Diffusion models are successful for synthesizing high-quality videos but are limited to generating short clips (e.g., 2-10 seconds). Synthesizing sustained footage (e.g. over minutes) still remains an open research question. In this paper, we propose MALT Diffusion (using Memory-Augmented Latent Transformers), a new diffusion model specialized for long video generation. MALT Diffusion (or just MALT) handles long videos by subdividing them into short segments and doing segment-level autoregressive generation. To achieve this, we first propose recurrent attention layers that encode multiple segments into a compact memory latent vector; by maintaining this memory vector over time, MALT is able to condition on it and continuously generate new footage based on a long temporal context. We also present several training techniques that enable the model to generate frames over a long horizon with consistent quality and minimal degradation. We validate the effectiveness of MALT through experiments on long video benchmarks. We first perform extensive analysis of MALT in long-contextual understanding capability and stability using popular long video benchmarks. For example, MALT achieves an FVD score of 220.4 on 128-frame video generation on UCF-101, outperforming the previous state-of-the-art of 648.4. Finally, we explore MALT's capabilities in a text-to-video generation setting and show that it can produce long videos compared with recent techniques for long text-to-video generation.
Popular Micro-videos, dominant on platforms like TikTok and YouTube, hold significant commercial value. The rise of high-quality AI-generated content has spurred interest in AI-driven micro-video creation. However, despite the advanced capabilities of large language models (LLMs) like ChatGPT and DeepSeek in text generation and reasoning, their potential to assist the creation of popular micro-videos remains largely unexplored. In this paper, we conduct an empirical study on LLM-assisted popular micro-video generation (LLMPopcorn). Specifically, we investigate the following research questions: (i) How can LLMs be effectively utilized to assist popular micro-video generation? (ii) To what extent can prompt-based enhancements optimize the LLM-generated content for higher popularity? (iii) How well do various LLMs and video generators perform in the popular micro-video generation task? By exploring these questions, we show that advanced LLMs like DeepSeek-V3 enable micro-video generation to achieve popularity comparable to human-created content. Prompt enhancements further boost popularity, and benchmarking highlights DeepSeek-V3 and DeepSeek-R1 among LLMs, while LTX-Video and HunyuanVideo lead in video generation. This pioneering work advances AI-assisted micro-video creation, uncovering new research opportunities. We will release the code and datasets to support future studies.
Recent advancements in camera-trajectory-guided image-to-video generation offer higher precision and better support for complex camera control compared to text-based approaches. However, they also introduce significant usability challenges, as users often struggle to provide precise camera parameters when working with arbitrary real-world images without knowledge of their depth nor scene scale. To address these real-world application issues, we propose RealCam-I2V, a novel diffusion-based video generation framework that integrates monocular metric depth estimation to establish 3D scene reconstruction in a preprocessing step. During training, the reconstructed 3D scene enables scaling camera parameters from relative to absolute values, ensuring compatibility and scale consistency across diverse real-world images. In inference, RealCam-I2V offers an intuitive interface where users can precisely draw camera trajectories by dragging within the 3D scene. To further enhance precise camera control and scene consistency, we propose scene-constrained noise shaping, which shapes high-level noise and also allows the framework to maintain dynamic, coherent video generation in lower noise stages. RealCam-I2V achieves significant improvements in controllability and video quality on the RealEstate10K and out-of-domain images. We further enables applications like camera-controlled looping video generation and generative frame interpolation. We will release our absolute-scale annotation, codes, and all checkpoints. Please see dynamic results in https://zgctroy.github.io/RealCam-I2V.
Accurate and flexible world models are crucial for autonomous systems to understand their environment and predict future events. Object-centric models, with structured latent spaces, have shown promise in modeling object dynamics and interactions, but often face challenges in scaling to complex datasets and incorporating external guidance, limiting their applicability in robotics. To address these limitations, we propose TextOCVP, an object-centric model for image-to-video generation guided by textual descriptions. TextOCVP parses an observed scene into object representations, called slots, and utilizes a text-conditioned transformer predictor to forecast future object states and video frames. Our approach jointly models object dynamics and interactions while incorporating textual guidance, thus leading to accurate and controllable predictions. Our method's structured latent space offers enhanced control over the prediction process, outperforming several image-to-video generative baselines. Additionally, we demonstrate that structured object-centric representations provide superior controllability and interpretability, facilitating the modeling of object dynamics and enabling more precise and understandable predictions. Videos and code are available at https://play-slot.github.io/TextOCVP/.
Generating long, high-quality videos remains a challenge due to the complex interplay of spatial and temporal dynamics and hardware limitations. In this work, we introduce \textbf{MaskFlow}, a unified video generation framework that combines discrete representations with flow-matching to enable efficient generation of high-quality long videos. By leveraging a frame-level masking strategy during training, MaskFlow conditions on previously generated unmasked frames to generate videos with lengths ten times beyond that of the training sequences. MaskFlow does so very efficiently by enabling the use of fast Masked Generative Model (MGM)-style sampling and can be deployed in both fully autoregressive as well as full-sequence generation modes. We validate the quality of our method on the FaceForensics (FFS) and Deepmind Lab (DMLab) datasets and report Fr\'echet Video Distance (FVD) competitive with state-of-the-art approaches. We also provide a detailed analysis on the sampling efficiency of our method and demonstrate that MaskFlow can be applied to both timestep-dependent and timestep-independent models in a training-free manner.
Despite the promise of synthesizing high-fidelity videos, Diffusion Transformers (DiTs) with 3D full attention suffer from expensive inference due to the complexity of attention computation and numerous sampling steps. For example, the popular Open-Sora-Plan model consumes more than 9 minutes for generating a single video of 29 frames. This paper addresses the inefficiency issue from two aspects: 1) Prune the 3D full attention based on the redundancy within video data; We identify a prevalent tile-style repetitive pattern in the 3D attention maps for video data, and advocate a new family of sparse 3D attention that holds a linear complexity w.r.t. the number of video frames. 2) Shorten the sampling process by adopting existing multi-step consistency distillation; We split the entire sampling trajectory into several segments and perform consistency distillation within each one to activate few-step generation capacities. We further devise a three-stage training pipeline to conjoin the low-complexity attention and few-step generation capacities. Notably, with 0.1% pretraining data, we turn the Open-Sora-Plan-1.2 model into an efficient one that is 7.4x -7.8x faster for 29 and 93 frames 720p video generation with a marginal performance trade-off in VBench. In addition, we demonstrate that our approach is amenable to distributed inference, achieving an additional 3.91x speedup when running on 4 GPUs with sequence parallelism.
In this paper, we propose the Dynamic Latent Frame Rate VAE (DLFR-VAE), a training-free paradigm that can make use of adaptive temporal compression in latent space. While existing video generative models apply fixed compression rates via pretrained VAE, we observe that real-world video content exhibits substantial temporal non-uniformity, with high-motion segments containing more information than static scenes. Based on this insight, DLFR-VAE dynamically adjusts the latent frame rate according to the content complexity. Specifically, DLFR-VAE comprises two core innovations: (1) A Dynamic Latent Frame Rate Scheduler that partitions videos into temporal chunks and adaptively determines optimal frame rates based on information-theoretic content complexity, and (2) A training-free adaptation mechanism that transforms pretrained VAE architectures into a dynamic VAE that can process features with variable frame rates. Our simple but effective DLFR-VAE can function as a plug-and-play module, seamlessly integrating with existing video generation models and accelerating the video generation process.
The continuous development of foundational models for video generation is evolving into various applications, with subject-consistent video generation still in the exploratory stage. We refer to this as Subject-to-Video, which extracts subject elements from reference images and generates subject-consistent video through textual instructions. We believe that the essence of subject-to-video lies in balancing the dual-modal prompts of text and image, thereby deeply and simultaneously aligning both text and visual content. To this end, we propose Phantom, a unified video generation framework for both single and multi-subject references. Building on existing text-to-video and image-to-video architectures, we redesign the joint text-image injection model and drive it to learn cross-modal alignment via text-image-video triplet data. In particular, we emphasize subject consistency in human generation, covering existing ID-preserving video generation while offering enhanced advantages. The project homepage is here https://phantom-video.github.io/Phantom/.
We present Step-Video-T2V, a state-of-the-art text-to-video pre-trained model with 30B parameters and the ability to generate videos up to 204 frames in length. A deep compression Variational Autoencoder, Video-VAE, is designed for video generation tasks, achieving 16x16 spatial and 8x temporal compression ratios, while maintaining exceptional video reconstruction quality. User prompts are encoded using two bilingual text encoders to handle both English and Chinese. A DiT with 3D full attention is trained using Flow Matching and is employed to denoise input noise into latent frames. A video-based DPO approach, Video-DPO, is applied to reduce artifacts and improve the visual quality of the generated videos. We also detail our training strategies and share key observations and insights. Step-Video-T2V's performance is evaluated on a novel video generation benchmark, Step-Video-T2V-Eval, demonstrating its state-of-the-art text-to-video quality when compared with both open-source and commercial engines. Additionally, we discuss the limitations of current diffusion-based model paradigm and outline future directions for video foundation models. We make both Step-Video-T2V and Step-Video-T2V-Eval available at https://github.com/stepfun-ai/Step-Video-T2V. The online version can be accessed from https://yuewen.cn/videos as well. Our goal is to accelerate the innovation of video foundation models and empower video content creators.
This paper presents FloVD, a novel optical-flow-based video diffusion model for camera-controllable video generation. FloVD leverages optical flow maps to represent motions of the camera and moving objects. This approach offers two key benefits. Since optical flow can be directly estimated from videos, our approach allows for the use of arbitrary training videos without ground-truth camera parameters. Moreover, as background optical flow encodes 3D correlation across different viewpoints, our method enables detailed camera control by leveraging the background motion. To synthesize natural object motion while supporting detailed camera control, our framework adopts a two-stage video synthesis pipeline consisting of optical flow generation and flow-conditioned video synthesis. Extensive experiments demonstrate the superiority of our method over previous approaches in terms of accurate camera control and natural object motion synthesis.
Recent advances in video generation models demonstrate their potential as world simulators, but they often struggle with videos deviating from physical laws, a key concern overlooked by most text-to-video benchmarks. We introduce a benchmark designed specifically to assess the Physical Coherence of generated videos, PhyCoBench. Our benchmark includes 120 prompts covering 7 categories of physical principles, capturing key physical laws observable in video content. We evaluated four state-of-the-art (SoTA) T2V models on PhyCoBench and conducted manual assessments. Additionally, we propose an automated evaluation model: PhyCoPredictor, a diffusion model that generates optical flow and video frames in a cascade manner. Through a consistency evaluation comparing automated and manual sorting, the experimental results show that PhyCoPredictor currently aligns most closely with human evaluation. Therefore, it can effectively evaluate the physical coherence of videos, providing insights for future model optimization. Our benchmark, which includes physical coherence prompts, automatic evaluation tool PhyCoPredictor, and generated video dataset, will all be released on GitHub shortly.
In this work, we present CineMaster, a novel framework for 3D-aware and controllable text-to-video generation. Our goal is to empower users with comparable controllability as professional film directors: precise placement of objects within the scene, flexible manipulation of both objects and camera in 3D space, and intuitive layout control over the rendered frames. To achieve this, CineMaster operates in two stages. In the first stage, we design an interactive workflow that allows users to intuitively construct 3D-aware conditional signals by positioning object bounding boxes and defining camera movements within the 3D space. In the second stage, these control signals--comprising rendered depth maps, camera trajectories and object class labels--serve as the guidance for a text-to-video diffusion model, ensuring to generate the user-intended video content. Furthermore, to overcome the scarcity of in-the-wild datasets with 3D object motion and camera pose annotations, we carefully establish an automated data annotation pipeline that extracts 3D bounding boxes and camera trajectories from large-scale video data. Extensive qualitative and quantitative experiments demonstrate that CineMaster significantly outperforms existing methods and implements prominent 3D-aware text-to-video generation. Project page: https://cinemaster-dev.github.io/.
With the rapid development of embodied artificial intelligence, significant progress has been made in vision-language-action (VLA) models for general robot decision-making. However, the majority of existing VLAs fail to account for the inevitable external perturbations encountered during deployment. These perturbations introduce unforeseen state information to the VLA, resulting in inaccurate actions and consequently, a significant decline in generalization performance. The classic internal model control (IMC) principle demonstrates that a closed-loop system with an internal model that includes external input signals can accurately track the reference input and effectively offset the disturbance. We propose a novel closed-loop VLA method GEVRM that integrates the IMC principle to enhance the robustness of robot visual manipulation. The text-guided video generation model in GEVRM can generate highly expressive future visual planning goals. Simultaneously, we evaluate perturbations by simulating responses, which are called internal embeddings and optimized through prototype contrastive learning. This allows the model to implicitly infer and distinguish perturbations from the external environment. The proposed GEVRM achieves state-of-the-art performance on both standard and perturbed CALVIN benchmarks and shows significant improvements in realistic robot tasks.
In this technical report, we present Magic 1-For-1 (Magic141), an efficient video generation model with optimized memory consumption and inference latency. The key idea is simple: factorize the text-to-video generation task into two separate easier tasks for diffusion step distillation, namely text-to-image generation and image-to-video generation. We verify that with the same optimization algorithm, the image-to-video task is indeed easier to converge over the text-to-video task. We also explore a bag of optimization tricks to reduce the computational cost of training the image-to-video (I2V) models from three aspects: 1) model convergence speedup by using a multi-modal prior condition injection; 2) inference latency speed up by applying an adversarial step distillation, and 3) inference memory cost optimization with parameter sparsification. With those techniques, we are able to generate 5-second video clips within 3 seconds. By applying a test time sliding window, we are able to generate a minute-long video within one minute with significantly improved visual quality and motion dynamics, spending less than 1 second for generating 1 second video clips on average. We conduct a series of preliminary explorations to find out the optimal tradeoff between computational cost and video quality during diffusion step distillation and hope this could be a good foundation model for open-source explorations. The code and the model weights are available at https://github.com/DA-Group-PKU/Magic-1-For-1.
DiT-based video generation has achieved remarkable results, but research into enhancing existing models remains relatively unexplored. In this work, we introduce a training-free approach to enhance the coherence and quality of DiT-based generated videos, named Enhance-A-Video. The core idea is enhancing the cross-frame correlations based on non-diagonal temporal attention distributions. Thanks to its simple design, our approach can be easily applied to most DiT-based video generation frameworks without any retraining or fine-tuning. Across various DiT-based video generation models, our approach demonstrates promising improvements in both temporal consistency and visual quality. We hope this research can inspire future explorations in video generation enhancement.
With the rapid development of AI-generated content (AIGC), the creation of high-quality AI-generated videos has become faster and easier, resulting in the Internet being flooded with all kinds of video content. However, the impact of these videos on the content ecosystem remains largely unexplored. Video information retrieval remains a fundamental approach for accessing video content. Building on the observation that retrieval models often favor AI-generated content in ad-hoc and image retrieval tasks, we investigate whether similar biases emerge in the context of challenging video retrieval, where temporal and visual factors may further influence model behavior. To explore this, we first construct a comprehensive benchmark dataset containing both real and AI-generated videos, along with a set of fair and rigorous metrics to assess bias. This benchmark consists of 13,000 videos generated by two state-of-the-art open-source video generation models. We meticulously design a suite of rigorous metrics to accurately measure this preference, accounting for potential biases arising from the limited frame rate and suboptimal quality of AIGC videos. We then applied three off-the-shelf video retrieval models to perform retrieval tasks on this hybrid dataset. Our findings reveal a clear preference for AI-generated videos in retrieval. Further investigation shows that incorporating AI-generated videos into the training set of retrieval models exacerbates this bias. Unlike the preference observed in image modalities, we find that video retrieval bias arises from both unseen visual and temporal information, making the root causes of video bias a complex interplay of these two factors. To mitigate this bias, we fine-tune the retrieval models using a contrastive learning approach. The results of this study highlight the potential implications of AI-generated videos on retrieval systems.
We present ATOP (Articulate That Object Part), a novel method based on motion personalization to articulate a 3D object with respect to a part and its motion as prescribed in a text prompt. Specifically, the text input allows us to tap into the power of modern-day video diffusion to generate plausible motion samples for the right object category and part. In turn, the input 3D object provides image prompting to personalize the generated video to that very object we wish to articulate. Our method starts with a few-shot finetuning for category-specific motion generation, a key first step to compensate for the lack of articulation awareness by current video diffusion models. For this, we finetune a pre-trained multi-view image generation model for controllable multi-view video generation, using a small collection of video samples obtained for the target object category. This is followed by motion video personalization that is realized by multi-view rendered images of the target 3D object. At last, we transfer the personalized video motion to the target 3D object via differentiable rendering to optimize part motion parameters by a score distillation sampling loss. We show that our method is capable of generating realistic motion videos and predict 3D motion parameters in a more accurate and generalizable way, compared to prior works.
Next-Token Prediction (NTP) is a de facto approach for autoregressive (AR) video generation, but it suffers from suboptimal unidirectional dependencies and slow inference speed. In this work, we propose a semi-autoregressive (semi-AR) framework, called Next-Block Prediction (NBP), for video generation. By uniformly decomposing video content into equal-sized blocks (e.g., rows or frames), we shift the generation unit from individual tokens to blocks, allowing each token in the current block to simultaneously predict the corresponding token in the next block. Unlike traditional AR modeling, our framework employs bidirectional attention within each block, enabling tokens to capture more robust spatial dependencies. By predicting multiple tokens in parallel, NBP models significantly reduce the number of generation steps, leading to faster and more efficient inference. Our model achieves FVD scores of 103.3 on UCF101 and 25.5 on K600, outperforming the vanilla NTP model by an average of 4.4. Furthermore, thanks to the reduced number of inference steps, the NBP model generates 8.89 frames (128x128 resolution) per second, achieving an 11x speedup. We also explored model scales ranging from 700M to 3B parameters, observing significant improvements in generation quality, with FVD scores dropping from 103.3 to 55.3 on UCF101 and from 25.5 to 19.5 on K600, demonstrating the scalability of our approach.
Co-speech gesture generation is crucial for creating lifelike avatars and enhancing human-computer interactions by synchronizing gestures with speech. Despite recent advancements, existing methods struggle with accurately identifying the rhythmic or semantic triggers from audio for generating contextualized gesture patterns and achieving pixel-level realism. To address these challenges, we introduce Contextual Gesture, a framework that improves co-speech gesture video generation through three innovative components: (1) a chronological speech-gesture alignment that temporally connects two modalities, (2) a contextualized gesture tokenization that incorporate speech context into motion pattern representation through distillation, and (3) a structure-aware refinement module that employs edge connection to link gesture keypoints to improve video generation. Our extensive experiments demonstrate that Contextual Gesture not only produces realistic and speech-aligned gesture videos but also supports long-sequence generation and video gesture editing applications, shown in Fig.1 Project Page: https://andypinxinliu.github.io/Contextual-Gesture/.
Diffusion models have been used extensively for high quality image and video generation tasks. In this paper, we propose a novel conditional diffusion model with spatial attention and latent embedding (cDAL) for medical image segmentation. In cDAL, a convolutional neural network (CNN) based discriminator is used at every time-step of the diffusion process to distinguish between the generated labels and the real ones. A spatial attention map is computed based on the features learned by the discriminator to help cDAL generate more accurate segmentation of discriminative regions in an input image. Additionally, we incorporated a random latent embedding into each layer of our model to significantly reduce the number of training and sampling time-steps, thereby making it much faster than other diffusion models for image segmentation. We applied cDAL on 3 publicly available medical image segmentation datasets (MoNuSeg, Chest X-ray and Hippocampus) and observed significant qualitative and quantitative improvements with higher Dice scores and mIoU over the state-of-the-art algorithms. The source code is publicly available at https://github.com/Hejrati/cDAL/.