Audio-driven talking face video generation has attracted increasing attention due to its huge industrial potential. Some previous methods focus on learning a direct mapping from audio to visual content. Despite progress, they often struggle with the ambiguity of the mapping process, leading to flawed results. An alternative strategy involves facial structural representations (e.g., facial landmarks) as intermediaries. This multi-stage approach better preserves the appearance details but suffers from error accumulation due to the independent optimization of different stages. Moreover, most previous methods rely on generative adversarial networks, prone to training instability and mode collapse. To address these challenges, our study proposes a novel landmark-based diffusion model for talking face generation, which leverages facial landmarks as intermediate representations while enabling end-to-end optimization. Specifically, we first establish the less ambiguous mapping from audio to landmark motion of lip and jaw. Then, we introduce an innovative conditioning module called TalkFormer to align the synthesized motion with the motion represented by landmarks via differentiable cross-attention, which enables end-to-end optimization for improved lip synchronization. Besides, TalkFormer employs implicit feature warping to align the reference image features with the target motion for preserving more appearance details. Extensive experiments demonstrate that our approach can synthesize high-fidelity and lip-synced talking face videos, preserving more subject appearance details from the reference image.
We present Puppet-Master, an interactive video generative model that can serve as a motion prior for part-level dynamics. At test time, given a single image and a sparse set of motion trajectories (i.e., drags), Puppet-Master can synthesize a video depicting realistic part-level motion faithful to the given drag interactions. This is achieved by fine-tuning a large-scale pre-trained video diffusion model, for which we propose a new conditioning architecture to inject the dragging control effectively. More importantly, we introduce the all-to-first attention mechanism, a drop-in replacement for the widely adopted spatial attention modules, which significantly improves generation quality by addressing the appearance and background issues in existing models. Unlike other motion-conditioned video generators that are trained on in-the-wild videos and mostly move an entire object, Puppet-Master is learned from Objaverse-Animation-HQ, a new dataset of curated part-level motion clips. We propose a strategy to automatically filter out sub-optimal animations and augment the synthetic renderings with meaningful motion trajectories. Puppet-Master generalizes well to real images across various categories and outperforms existing methods in a zero-shot manner on a real-world benchmark. See our project page for more results: vgg-puppetmaster.github.io.
The quality of video-text pairs fundamentally determines the upper bound of text-to-video models. Currently, the datasets used for training these models suffer from significant shortcomings, including low temporal consistency, poor-quality captions, substandard video quality, and imbalanced data distribution. The prevailing video curation process, which depends on image models for tagging and manual rule-based curation, leads to a high computational load and leaves behind unclean data. As a result, there is a lack of appropriate training datasets for text-to-video models. To address this problem, we present VidGen-1M, a superior training dataset for text-to-video models. Produced through a coarse-to-fine curation strategy, this dataset guarantees high-quality videos and detailed captions with excellent temporal consistency. When used to train the video generation model, this dataset has led to experimental results that surpass those obtained with other models.
Recent years have seen a tremendous improvement in the quality of video generation and editing approaches. While several techniques focus on editing appearance, few address motion. Current approaches using text, trajectories, or bounding boxes are limited to simple motions, so we specify motions with a single motion reference video instead. We further propose to use a pre-trained image-to-video model rather than a text-to-video model. This approach allows us to preserve the exact appearance and position of a target object or scene and helps disentangle appearance from motion. Our method, called motion-textual inversion, leverages our observation that image-to-video models extract appearance mainly from the (latent) image input, while the text/image embedding injected via cross-attention predominantly controls motion. We thus represent motion using text/image embedding tokens. By operating on an inflated motion-text embedding containing multiple text/image embedding tokens per frame, we achieve a high temporal motion granularity. Once optimized on the motion reference video, this embedding can be applied to various target images to generate videos with semantically similar motions. Our approach does not require spatial alignment between the motion reference video and target image, generalizes across various domains, and can be applied to various tasks such as full-body and face reenactment, as well as controlling the motion of inanimate objects and the camera. We empirically demonstrate the effectiveness of our method in the semantic video motion transfer task, significantly outperforming existing methods in this context. Project website: https://mkansy.github.io/reenact-anything/
Recent advancements in Diffusion Transformer (DiT) have demonstrated remarkable proficiency in producing high-quality video content. Nonetheless, the potential of transformer-based diffusion models for effectively generating videos with controllable motion remains an area of limited exploration. This paper introduces Tora, the first trajectory-oriented DiT framework that concurrently integrates textual, visual, and trajectory conditions, thereby enabling scalable video generation with effective motion guidance. Specifically, Tora consists of a Trajectory Extractor(TE), a Spatial-Temporal DiT, and a Motion-guidance Fuser(MGF). The TE encodes arbitrary trajectories into hierarchical spacetime motion patches with a 3D video compression network. The MGF integrates the motion patches into the DiT blocks to generate consistent videos that accurately follow designated trajectories. Our design aligns seamlessly with DiT's scalability, allowing precise control of video content's dynamics with diverse durations, aspect ratios, and resolutions. Extensive experiments demonstrate Tora's excellence in achieving high motion fidelity, while also meticulously simulating the intricate movement of the physical world. Code is available at: https://github.com/alibaba/Tora.
Incorporating a temporal dimension into pretrained image diffusion models for video generation is a prevalent approach. However, this method is computationally demanding and necessitates large-scale video datasets. More critically, the heterogeneity between image and video datasets often results in catastrophic forgetting of the image expertise. Recent attempts to directly extract video snippets from image diffusion models have somewhat mitigated these problems. Nevertheless, these methods can only generate brief video clips with simple movements and fail to capture fine-grained motion or non-grid deformation. In this paper, we propose a novel Zero-Shot video Sampling algorithm, denoted as $\mathcal{ZS}^2$, capable of directly sampling high-quality video clips from existing image synthesis methods, such as Stable Diffusion, without any training or optimization. Specifically, $\mathcal{ZS}^2$ utilizes the dependency noise model and temporal momentum attention to ensure content consistency and animation coherence, respectively. This ability enables it to excel in related tasks, such as conditional and context-specialized video generation and instruction-guided video editing. Experimental results demonstrate that $\mathcal{ZS}^2$ achieves state-of-the-art performance in zero-shot video generation, occasionally outperforming recent supervised methods. Homepage: \url{https://densechen.github.io/zss/}.
In recent years, artificial intelligence (AI)-driven video generation has gained significant attention. Consequently, there is a growing need for accurate video quality assessment (VQA) metrics to evaluate the perceptual quality of AI-generated content (AIGC) videos and optimize video generation models. However, assessing the quality of AIGC videos remains a significant challenge because these videos often exhibit highly complex distortions, such as unnatural actions and irrational objects. To address this challenge, we systematically investigate the AIGC-VQA problem, considering both subjective and objective quality assessment perspectives. For the subjective perspective, we construct the Large-scale Generated Video Quality assessment (LGVQ) dataset, consisting of 2,808 AIGC videos generated by 6 video generation models using 468 carefully curated text prompts. We evaluate the perceptual quality of AIGC videos from three critical dimensions: spatial quality, temporal quality, and text-video alignment. For the objective perspective, we establish a benchmark for evaluating existing quality assessment metrics on the LGVQ dataset. Our findings show that current metrics perform poorly on this dataset, highlighting a gap in effective evaluation tools. To bridge this gap, we propose the Unify Generated Video Quality assessment (UGVQ) model, designed to accurately evaluate the multi-dimensional quality of AIGC videos. The UGVQ model integrates the visual and motion features of videos with the textual features of their corresponding prompts, forming a unified quality-aware feature representation tailored to AIGC videos. Experimental results demonstrate that UGVQ achieves state-of-the-art performance on the LGVQ dataset across all three quality dimensions. Both the LGVQ dataset and the UGVQ model are publicly available on https://github.com/zczhang-sjtu/UGVQ.git.
Echocardiography video is a primary modality for diagnosing heart diseases, but the limited data poses challenges for both clinical teaching and machine learning training. Recently, video generative models have emerged as a promising strategy to alleviate this issue. However, previous methods often relied on holistic conditions during generation, hindering the flexible movement control over specific cardiac structures. In this context, we propose an explainable and controllable method for echocardiography video generation, taking an initial frame and a motion curve as guidance. Our contributions are three-fold. First, we extract motion information from each heart substructure to construct motion curves, enabling the diffusion model to synthesize customized echocardiography videos by modifying these curves. Second, we propose the structure-to-motion alignment module, which can map semantic features onto motion curves across cardiac structures. Third, The position-aware attention mechanism is designed to enhance video consistency utilizing Gaussian masks with structural position information. Extensive experiments on three echocardiography datasets show that our method outperforms others regarding fidelity and consistency. The full code will be released at https://github.com/mlmi-2024-72/ECM.
In recent years, large-scale pre-trained diffusion models have demonstrated their outstanding capabilities in image and video generation tasks. However, existing models tend to produce visual objects commonly found in the training dataset, which diverges from user input prompts. The underlying reason behind the inaccurate generated results lies in the model's difficulty in sampling from specific intervals of the initial noise distribution corresponding to the prompt. Moreover, it is challenging to directly optimize the initial distribution, given that the diffusion process involves multiple denoising steps. In this paper, we introduce a Fine-tuning Initial Noise Distribution (FIND) framework with policy optimization, which unleashes the powerful potential of pre-trained diffusion networks by directly optimizing the initial distribution to align the generated contents with user-input prompts. To this end, we first reformulate the diffusion denoising procedure as a one-step Markov decision process and employ policy optimization to directly optimize the initial distribution. In addition, a dynamic reward calibration module is proposed to ensure training stability during optimization. Furthermore, we introduce a ratio clipping algorithm to utilize historical data for network training and prevent the optimized distribution from deviating too far from the original policy to restrain excessive optimization magnitudes. Extensive experiments demonstrate the effectiveness of our method in both text-to-image and text-to-video tasks, surpassing SOTA methods in achieving consistency between prompts and the generated content. Our method achieves 10 times faster than the SOTA approach. Our homepage is available at \url{https://github.com/vpx-ecnu/FIND-website}.
Video diffusion models have made substantial progress in various video generation applications. However, training models for long video generation tasks require significant computational and data resources, posing a challenge to developing long video diffusion models. This paper investigates a straightforward and training-free approach to extend an existing short video diffusion model (e.g. pre-trained on 16-frame videos) for consistent long video generation (e.g. 128 frames). Our preliminary observation has found that directly applying the short video diffusion model to generate long videos can lead to severe video quality degradation. Further investigation reveals that this degradation is primarily due to the distortion of high-frequency components in long videos, characterized by a decrease in spatial high-frequency components and an increase in temporal high-frequency components. Motivated by this, we propose a novel solution named FreeLong to balance the frequency distribution of long video features during the denoising process. FreeLong blends the low-frequency components of global video features, which encapsulate the entire video sequence, with the high-frequency components of local video features that focus on shorter subsequences of frames. This approach maintains global consistency while incorporating diverse and high-quality spatiotemporal details from local videos, enhancing both the consistency and fidelity of long video generation. We evaluated FreeLong on multiple base video diffusion models and observed significant improvements. Additionally, our method supports coherent multi-prompt generation, ensuring both visual coherence and seamless transitions between scenes.
This paper investigates the role of CLIP image embeddings within the Stable Video Diffusion (SVD) framework, focusing on their impact on video generation quality and computational efficiency. Our findings indicate that CLIP embeddings, while crucial for aesthetic quality, do not significantly contribute towards the subject and background consistency of video outputs. Moreover, the computationally expensive cross-attention mechanism can be effectively replaced by a simpler linear layer. This layer is computed only once at the first diffusion inference step, and its output is then cached and reused throughout the inference process, thereby enhancing efficiency while maintaining high-quality outputs. Building on these insights, we introduce the VCUT, a training-free approach optimized for efficiency within the SVD architecture. VCUT eliminates temporal cross-attention and replaces spatial cross-attention with a one-time computed linear layer, significantly reducing computational load. The implementation of VCUT leads to a reduction of up to 322T Multiple-Accumulate Operations (MACs) per video and a decrease in model parameters by up to 50M, achieving a 20% reduction in latency compared to the baseline. Our approach demonstrates that conditioning during the Semantic Binding stage is sufficient, eliminating the need for continuous computation across all inference steps and setting a new standard for efficient video generation.
We present Stable Video 4D (SV4D), a latent video diffusion model for multi-frame and multi-view consistent dynamic 3D content generation. Unlike previous methods that rely on separately trained generative models for video generation and novel view synthesis, we design a unified diffusion model to generate novel view videos of dynamic 3D objects. Specifically, given a monocular reference video, SV4D generates novel views for each video frame that are temporally consistent. We then use the generated novel view videos to optimize an implicit 4D representation (dynamic NeRF) efficiently, without the need for cumbersome SDS-based optimization used in most prior works. To train our unified novel view video generation model, we curated a dynamic 3D object dataset from the existing Objaverse dataset. Extensive experimental results on multiple datasets and user studies demonstrate SV4D's state-of-the-art performance on novel-view video synthesis as well as 4D generation compared to prior works.
Human image animation involves generating videos from a character photo, allowing user control and unlocking the potential for video and movie production. While recent approaches yield impressive results using high-quality training data, the inaccessibility of these datasets hampers fair and transparent benchmarking. Moreover, these approaches prioritize 2D human motion and overlook the significance of camera motions in videos, leading to limited control and unstable video generation. To demystify the training data, we present HumanVid, the first large-scale high-quality dataset tailored for human image animation, which combines crafted real-world and synthetic data. For the real-world data, we compile a vast collection of real-world videos from the internet. We developed and applied careful filtering rules to ensure video quality, resulting in a curated collection of 20K high-resolution (1080P) human-centric videos. Human and camera motion annotation is accomplished using a 2D pose estimator and a SLAM-based method. To expand our synthetic dataset, we collected 10K 3D avatar assets and leveraged existing assets of body shapes, skin textures and clothings. Notably, we introduce a rule-based camera trajectory generation method, enabling the synthetic pipeline to incorporate diverse and precise camera motion annotation, which can rarely be found in real-world data. To verify the effectiveness of HumanVid, we establish a baseline model named CamAnimate, short for Camera-controllable Human Animation, that considers both human and camera motions as conditions. Through extensive experimentation, we demonstrate that such simple baseline training on our HumanVid achieves state-of-the-art performance in controlling both human pose and camera motions, setting a new benchmark. Demo, data and code could be found in the project website: https://humanvid.github.io/.
Diffusion Transformer, the backbone of Sora for video generation, successfully scales the capacity of diffusion models, pioneering new avenues for high-fidelity sequential data generation. Unlike static data such as images, sequential data consists of consecutive data frames indexed by time, exhibiting rich spatial and temporal dependencies. These dependencies represent the underlying dynamic model and are critical to validate the generated data. In this paper, we make the first theoretical step towards bridging diffusion transformers for capturing spatial-temporal dependencies. Specifically, we establish score approximation and distribution estimation guarantees of diffusion transformers for learning Gaussian process data with covariance functions of various decay patterns. We highlight how the spatial-temporal dependencies are captured and affect learning efficiency. Our study proposes a novel transformer approximation theory, where the transformer acts to unroll an algorithm. We support our theoretical results by numerical experiments, providing strong evidence that spatial-temporal dependencies are captured within attention layers, aligning with our approximation theory.
Recent advancements in video generation have primarily leveraged diffusion models for short-duration content. However, these approaches often fall short in modeling complex narratives and maintaining character consistency over extended periods, which is essential for long-form video production like movies. We propose MovieDreamer, a novel hierarchical framework that integrates the strengths of autoregressive models with diffusion-based rendering to pioneer long-duration video generation with intricate plot progressions and high visual fidelity. Our approach utilizes autoregressive models for global narrative coherence, predicting sequences of visual tokens that are subsequently transformed into high-quality video frames through diffusion rendering. This method is akin to traditional movie production processes, where complex stories are factorized down into manageable scene capturing. Further, we employ a multimodal script that enriches scene descriptions with detailed character information and visual style, enhancing continuity and character identity across scenes. We present extensive experiments across various movie genres, demonstrating that our approach not only achieves superior visual and narrative quality but also effectively extends the duration of generated content significantly beyond current capabilities. Homepage: https://aim-uofa.github.io/MovieDreamer/.
Significant advancements have been made in video generative models recently. Unlike image generation, video generation presents greater challenges, requiring not only generating high-quality frames but also ensuring temporal consistency across these frames. Despite the impressive progress, research on metrics for evaluating the quality of generated videos, especially concerning temporal and motion consistency, remains underexplored. To bridge this research gap, we propose Fr\'echet Video Motion Distance (FVMD) metric, which focuses on evaluating motion consistency in video generation. Specifically, we design explicit motion features based on key point tracking, and then measure the similarity between these features via the Fr\'echet distance. We conduct sensitivity analysis by injecting noise into real videos to verify the effectiveness of FVMD. Further, we carry out a large-scale human study, demonstrating that our metric effectively detects temporal noise and aligns better with human perceptions of generated video quality than existing metrics. Additionally, our motion features can consistently improve the performance of Video Quality Assessment (VQA) models, indicating that our approach is also applicable to unary video quality evaluation. Code is available at https://github.com/ljh0v0/FMD-frechet-motion-distance.
Video generation has drawn significant interest recently, pushing the development of large-scale models capable of producing realistic videos with coherent motion. Due to memory constraints, these models typically generate short video segments that are then combined into long videos. The merging process poses a significant challenge, as it requires ensuring smooth transitions and overall consistency. In this paper, we introduce Anchored Diffusion, a novel method for synthesizing relatively long and seamless videos. We extend Diffusion Transformers (DiTs) to incorporate temporal information, creating our sequence-DiT (sDiT) model for generating short video segments. Unlike previous works, we train our model on video sequences with random non-uniform temporal spacing and incorporate temporal information via external guidance, increasing flexibility and allowing it to capture both short and long-term relationships. Furthermore, during inference, we leverage the transformer architecture to modify the diffusion process, generating a batch of non-uniform sequences anchored to a common frame, ensuring consistency regardless of temporal distance. To demonstrate our method, we focus on face reenactment, the task of creating a video from a source image that replicates the facial expressions and movements from a driving video. Through comprehensive experiments, we show our approach outperforms current techniques in producing longer consistent high-quality videos while offering editing capabilities.
With the advancement of computer vision and natural language processing, text-to-video generation, enabled by text-to-video diffusion models, has become more prevalent. These models are trained using a large amount of data from the internet. However, the training data often contain copyrighted content, including cartoon character icons and artist styles, private portraits, and unsafe videos. Since filtering the data and retraining the model is challenging, methods for unlearning specific concepts from text-to-video diffusion models have been investigated. However, due to the high computational complexity and relative large optimization scale, there is little work on unlearning methods for text-to-video diffusion models. We propose a novel concept-unlearning method by transferring the unlearning capability of the text encoder of text-to-image diffusion models to text-to-video diffusion models. Specifically, the method optimizes the text encoder using few-shot unlearning, where several generated images are used. We then use the optimized text encoder in text-to-video diffusion models to generate videos. Our method costs low computation resources and has small optimization scale. We discuss the generated videos after unlearning a concept. The experiments demonstrates that our method can unlearn copyrighted cartoon characters, artist styles, objects and people's facial characteristics. Our method can unlearn a concept within about 100 seconds on an RTX 3070. Since there was no concept unlearning method for text-to-video diffusion models before, we make concept unlearning feasible and more accessible in the text-to-video domain.
Text-to-video (T2V) generative models have advanced significantly, yet their ability to compose different objects, attributes, actions, and motions into a video remains unexplored. Previous text-to-video benchmarks also neglect this important ability for evaluation. In this work, we conduct the first systematic study on compositional text-to-video generation. We propose T2V-CompBench, the first benchmark tailored for compositional text-to-video generation. T2V-CompBench encompasses diverse aspects of compositionality, including consistent attribute binding, dynamic attribute binding, spatial relationships, motion binding, action binding, object interactions, and generative numeracy. We further carefully design evaluation metrics of multimodal large language model (MLLM)-based, detection-based, and tracking-based metrics, which can better reflect the compositional text-to-video generation quality of seven proposed categories with 1400 text prompts. The effectiveness of the proposed metrics is verified by correlation with human evaluations. We also benchmark various text-to-video generative models and conduct in-depth analysis across different models and various compositional categories. We find that compositional text-to-video generation is highly challenging for current models, and we hope our attempt could shed light on future research in this direction.
We present a method for generating Streetscapes-long sequences of views through an on-the-fly synthesized city-scale scene. Our generation is conditioned by language input (e.g., city name, weather), as well as an underlying map/layout hosting the desired trajectory. Compared to recent models for video generation or 3D view synthesis, our method can scale to much longer-range camera trajectories, spanning several city blocks, while maintaining visual quality and consistency. To achieve this goal, we build on recent work on video diffusion, used within an autoregressive framework that can easily scale to long sequences. In particular, we introduce a new temporal imputation method that prevents our autoregressive approach from drifting from the distribution of realistic city imagery. We train our Streetscapes system on a compelling source of data-posed imagery from Google Street View, along with contextual map data-which allows users to generate city views conditioned on any desired city layout, with controllable camera poses. Please see more results at our project page at https://boyangdeng.com/streetscapes.