However, existing methods mainly focus on vision while neglecting language and alignment. In this paper, we propose Discrepancy Reduction in Vision, Language, and Alignment (DiscoVLA), which simultaneously mitigates all three discrepancies. Specifically, we introduce Image-Video Features Fusion to integrate image-level and video-level features, effectively tackling both vision and language discrepancies. Additionally, we generate pseudo image captions to learn fine-grained image-level alignment. To mitigate alignment discrepancies, we propose Image-to-Video Alignment Distillation, which leverages image-level alignment knowledge to enhance video-level alignment. Extensive experiments demonstrate the superiority of our DiscoVLA. In particular, on MSRVTT with CLIP (ViT-B/16), DiscoVLA outperforms previous methods by 1.5% in R@1, reaching a final score of 50.5% R@1. The code is available at https://github.com/LunarShen/DsicoVLA.
Generating highly dynamic and photorealistic portrait animations driven by audio and skeletal motion remains challenging due to the need for precise lip synchronization, natural facial expressions, and high-fidelity body motion dynamics. We propose a human-preference-aligned diffusion framework that addresses these challenges through two key innovations. First, we introduce direct preference optimization tailored for human-centric animation, leveraging a curated dataset of human preferences to align generated outputs with perceptual metrics for portrait motion-video alignment and naturalness of expression. Second, the proposed temporal motion modulation resolves spatiotemporal resolution mismatches by reshaping motion conditions into dimensionally aligned latent features through temporal channel redistribution and proportional feature expansion, preserving the fidelity of high-frequency motion details in diffusion-based synthesis. The proposed mechanism is complementary to existing UNet and DiT-based portrait diffusion approaches, and experiments demonstrate obvious improvements in lip-audio synchronization, expression vividness, body motion coherence over baseline methods, alongside notable gains in human preference metrics. Our model and source code can be found at: https://github.com/xyz123xyz456/hallo4.
Recent advancements in large multimodal models (LMMs) have driven substantial progress in both text-to-video (T2V) generation and video-to-text (V2T) interpretation tasks. However, current AI-generated videos (AIGVs) still exhibit limitations in terms of perceptual quality and text-video alignment. Therefore, a reliable and scalable automatic model for AIGV evaluation is desirable, which heavily relies on the scale and quality of human annotations. To this end, we present AIGVE-60K, a comprehensive dataset and benchmark for AI-Generated Video Evaluation, which features (i) comprehensive tasks, encompassing 3,050 extensive prompts across 20 fine-grained task dimensions, (ii) the largest human annotations, including 120K mean-opinion scores (MOSs) and 60K question-answering (QA) pairs annotated on 58,500 videos generated from 30 T2V models, and (iii) bidirectional benchmarking and evaluating for both T2V generation and V2T interpretation capabilities. Based on AIGVE-60K, we propose LOVE, a LMM-based metric for AIGV Evaluation from multiple dimensions including perceptual preference, text-video correspondence, and task-specific accuracy in terms of both instance level and model level. Comprehensive experiments demonstrate that LOVE not only achieves state-of-the-art performance on the AIGVE-60K dataset, but also generalizes effectively to a wide range of other AIGV evaluation benchmarks. These findings highlight the significance of the AIGVE-60K dataset. Database and codes are anonymously available at https://github.com/IntMeGroup/LOVE.
Video generation based on diffusion models presents a challenging multimodal task, with video editing emerging as a pivotal direction in this field. Recent video editing approaches primarily fall into two categories: training-required and training-free methods. While training-based methods incur high computational costs, training-free alternatives often yield suboptimal performance. To address these limitations, we propose DAPE, a high-quality yet cost-effective two-stage parameter-efficient fine-tuning (PEFT) framework for video editing. In the first stage, we design an efficient norm-tuning method to enhance temporal consistency in generated videos. The second stage introduces a vision-friendly adapter to improve visual quality. Additionally, we identify critical shortcomings in existing benchmarks, including limited category diversity, imbalanced object distribution, and inconsistent frame counts. To mitigate these issues, we curate a large dataset benchmark comprising 232 videos with rich annotations and 6 editing prompts, enabling objective and comprehensive evaluation of advanced methods. Extensive experiments on existing datasets (BalanceCC, LOVEU-TGVE, RAVE) and our proposed benchmark demonstrate that DAPE significantly improves temporal coherence and text-video alignment while outperforming previous state-of-the-art approaches.
Customized video generation aims to produce videos featuring specific subjects under flexible user-defined conditions, yet existing methods often struggle with identity consistency and limited input modalities. In this paper, we propose HunyuanCustom, a multi-modal customized video generation framework that emphasizes subject consistency while supporting image, audio, video, and text conditions. Built upon HunyuanVideo, our model first addresses the image-text conditioned generation task by introducing a text-image fusion module based on LLaVA for enhanced multi-modal understanding, along with an image ID enhancement module that leverages temporal concatenation to reinforce identity features across frames. To enable audio- and video-conditioned generation, we further propose modality-specific condition injection mechanisms: an AudioNet module that achieves hierarchical alignment via spatial cross-attention, and a video-driven injection module that integrates latent-compressed conditional video through a patchify-based feature-alignment network. Extensive experiments on single- and multi-subject scenarios demonstrate that HunyuanCustom significantly outperforms state-of-the-art open- and closed-source methods in terms of ID consistency, realism, and text-video alignment. Moreover, we validate its robustness across downstream tasks, including audio and video-driven customized video generation. Our results highlight the effectiveness of multi-modal conditioning and identity-preserving strategies in advancing controllable video generation. All the code and models are available at https://hunyuancustom.github.io.
Compositional text-to-video generation, which requires synthesizing dynamic scenes with multiple interacting entities and precise spatial-temporal relationships, remains a critical challenge for diffusion-based models. Existing methods struggle with layout discontinuity, entity identity drift, and implausible interaction dynamics due to unconstrained cross-attention mechanisms and inadequate physics-aware reasoning. To address these limitations, we propose DyST-XL, a \textbf{training-free} framework that enhances off-the-shelf text-to-video models (e.g., CogVideoX-5B) through frame-aware control. DyST-XL integrates three key innovations: (1) A Dynamic Layout Planner that leverages large language models (LLMs) to parse input prompts into entity-attribute graphs and generates physics-aware keyframe layouts, with intermediate frames interpolated via trajectory optimization; (2) A Dual-Prompt Controlled Attention Mechanism that enforces localized text-video alignment through frame-aware attention masking, achieving the precise control over individual entities; and (3) An Entity-Consistency Constraint strategy that propagates first-frame feature embeddings to subsequent frames during denoising, preserving object identity without manual annotation. Experiments demonstrate that DyST-XL excels in compositional text-to-video generation, significantly improving performance on complex prompts and bridging a crucial gap in training-free video synthesis.
The advancement of 4D (i.e., sequential 3D) generation opens up new possibilities for lifelike experiences in various applications, where users can explore dynamic objects or characters from any viewpoint. Meanwhile, video generative models are receiving particular attention given their ability to produce realistic and imaginative frames. These models are also observed to exhibit strong 3D consistency, indicating the potential to act as world simulators. In this work, we present Video4DGen, a novel framework that excels in generating 4D representations from single or multiple generated videos as well as generating 4D-guided videos. This framework is pivotal for creating high-fidelity virtual contents that maintain both spatial and temporal coherence. The 4D outputs generated by Video4DGen are represented using our proposed Dynamic Gaussian Surfels (DGS), which optimizes time-varying warping functions to transform Gaussian surfels (surface elements) from a static state to a dynamically warped state. We design warped-state geometric regularization and refinements on Gaussian surfels, to preserve the structural integrity and fine-grained appearance details. To perform 4D generation from multiple videos and capture representation across spatial, temporal, and pose dimensions, we design multi-video alignment, root pose optimization, and pose-guided frame sampling strategies. The leveraging of continuous warping fields also enables a precise depiction of pose, motion, and deformation over per-video frames. Further, to improve the overall fidelity from the observation of all camera poses, Video4DGen performs novel-view video generation guided by the 4D content, with the proposed confidence-filtered DGS to enhance the quality of generated sequences. With the ability of 4D and video generation, Video4DGen offers a powerful tool for applications in virtual reality, animation, and beyond.
We propose a new task, video referring matting, which obtains the alpha matte of a specified instance by inputting a referring caption. We treat the dense prediction task of matting as video generation, leveraging the text-to-video alignment prior of video diffusion models to generate alpha mattes that are temporally coherent and closely related to the corresponding semantic instances. Moreover, we propose a new Latent-Constructive loss to further distinguish different instances, enabling more controllable interactive matting. Additionally, we introduce a large-scale video referring matting dataset with 10,000 videos. To the best of our knowledge, this is the first dataset that concurrently contains captions, videos, and instance-level alpha mattes. Extensive experiments demonstrate the effectiveness of our method. The dataset and code are available at https://github.com/Hansxsourse/VRMDiff.
The objective of this work is to align asynchronous subtitles in sign language videos with limited labelled data. To achieve this goal, we propose a novel framework with the following contributions: (1) we leverage fundamental grammatical rules of British Sign Language (BSL) to pre-process the input subtitles, (2) we design a selective alignment loss to optimise the model for predicting the temporal location of signs only when the queried sign actually occurs in a scene, and (3) we conduct self-training with refined pseudo-labels which are more accurate than the heuristic audio-aligned labels. From this, our model not only better understands the correlation between the text and the signs, but also holds potential for application in the translation of sign languages, particularly in scenarios where manual labelling of large-scale sign data is impractical or challenging. Extensive experimental results demonstrate that our approach achieves state-of-the-art results, surpassing previous baselines by substantial margins in terms of both frame-level accuracy and F1-score. This highlights the effectiveness and practicality of our framework in advancing the field of sign language video alignment and translation.
The remarkable progress in text-to-video diffusion models enables photorealistic generations, although the contents of the generated video often include unnatural movement or deformation, reverse playback, and motionless scenes. Recently, an alignment problem has attracted huge attention, where we steer the output of diffusion models based on some quantity on the goodness of the content. Because there is a large room for improvement of perceptual quality along the frame direction, we should address which metrics we should optimize and how we can optimize them in the video generation. In this paper, we propose diffusion latent beam search with lookahead estimator, which can select better diffusion latent to maximize a given alignment reward, at inference time. We then point out that the improvement of perceptual video quality considering the alignment to prompts requires reward calibration by weighting existing metrics. When evaluating outputs by using vision language models as a proxy of humans, many previous metrics to quantify the naturalness of video do not always correlate with evaluation and also depend on the degree of dynamic descriptions in evaluation prompts. We demonstrate that our method improves the perceptual quality based on the calibrated reward, without model parameter update, and outputs the best generation compared to greedy search and best-of-N sampling. We provide practical guidelines on which axes, among search budget, lookahead steps for reward estimate, and denoising steps, in the reverse diffusion process, we should allocate the inference-time computation.
The video-to-audio (V2A) generation task has drawn attention in the field of multimedia due to the practicality in producing Foley sound. Semantic and temporal conditions are fed to the generation model to indicate sound events and temporal occurrence. Recent studies on synthesizing immersive and synchronized audio are faced with challenges on videos with moving visual presence. The temporal condition is not accurate enough while low-resolution semantic condition exacerbates the problem. To tackle these challenges, we propose Smooth-Foley, a V2A generative model taking semantic guidance from the textual label across the generation to enhance both semantic and temporal alignment in audio. Two adapters are trained to leverage pre-trained text-to-audio generation models. A frame adapter integrates high-resolution frame-wise video features while a temporal adapter integrates temporal conditions obtained from similarities of visual frames and textual labels. The incorporation of semantic guidance from textual labels achieves precise audio-video alignment. We conduct extensive quantitative and qualitative experiments. Results show that Smooth-Foley performs better than existing models on both continuous sound scenarios and general scenarios. With semantic guidance, the audio generated by Smooth-Foley exhibits higher quality and better adherence to physical laws.
Recent advancements in video generation have significantly impacted daily life for both individuals and industries. However, the leading video generation models remain closed-source, resulting in a notable performance gap between industry capabilities and those available to the public. In this report, we introduce HunyuanVideo, an innovative open-source video foundation model that demonstrates performance in video generation comparable to, or even surpassing, that of leading closed-source models. HunyuanVideo encompasses a comprehensive framework that integrates several key elements, including data curation, advanced architectural design, progressive model scaling and training, and an efficient infrastructure tailored for large-scale model training and inference. As a result, we successfully trained a video generative model with over 13 billion parameters, making it the largest among all open-source models. We conducted extensive experiments and implemented a series of targeted designs to ensure high visual quality, motion dynamics, text-video alignment, and advanced filming techniques. According to evaluations by professionals, HunyuanVideo outperforms previous state-of-the-art models, including Runway Gen-3, Luma 1.6, and three top-performing Chinese video generative models. By releasing the code for the foundation model and its applications, we aim to bridge the gap between closed-source and open-source communities. This initiative will empower individuals within the community to experiment with their ideas, fostering a more dynamic and vibrant video generation ecosystem. The code is publicly available at https://github.com/Tencent/HunyuanVideo.
Large text-to-video models hold immense potential for a wide range of downstream applications. However, these models struggle to accurately depict dynamic object interactions, often resulting in unrealistic movements and frequent violations of real-world physics. One solution inspired by large language models is to align generated outputs with desired outcomes using external feedback. This enables the model to refine its responses autonomously, eliminating extensive manual data collection. In this work, we investigate the use of feedback to enhance the object dynamics in text-to-video models. We aim to answer a critical question: what types of feedback, paired with which specific self-improvement algorithms, can most effectively improve text-video alignment and realistic object interactions? We begin by deriving a unified probabilistic objective for offline RL finetuning of text-to-video models. This perspective highlights how design elements in existing algorithms like KL regularization and policy projection emerge as specific choices within a unified framework. We then use derived methods to optimize a set of text-video alignment metrics (e.g., CLIP scores, optical flow), but notice that they often fail to align with human perceptions of generation quality. To address this limitation, we propose leveraging vision-language models to provide more nuanced feedback specifically tailored to object dynamics in videos. Our experiments demonstrate that our method can effectively optimize a wide variety of rewards, with binary AI feedback driving the most significant improvements in video quality for dynamic interactions, as confirmed by both AI and human evaluations. Notably, we observe substantial gains when using reward signals derived from AI feedback, particularly in scenarios involving complex interactions between multiple objects and realistic depictions of objects falling.
Recent advancements in text-to-video models such as Sora, Gen-3, MovieGen, and CogVideoX are pushing the boundaries of synthetic video generation, with adoption seen in fields like robotics, autonomous driving, and entertainment. As these models become prevalent, various metrics and benchmarks have emerged to evaluate the quality of the generated videos. However, these metrics emphasize visual quality and smoothness, neglecting temporal fidelity and text-to-video alignment, which are crucial for safety-critical applications. To address this gap, we introduce NeuS-V, a novel synthetic video evaluation metric that rigorously assesses text-to-video alignment using neuro-symbolic formal verification techniques. Our approach first converts the prompt into a formally defined Temporal Logic (TL) specification and translates the generated video into an automaton representation. Then, it evaluates the text-to-video alignment by formally checking the video automaton against the TL specification. Furthermore, we present a dataset of temporally extended prompts to evaluate state-of-the-art video generation models against our benchmark. We find that NeuS-V demonstrates a higher correlation by over 5x with human evaluations when compared to existing metrics. Our evaluation further reveals that current video generation models perform poorly on these temporally complex prompts, highlighting the need for future work in improving text-to-video generation capabilities.
Recent text-to-video (T2V) diffusion models have demonstrated impressive generation capabilities across various domains. However, these models often generate videos that have misalignments with text prompts, especially when the prompts describe complex scenes with multiple objects and attributes. To address this, we introduce VideoRepair, a novel model-agnostic, training-free video refinement framework that automatically identifies fine-grained text-video misalignments and generates explicit spatial and textual feedback, enabling a T2V diffusion model to perform targeted, localized refinements. VideoRepair consists of four stages: In (1) video evaluation, we detect misalignments by generating fine-grained evaluation questions and answering those questions with MLLM. In (2) refinement planning, we identify accurately generated objects and then create localized prompts to refine other areas in the video. Next, in (3) region decomposition, we segment the correctly generated area using a combined grounding module. We regenerate the video by adjusting the misaligned regions while preserving the correct regions in (4) localized refinement. On two popular video generation benchmarks (EvalCrafter and T2V-CompBench), VideoRepair substantially outperforms recent baselines across various text-video alignment metrics. We provide a comprehensive analysis of VideoRepair components and qualitative examples.
As visual generation technologies continue to advance, the scale of video datasets has expanded rapidly, and the quality of these datasets is critical to the performance of video generation models. We argue that temporal splitting, detailed captions, and video quality filtering are three key factors that determine dataset quality. However, existing datasets exhibit various limitations in these areas. To address these challenges, we introduce Koala-36M, a large-scale, high-quality video dataset featuring accurate temporal splitting, detailed captions, and superior video quality. The core of our approach lies in improving the consistency between fine-grained conditions and video content. Specifically, we employ a linear classifier on probability distributions to enhance the accuracy of transition detection, ensuring better temporal consistency. We then provide structured captions for the splitted videos, with an average length of 200 words, to improve text-video alignment. Additionally, we develop a Video Training Suitability Score (VTSS) that integrates multiple sub-metrics, allowing us to filter high-quality videos from the original corpus. Finally, we incorporate several metrics into the training process of the generation model, further refining the fine-grained conditions. Our experiments demonstrate the effectiveness of our data processing pipeline and the quality of the proposed Koala-36M dataset. Our dataset and code will be released at https://koala36m.github.io/.
In this paper, we focus on enhancing a diffusion-based text-to-video (T2V) model during the post-training phase by distilling a highly capable consistency model from a pretrained T2V model. Our proposed method, T2V-Turbo-v2, introduces a significant advancement by integrating various supervision signals, including high-quality training data, reward model feedback, and conditional guidance, into the consistency distillation process. Through comprehensive ablation studies, we highlight the crucial importance of tailoring datasets to specific learning objectives and the effectiveness of learning from diverse reward models for enhancing both the visual quality and text-video alignment. Additionally, we highlight the vast design space of conditional guidance strategies, which centers on designing an effective energy function to augment the teacher ODE solver. We demonstrate the potential of this approach by extracting motion guidance from the training datasets and incorporating it into the ODE solver, showcasing its effectiveness in improving the motion quality of the generated videos with the improved motion-related metrics from VBench and T2V-CompBench. Empirically, our T2V-Turbo-v2 establishes a new state-of-the-art result on VBench, with a Total score of 85.13, surpassing proprietary systems such as Gen-3 and Kling.
Learning to localize temporal boundaries of procedure steps in instructional videos is challenging due to the limited availability of annotated large-scale training videos. Recent works focus on learning the cross-modal alignment between video segments and ASR-transcripted narration texts through contrastive learning. However, these methods fail to account for the alignment noise, i.e., irrelevant narrations to the instructional task in videos and unreliable timestamps in narrations. To address these challenges, this work proposes a novel training framework. Motivated by the strong capabilities of Large Language Models (LLMs) in procedure understanding and text summarization, we first apply an LLM to filter out task-irrelevant information and summarize task-related procedure steps (LLM-steps) from narrations. To further generate reliable pseudo-matching between the LLM-steps and the video for training, we propose the Multi-Pathway Text-Video Alignment (MPTVA) strategy. The key idea is to measure alignment between LLM-steps and videos via multiple pathways, including: (1) step-narration-video alignment using narration timestamps, (2) direct step-to-video alignment based on their long-term semantic similarity, and (3) direct step-to-video alignment focusing on short-term fine-grained semantic similarity learned from general video domains. The results from different pathways are fused to generate reliable pseudo step-video matching. We conducted extensive experiments across various tasks and problem settings to evaluate our proposed method. Our approach surpasses state-of-the-art methods in three downstream tasks: procedure step grounding, step localization, and narration grounding by 5.9\%, 3.1\%, and 2.8\%.
Robust frame-wise embeddings are essential to perform video analysis and understanding tasks. We present a self-supervised method for representation learning based on aligning temporal video sequences. Our framework uses a transformer-based encoder to extract frame-level features and leverages them to find the optimal alignment path between video sequences. We introduce the novel Local-Alignment Contrastive (LAC) loss, which combines a differentiable local alignment loss to capture local temporal dependencies with a contrastive loss to enhance discriminative learning. Prior works on video alignment have focused on using global temporal ordering across sequence pairs, whereas our loss encourages identifying the best-scoring subsequence alignment. LAC uses the differentiable Smith-Waterman (SW) affine method, which features a flexible parameterization learned through the training phase, enabling the model to adjust the temporal gap penalty length dynamically. Evaluations show that our learned representations outperform existing state-of-the-art approaches on action recognition tasks.