Teng Hu

Qiming Peng, Yinxu Pan, Wenjin Wang et al.

Recent years have witnessed the rise and success of pre-training techniques in visually-rich document understanding. However, most existing methods lack the systematic mining and utilization of layout-centered knowledge, leading to sub-optimal performances. In this paper, we propose ERNIE-Layout, a novel document pre-training solution with layout knowledge enhancement in the whole workflow, to learn better representations that combine the features from text, layout, and image. Specifically, we first rearrange input sequences in the serialization stage, and then present a correlative pre-training task, reading order prediction, to learn the proper reading order of documents. To improve the layout awareness of the model, we integrate a spatial-aware disentangled attention into the multi-modal transformer and a replaced regions prediction task into the pre-training phase. Experimental results show that ERNIE-Layout achieves superior performance on various downstream tasks, setting new state-of-the-art on key information extraction, document image classification, and document question answering datasets. The code and models are publicly available at http://github.com/PaddlePaddle/PaddleNLP/tree/develop/model_zoo/ernie-layout.

Teng Hu, Ran Yi, Haokun Zhu et al. · tsinghua

Stroke-based rendering aims to recreate an image with a set of strokes. Most existing methods render complex images using an uniform-block-dividing strategy, which leads to boundary inconsistency artifacts. To solve the problem, we propose Compositional Neural Painter, a novel stroke-based rendering framework which dynamically predicts the next painting region based on the current canvas, instead of dividing the image plane uniformly into painting regions. We start from an empty canvas and divide the painting process into several steps. At each step, a compositor network trained with a phasic RL strategy first predicts the next painting region, then a painter network trained with a WGAN discriminator predicts stroke parameters, and a stroke renderer paints the strokes onto the painting region of the current canvas. Moreover, we extend our method to stroke-based style transfer with a novel differentiable distance transform loss, which helps preserve the structure of the input image during stroke-based stylization. Extensive experiments show our model outperforms the existing models in both stroke-based neural painting and stroke-based stylization. Code is available at https://github.com/sjtuplayer/Compositional_Neural_Painter

Teng Hu, Jiangning Zhang, Liang Liu et al. · tsinghua

Training a generative model with limited number of samples is a challenging task. Current methods primarily rely on few-shot model adaption to train the network. However, in scenarios where data is extremely limited (less than 10), the generative network tends to overfit and suffers from content degradation. To address these problems, we propose a novel phasic content fusing few-shot diffusion model with directional distribution consistency loss, which targets different learning objectives at distinct training stages of the diffusion model. Specifically, we design a phasic training strategy with phasic content fusion to help our model learn content and style information when t is large, and learn local details of target domain when t is small, leading to an improvement in the capture of content, style and local details. Furthermore, we introduce a novel directional distribution consistency loss that ensures the consistency between the generated and source distributions more efficiently and stably than the prior methods, preventing our model from overfitting. Finally, we propose a cross-domain structure guidance strategy that enhances structure consistency during domain adaptation. Theoretical analysis, qualitative and quantitative experiments demonstrate the superiority of our approach in few-shot generative model adaption tasks compared to state-of-the-art methods. The source code is available at: https://github.com/sjtuplayer/few-shot-diffusion.

Yuheng Chen, Teng Hu, Yuji Wang et al.

Identity-preserving video generation (IPVG) aims to synthesize high-fidelity videos that follow text prompts while faithfully preserving a reference identity. Despite recent progress, existing IPVG methods still struggle to balance high-level semantic control and low-level identity fidelity. To bridge this gap, we propose ST-DRC, an effective Spatial-Temporal Decoupled Reference Conditioning framework for identity-preserving text-to-video generation. At the framework level, ST-DRC performs latent in-context feature injection by encoding the reference image with the video VAE and concatenating it with noisy video latents, enabling rich low-level identity details to be accessed without additional adapters. To separate identity-aware reference retrieval from appearance copying, we introduce TASS-RoPE, a Temporal-Adjacent Spatial-Shifted RoPE scheme that places reference tokens near the video sequence in time but shifts them in space, allowing reference information to flow through spatio-temporal attention while suppressing pixel-level copy-paste shortcuts. To further prevent shortcut learning and strengthen the otherwise diluted identity supervision in the diffusion objective, we combine appearance-invariant reference augmentation with face-guided identity objectives, encouraging the model to preserve identity under variations in color, pose, and layout. At inference time, we introduce a three-stream reference classifier-free guidance strategy that independently controls text adherence and reference fidelity. Experiments demonstrate that ST-DRC achieves strong identity preservation, prompt alignment, temporal consistency, and video quality with a lightweight design built on LTX-2.3. Our method ranks among the top submissions in the facial identity-preserving video generation track, validating the effectiveness of spatial-temporal decoupled reference conditioning.

Teng Hu, Mingchun Lu, Yating Wang et al.

Video world models are a foundational generative technology for embodied AI and the Metaverse, yet existing approaches are inherently limited to a single agent observing from a single perspective. Extending these models to multi-agent settings introduces two critical challenges: data scarcity (coordinated multi-view recordings are prohibitively expensive to collect for general open-domain scenarios) and world state alignment (independently generated video streams cannot ensure that shared physical environments and events evolve consistently across views). To address these challenges, we propose MetaWorld, a novel framework that scales multi-agent video world models to open-domain environments directly from single-view videos. First, we introduce Monocular World-State Unrolling (MWSU) to explicitly decompose monocular footage into the camera operator's ego-motion and the visible subject's spatial trajectory. This camera-trajectory decomposition naturally extracts synchronized multi-agent motion data within a shared 3D space, completely bypassing the need for multi-camera setups. Second, for precise visual control, we develop the Subject-Aware World Generator to enable appearance-driven simulation conditioned on per-agent identity images. Finally, to ensure both views are grounded in the identical physical reality, we propose World-State Alignment, a per-frame inter-branch cross-attention mechanism inserted at every transformer layer of the video DiT. By jointly synchronizing the denoising process, WSA enforces both static geometric consistency and dynamic motion consistency, encouraging that the shared 3D environment and physical events remain well-aligned across both egocentric views. Extensive experiments demonstrate that MetaWorld achieves superior cross-view consistency and identity fidelity, establishing a highly scalable, physics-driven paradigm for multi-agent video world modeling.

Tong Zhang, Jiangning Zhang, Zhucun Xue et al.

Balancing convergence speed, generalization capability, and computational efficiency remains a core challenge in deep learning optimization. First-order gradient descent methods, epitomized by stochastic gradient descent (SGD) and Adam, serve as the cornerstone of modern training pipelines. However, large-scale model training, stringent differential privacy requirements, and distributed learning paradigms expose critical limitations in these conventional approaches regarding privacy protection and memory efficiency. To mitigate these bottlenecks, researchers explore second-order optimization techniques to surpass first-order performance ceilings, while zeroth-order methods reemerge to alleviate memory constraints inherent to large-scale training. Despite this proliferation of methodologies, the field lacks a cohesive framework that unifies underlying principles and delineates application scenarios for these disparate approaches. In this work, we retrospectively analyze the evolutionary trajectory of deep learning optimization algorithms and present a comprehensive empirical evaluation of mainstream optimizers across diverse model architectures and training scenarios. We distill key emerging trends and fundamental design trade-offs, pinpointing promising directions for future research. By synthesizing theoretical insights with extensive empirical evidence, we provide actionable guidance for designing next-generation highly efficient, robust, and trustworthy optimization methods. The code is available at https://github.com/APRIL-AIGC/Awesome-Optimizer.

Ying Jin, Jinlong Peng, Qingdong He et al.

The performance of anomaly inspection in industrial manufacturing is constrained by the scarcity of anomaly data. To overcome this challenge, researchers have started employing anomaly generation approaches to augment the anomaly dataset. However, existing anomaly generation methods suffer from limited diversity in the generated anomalies and struggle to achieve a seamless blending of this anomaly with the original image. Moreover, the generated mask is usually not aligned with the generated anomaly. In this paper, we overcome these challenges from a new perspective, simultaneously generating a pair of the overall image and the corresponding anomaly part. We propose DualAnoDiff, a novel diffusion-based few-shot anomaly image generation model, which can generate diverse and realistic anomaly images by using a dual-interrelated diffusion model, where one of them is employed to generate the whole image while the other one generates the anomaly part. Moreover, we extract background and shape information to mitigate the distortion and blurriness phenomenon in few-shot image generation. Extensive experiments demonstrate the superiority of our proposed model over state-of-the-art methods in terms of diversity, realism and the accuracy of mask. Overall, our approach significantly improves the performance of downstream anomaly inspection tasks, including anomaly detection, anomaly localization, and anomaly classification tasks.

Yicheng Xu, Jiangning Zhang, Zhucun Xue et al.

In-context Learning enables training-free adaptation via demonstrations but remains highly sensitive to example selection and formatting. In unified multimodal models spanning understanding and generation, this sensitivity is exacerbated by cross-modal interference and varying cognitive demands. Consequently, In-context Learning efficacy is often non-monotonic and highly task-dependent. To diagnose these behaviors, we introduce a six-level capability-oriented taxonomy that categorizes the functional role of demonstrations from basic perception to high-order discernment. Guided by this cognitive framework, we construct UniICL-760K, a large-scale corpus featuring curated 8-shot In-context Learning episodes across 15 subtasks, alongside UniICL-Bench for rigorous, controlled evaluation. As an architectural intervention to stabilize few-shot adaptation, we propose the Context-Adaptive Prototype Modulator, a lightweight, plug-and-play module. Evaluations on UniICL-Bench show that our approach yields highly competitive unified results, outperforming larger-parameter multimodal large language model baselines on most understanding In-context Learning tasks. Data and code will be available soon at https://github.com/xuyicheng-zju/UniICL.

Teng Hu, Jiangning Zhang, Hongrui Huang et al.

The rapid advancement of Artificial Intelligence Generated Content (AIGC) has revolutionized video generation, enabling systems ranging from proprietary pioneers like OpenAI's Sora, Google's Veo3, and Bytedance's Seedance to powerful open-source contenders like Wan and HunyuanVideo to synthesize temporally coherent and semantically rich videos. These advancements pave the way for building "world models" that simulate real-world dynamics, with applications spanning entertainment, education, and virtual reality. However, existing reviews on video generation often focus on narrow technical fields, e.g., Generative Adversarial Networks (GAN) and diffusion models, or specific tasks (e. g., video editing), lacking a comprehensive perspective on the field's evolution, especially regarding Auto-Regressive (AR) models and integration of multimodal information. To address these gaps, this survey firstly provides a systematic review of the development of video generation technology, tracing its evolution from early GANs to dominant diffusion models, and further to emerging AR-based and multimodal techniques. We conduct an in-depth analysis of the foundational principles, key advancements, and comparative strengths/limitations. Then, we explore emerging trends in multimodal video generation, emphasizing the integration of diverse data types to enhance contextual awareness. Finally, by bridging historical developments and contemporary innovations, this survey offers insights to guide future research in video generation and its applications, including virtual/augmented reality, personalized education, autonomous driving simulations, digital entertainment, and advanced world models, in this rapidly evolving field. For more details, please refer to the project at https://github.com/sjtuplayer/Awesome-Video-Foundations.

Zhucun Xue, Jiangning Zhang, Juntao Jiang et al.

The rapid expansion of research across machine learning, vision, and language has produced a volume of publications that is increasingly difficult to synthesize. Traditional bibliometric tools rely mainly on metadata and offer limited visibility into the semantic content of papers, making it hard to track how research themes evolve over time or how different areas influence one another. To obtain a clearer picture of recent developments, we compile a unified corpus of more than 100,000 papers from 22 major conferences between 2020 and 2025 and construct a multidimensional profiling pipeline to organize and analyze their textual content. By combining topic clustering, LLM-assisted parsing, and structured retrieval, we derive a comprehensive representation of research activity that supports the study of topic lifecycles, methodological transitions, dataset and model usage patterns, and institutional research directions. Our analysis highlights several notable shifts, including the growth of safety, multimodal reasoning, and agent-oriented studies, as well as the gradual stabilization of areas such as neural machine translation and graph-based methods. These findings provide an evidence-based view of how AI research is evolving and offer a resource for understanding broader trends and identifying emerging directions. Code and dataset: https://github.com/xzc-zju/Profiling_Scientific_Literature

Haifeng Wang, Hua Wu, Tian Wu et al.

In this report, we introduce ERNIE 5.0, a natively autoregressive foundation model desinged for unified multimodal understanding and generation across text, image, video, and audio. All modalities are trained from scratch under a unified next-group-of-tokens prediction objective, based on an ultra-sparse mixture-of-experts (MoE) architecture with modality-agnostic expert routing. To address practical challenges in large-scale deployment under diverse resource constraints, ERNIE 5.0 adopts a novel elastic training paradigm. Within a single pre-training run, the model learns a family of sub-models with varying depths, expert capacities, and routing sparsity, enabling flexible trade-offs among performance, model size, and inference latency in memory- or time-constrained scenarios. Moreover, we systematically address the challenges of scaling reinforcement learning to unified foundation models, thereby guaranteeing efficient and stable post-training under ultra-sparse MoE architectures and diverse multimodal settings. Extensive experiments demonstrate that ERNIE 5.0 achieves strong and balanced performance across multiple modalities. To the best of our knowledge, among publicly disclosed models, ERNIE 5.0 represents the first production-scale realization of a trillion-parameter unified autoregressive model that supports both multimodal understanding and generation. To facilitate further research, we present detailed visualizations of modality-agnostic expert routing in the unified model, alongside comprehensive empirical analysis of elastic training, aiming to offer profound insights to the community.

Guozhen Zhang, Zixiang Zhou, Teng Hu et al.

Due to the lack of effective cross-modal modeling, existing open-source audio-video generation methods often exhibit compromised lip synchronization and insufficient semantic consistency. To mitigate these drawbacks, we propose UniAVGen, a unified framework for joint audio and video generation. UniAVGen is anchored in a dual-branch joint synthesis architecture, incorporating two parallel Diffusion Transformers (DiTs) to build a cohesive cross-modal latent space. At its heart lies an Asymmetric Cross-Modal Interaction mechanism, which enables bidirectional, temporally aligned cross-attention, thus ensuring precise spatiotemporal synchronization and semantic consistency. Furthermore, this cross-modal interaction is augmented by a Face-Aware Modulation module, which dynamically prioritizes salient regions in the interaction process. To enhance generative fidelity during inference, we additionally introduce Modality-Aware Classifier-Free Guidance, a novel strategy that explicitly amplifies cross-modal correlation signals. Notably, UniAVGen's robust joint synthesis design enables seamless unification of pivotal audio-video tasks within a single model, such as joint audio-video generation and continuation, video-to-audio dubbing, and audio-driven video synthesis. Comprehensive experiments validate that, with far fewer training samples (1.3M vs. 30.1M), UniAVGen delivers overall advantages in audio-video synchronization, timbre consistency, and emotion consistency.

Teng Hu, Jiangning Zhang, Ran Yi et al.

In recent years, the development of diffusion models has led to significant progress in image and video generation tasks, with pre-trained models like the Stable Diffusion series playing a crucial role. Inspired by model pruning which lightens large pre-trained models by removing unimportant parameters, we propose a novel model fine-tuning method to make full use of these ineffective parameters and enable the pre-trained model with new task-specified capabilities. In this work, we first investigate the importance of parameters in pre-trained diffusion models, and discover that the smallest 10% to 20% of parameters by absolute values do not contribute to the generation process. Based on this observation, we propose a method termed SaRA that re-utilizes these temporarily ineffective parameters, equating to optimizing a sparse weight matrix to learn the task-specific knowledge. To mitigate overfitting, we propose a nuclear-norm-based low-rank sparse training scheme for efficient fine-tuning. Furthermore, we design a new progressive parameter adjustment strategy to make full use of the re-trained/finetuned parameters. Finally, we propose a novel unstructural backpropagation strategy, which significantly reduces memory costs during fine-tuning. Our method enhances the generative capabilities of pre-trained models in downstream applications and outperforms traditional fine-tuning methods like LoRA in maintaining model's generalization ability. We validate our approach through fine-tuning experiments on SD models, demonstrating significant improvements. SaRA also offers a practical advantage that requires only a single line of code modification for efficient implementation and is seamlessly compatible with existing methods.

Sijin Wu, Dan Zhang, Teng Hu et al.

In this paper, we propose Docprompt for document question answering tasks with powerful zero-shot and few-shot performance. We proposed a novel weakly supervised data generation method, a novel multl-stage training method and a novel understanding model \& generation model ensemble method. We achieved state-of-the-art performance on 4 document question answering tasks. This method greatly improves the delivery efficiency and model performance of document question answering customer projects, reducing annotation costs and labor costs. Our demo can be found at https://huggingface.co/spaces/PaddlePaddle/ERNIE-Layout.

Jiangning Zhang, Junwei Zhu, Teng Hu et al.

Native 4K (2160$\times$3840) video generation remains a critical challenge due to the quadratic computational explosion of full-attention as spatiotemporal resolution increases, making it difficult for models to strike a balance between efficiency and quality. This paper proposes a novel Transformer retrofit strategy termed $\textbf{T3}$ ($\textbf{T}$ransform $\textbf{T}$rained $\textbf{T}$ransformer) that, without altering the core architecture of full-attention pretrained models, significantly reduces compute requirements by optimizing their forward logic. Specifically, $\textbf{T3-Video}$ introduces a multi-scale weight-sharing window attention mechanism and, via hierarchical blocking together with an axis-preserving full-attention design, can effect an "attention pattern" transformation of a pretrained model using only modest compute and data. Results on 4K-VBench show that $\textbf{T3-Video}$ substantially outperforms existing approaches: while delivering performance improvements (+4.29$\uparrow$ VQA and +0.08$\uparrow$ VTC), it accelerates native 4K video generation by more than 10$\times$. Project page at https://zhangzjn.github.io/projects/T3-Video

Yuheng Chen, Qingdong He, Teng Hu et al.

The landscape of joint audio and video generation has been fundamentally transformed by the advent of powerful foundation models. Despite these strides, achieving cohesive multimodal customization for the simultaneous preservation of visual identities and vocal timbres across multiple interacting subjects remains largely underexplored. To bridge this gap, we present Omni-Customizer, an end-to-end framework targeted at the precise binding and seamless fusion of multimodal identity information. Specifically, we introduce an Omni-Context Fusion (OCF) module that effectively enriches the base textual prompt with dense, multimodal identity cues, along with a Masked TTS Cross-Attention (MTP-CA) mechanism explicitly designed to prevent the severe "speech leakage" problem. Within this architecture, we propose Semantic-Anchored Multimodal RoPE (SA-MRoPE) to anchor visual and audio reference tokens, along with TTS embeddings, to their corresponding semantic descriptions, enabling structured multimodal fusion and robust identity binding. Furthermore, we devise a comprehensive training strategy that incorporates interleaved audio-video scheduling to rapidly adapt the audio branch to multilingual scenarios without degrading foundational priors, and a progressive in-pair to cross-pair curriculum to facilitate the learning of high-level and robust identity features. Extensive experiments demonstrate that Omni-Customizer achieves state-of-the-art performance in dual-modal customized generation, excelling across visual identity similarity, timbre consistency, precise audio-video synchronization, and overall video-audio fidelity.

Haokun Zhu, Juang Ian Chong, Teng Hu et al.

Vector graphics are widely used in graphical designs and have received more and more attention. However, unlike raster images which can be easily obtained, acquiring high-quality vector graphics, typically through automatically converting from raster images remains a significant challenge, especially for more complex images such as photos or artworks. In this paper, we propose SAMVG, a multi-stage model to vectorize raster images into SVG (Scalable Vector Graphics). Firstly, SAMVG uses general image segmentation provided by the Segment-Anything Model and uses a novel filtering method to identify the best dense segmentation map for the entire image. Secondly, SAMVG then identifies missing components and adds more detailed components to the SVG. Through a series of extensive experiments, we demonstrate that SAMVG can produce high quality SVGs in any domain while requiring less computation time and complexity compared to previous state-of-the-art methods.

Zhucun Xue, Jiangning Zhang, Teng Hu et al.

The quality of the video dataset (image quality, resolution, and fine-grained caption) greatly influences the performance of the video generation model. The growing demand for video applications sets higher requirements for high-quality video generation models. For example, the generation of movie-level Ultra-High Definition (UHD) videos and the creation of 4K short video content. However, the existing public datasets cannot support related research and applications. In this paper, we first propose a high-quality open-sourced UHD-4K (22.4\% of which are 8K) text-to-video dataset named UltraVideo, which contains a wide range of topics (more than 100 kinds), and each video has 9 structured captions with one summarized caption (average of 824 words). Specifically, we carefully design a highly automated curation process with four stages to obtain the final high-quality dataset: \textit{i)} collection of diverse and high-quality video clips. \textit{ii)} statistical data filtering. \textit{iii)} model-based data purification. \textit{iv)} generation of comprehensive, structured captions. In addition, we expand Wan to UltraWan-1K/-4K, which can natively generate high-quality 1K/4K videos with more consistent text controllability, demonstrating the effectiveness of our data curation.We believe that this work can make a significant contribution to future research on UHD video generation. UltraVideo dataset and UltraWan models are available at https://xzc-zju.github.io/projects/UltraVideo.

Ruiyan Wang, Teng Hu, Kaihui Huang et al.

Pose-guided video generation refers to controlling the motion of subjects in generated video through a sequence of poses. It enables precise control over subject motion and has important applications in animation. However, current pose-guided video generation methods are limited to accepting only human poses as input, thus generalizing poorly to pose of other subjects. To address this issue, we propose PoseAnything, the first universal pose-guided video generation framework capable of handling both human and non-human characters, supporting arbitrary skeletal inputs. To enhance consistency preservation during motion, we introduce Part-aware Temporal Coherence Module, which divides the subject into different parts, establishes part correspondences, and computes cross-attention between corresponding parts across frames to achieve fine-grained part-level consistency. Additionally, we propose Subject and Camera Motion Decoupled CFG, a novel guidance strategy that, for the first time, enables independent camera movement control in pose-guided video generation, by separately injecting subject and camera motion control information into the positive and negative anchors of CFG. Furthermore, we present XPose, a high-quality public dataset containing 50,000 non-human pose-video pairs, along with an automated pipeline for annotation and filtering. Extensive experiments demonstrate that Pose-Anything significantly outperforms state-of-the-art methods in both effectiveness and generalization.

Teng Hu, Jiangning Zhang, Ran Yi et al.

Employing LLMs for visual generation has recently become a research focus. However, the existing methods primarily transfer the LLM architecture to visual generation but rarely investigate the fundamental differences between language and vision. This oversight may lead to suboptimal utilization of visual generation capabilities within the LLM framework. In this paper, we explore the characteristics of visual embedding space under the LLM framework and discover that the correlation between visual embeddings can help achieve more stable and robust generation results. We present IAR, an Improved AutoRegressive Visual Generation Method that enhances the training efficiency and generation quality of LLM-based visual generation models. Firstly, we propose a Codebook Rearrangement strategy that uses balanced k-means clustering algorithm to rearrange the visual codebook into clusters, ensuring high similarity among visual features within each cluster. Leveraging the rearranged codebook, we propose a Cluster-oriented Cross-entropy Loss that guides the model to correctly predict the cluster where the token is located. This approach ensures that even if the model predicts the wrong token index, there is a high probability the predicted token is located in the correct cluster, which significantly enhances the generation quality and robustness. Extensive experiments demonstrate that our method consistently enhances the model training efficiency and performance from 100M to 1.4B, reducing the training time by half while achieving the same FID. Additionally, our approach can be applied to various LLM-based visual generation models and adheres to the scaling law, providing a promising direction for future research in LLM-based visual generation. The code is available at: https://github.com/sjtuplayer/IAR.

Sen Liang, Zhentao Yu, Zhengguang Zhou et al.

The emergence of Diffusion Transformers (DiT) has brought significant advancements to video generation, especially in text-to-video and image-to-video tasks. Although video generation is widely applied in various fields, most existing models are limited to single scenarios and cannot perform diverse video generation and editing through dynamic content manipulation. We propose OmniV2V, a video model capable of generating and editing videos across different scenarios based on various operations, including: object movement, object addition, mask-guided video edit, try-on, inpainting, outpainting, human animation, and controllable character video synthesis. We explore a unified dynamic content manipulation injection module, which effectively integrates the requirements of the above tasks. In addition, we design a visual-text instruction module based on LLaVA, enabling the model to effectively understand the correspondence between visual content and instructions. Furthermore, we build a comprehensive multi-task data processing system. Since there is data overlap among various tasks, this system can efficiently provide data augmentation. Using this system, we construct a multi-type, multi-scenario OmniV2V dataset and its corresponding OmniV2V-Test benchmark. Extensive experiments show that OmniV2V works as well as, and sometimes better than, the best existing open-source and commercial models for many video generation and editing tasks.

Teng Hu, Zhentao Yu, Zhengguang Zhou et al.

Despite recent advances in video generation, existing models still lack fine-grained controllability, especially for multi-subject customization with consistent identity and interaction. In this paper, we propose PolyVivid, a multi-subject video customization framework that enables flexible and identity-consistent generation. To establish accurate correspondences between subject images and textual entities, we design a VLLM-based text-image fusion module that embeds visual identities into the textual space for precise grounding. To further enhance identity preservation and subject interaction, we propose a 3D-RoPE-based enhancement module that enables structured bidirectional fusion between text and image embeddings. Moreover, we develop an attention-inherited identity injection module to effectively inject fused identity features into the video generation process, mitigating identity drift. Finally, we construct an MLLM-based data pipeline that combines MLLM-based grounding, segmentation, and a clique-based subject consolidation strategy to produce high-quality multi-subject data, effectively enhancing subject distinction and reducing ambiguity in downstream video generation. Extensive experiments demonstrate that PolyVivid achieves superior performance in identity fidelity, video realism, and subject alignment, outperforming existing open-source and commercial baselines.

Jiangning Zhang, Teng Hu, Haoyang He et al.

This work focuses on developing parameter-efficient and lightweight models for dense predictions while trading off parameters, FLOPs, and performance. Our goal is to set up the new frontier of the 5M magnitude lightweight model on various downstream tasks. Inverted Residual Block (IRB) serves as the infrastructure for lightweight CNNs, but no counterparts have been recognized by attention-based design. Our work rethinks the lightweight infrastructure of efficient IRB and practical components in Transformer from a unified perspective, extending CNN-based IRB to attention-based models and abstracting a one-residual Meta Mobile Block (MMBlock) for lightweight model design. Following neat but effective design criterion, we deduce a modern Improved Inverted Residual Mobile Block (i2RMB) and improve a hierarchical Efficient MOdel (EMOv2) with no elaborate complex structures. Considering the imperceptible latency for mobile users when downloading models under 4G/5G bandwidth and ensuring model performance, we investigate the performance upper limit of lightweight models with a magnitude of 5M. Extensive experiments on various vision recognition, dense prediction, and image generation tasks demonstrate the superiority of our EMOv2 over state-of-the-art methods, e.g., EMOv2-1M/2M/5M achieve 72.3, 75.8, and 79.4 Top-1 that surpass equal-order CNN-/Attention-based models significantly. At the same time, EMOv2-5M equipped RetinaNet achieves 41.5 mAP for object detection tasks that surpasses the previous EMO-5M by +2.6. When employing the more robust training recipe, our EMOv2-5M eventually achieves 82.9 Top-1 accuracy, which elevates the performance of 5M magnitude models to a new level. Code is available at https://github.com/zhangzjn/EMOv2.

Youliang Zhang, Zhengguang Zhou, Zhentao Yu et al.

Generating talking avatars is a fundamental task in video generation. Although existing methods can generate full-body talking avatars with simple human motion, extending this task to grounded human-object interaction (GHOI) remains an open challenge, requiring the avatar to perform text-aligned interactions with surrounding objects. This challenge stems from the need for environmental perception and the control-quality dilemma in GHOI generation. To address this, we propose a novel dual-stream framework, InteractAvatar, which decouples perception and planning from video synthesis for grounded human-object interaction. Leveraging detection to enhance environmental perception, we introduce a Perception and Interaction Module (PIM) to generate text-aligned interaction motions. Additionally, an Audio-Interaction Aware Generation Module (AIM) is proposed to synthesize vivid talking avatars performing object interactions. With a specially designed motion-to-video aligner, PIM and AIM share a similar network structure and enable parallel co-generation of motions and plausible videos, effectively mitigating the control-quality dilemma. Finally, we establish a benchmark, GroundedInter, for evaluating GHOI video generation. Extensive experiments and comparisons demonstrate the effectiveness of our method in generating grounded human-object interactions for talking avatars. Project page: https://interactavatar.github.io

Teng Hu, Zhentao Yu, Guozhen Zhang et al.

The synthesis of synchronized audio-visual content is a key challenge in generative AI, with open-source models facing challenges in robust audio-video alignment. Our analysis reveals that this issue is rooted in three fundamental challenges of the joint diffusion process: (1) Correspondence Drift, where concurrently evolving noisy latents impede stable learning of alignment; (2) inefficient global attention mechanisms that fail to capture fine-grained temporal cues; and (3) the intra-modal bias of conventional Classifier-Free Guidance (CFG), which enhances conditionality but not cross-modal synchronization. To overcome these challenges, we introduce Harmony, a novel framework that mechanistically enforces audio-visual synchronization. We first propose a Cross-Task Synergy training paradigm to mitigate drift by leveraging strong supervisory signals from audio-driven video and video-driven audio generation tasks. Then, we design a Global-Local Decoupled Interaction Module for efficient and precise temporal-style alignment. Finally, we present a novel Synchronization-Enhanced CFG (SyncCFG) that explicitly isolates and amplifies the alignment signal during inference. Extensive experiments demonstrate that Harmony establishes a new state-of-the-art, significantly outperforming existing methods in both generation fidelity and, critically, in achieving fine-grained audio-visual synchronization.

Teng Hu, Ran Yi, Baihong Qian et al.

SVG (Scalable Vector Graphics) is a widely used graphics format that possesses excellent scalability and editability. Image vectorization, which aims to convert raster images to SVGs, is an important yet challenging problem in computer vision and graphics. Existing image vectorization methods either suffer from low reconstruction accuracy for complex images or require long computation time. To address this issue, we propose SuperSVG, a superpixel-based vectorization model that achieves fast and high-precision image vectorization. Specifically, we decompose the input image into superpixels to help the model focus on areas with similar colors and textures. Then, we propose a two-stage self-training framework, where a coarse-stage model is employed to reconstruct the main structure and a refinement-stage model is used for enriching the details. Moreover, we propose a novel dynamic path warping loss to help the refinement-stage model to inherit knowledge from the coarse-stage model. Extensive qualitative and quantitative experiments demonstrate the superior performance of our method in terms of reconstruction accuracy and inference time compared to state-of-the-art approaches. The code is available in \url{https://github.com/sjtuplayer/SuperSVG}.

Yinan Chen, Jiangning Zhang, Yali Bi et al.

Image inversion is a fundamental task in generative models, aiming to map images back to their latent representations to enable downstream applications such as editing, restoration, and style transfer. This paper provides a comprehensive review of the latest advancements in image inversion techniques, focusing on two main paradigms: Generative Adversarial Network (GAN) inversion and diffusion model inversion. We categorize these techniques based on their optimization methods. For GAN inversion, we systematically classify existing methods into encoder-based approaches, latent optimization approaches, and hybrid approaches, analyzing their theoretical foundations, technical innovations, and practical trade-offs. For diffusion model inversion, we explore training-free strategies, fine-tuning methods, and the design of additional trainable modules, highlighting their unique advantages and limitations. Additionally, we discuss several popular downstream applications and emerging applications beyond image tasks, identifying current challenges and future research directions. By synthesizing the latest developments, this paper aims to provide researchers and practitioners with a valuable reference resource, promoting further advancements in the field of image inversion. We keep track of the latest works at https://github.com/RyanChenYN/ImageInversion

Teng Hu, Jiangning Zhang, Ran Yi et al.

Anomaly inspection plays an important role in industrial manufacture. Existing anomaly inspection methods are limited in their performance due to insufficient anomaly data. Although anomaly generation methods have been proposed to augment the anomaly data, they either suffer from poor generation authenticity or inaccurate alignment between the generated anomalies and masks. To address the above problems, we propose AnomalyDiffusion, a novel diffusion-based few-shot anomaly generation model, which utilizes the strong prior information of latent diffusion model learned from large-scale dataset to enhance the generation authenticity under few-shot training data. Firstly, we propose Spatial Anomaly Embedding, which consists of a learnable anomaly embedding and a spatial embedding encoded from an anomaly mask, disentangling the anomaly information into anomaly appearance and location information. Moreover, to improve the alignment between the generated anomalies and the anomaly masks, we introduce a novel Adaptive Attention Re-weighting Mechanism. Based on the disparities between the generated anomaly image and normal sample, it dynamically guides the model to focus more on the areas with less noticeable generated anomalies, enabling generation of accurately-matched anomalous image-mask pairs. Extensive experiments demonstrate that our model significantly outperforms the state-of-the-art methods in generation authenticity and diversity, and effectively improves the performance of downstream anomaly inspection tasks. The code and data are available in https://github.com/sjtuplayer/anomalydiffusion.

Teng Hu, Jiangning Zhang, Ran Yi et al.

The emergence of diffusion models has greatly propelled the progress in image and video generation. Recently, some efforts have been made in controllable video generation, including text-to-video generation and video motion control, among which camera motion control is an important topic. However, existing camera motion control methods rely on training a temporal camera module, and necessitate substantial computation resources due to the large amount of parameters in video generation models. Moreover, existing methods pre-define camera motion types during training, which limits their flexibility in camera control. Therefore, to reduce training costs and achieve flexible camera control, we propose COMD, a novel training-free video motion transfer model, which disentangles camera motions and object motions in source videos and transfers the extracted camera motions to new videos. We first propose a one-shot camera motion disentanglement method to extract camera motion from a single source video, which separates the moving objects from the background and estimates the camera motion in the moving objects region based on the motion in the background by solving a Poisson equation. Furthermore, we propose a few-shot camera motion disentanglement method to extract the common camera motion from multiple videos with similar camera motions, which employs a window-based clustering technique to extract the common features in temporal attention maps of multiple videos. Finally, we propose a motion combination method to combine different types of camera motions together, enabling our model a more controllable and flexible camera control. Extensive experiments demonstrate that our training-free approach can effectively decouple camera-object motion and apply the decoupled camera motion to a wide range of controllable video generation tasks, achieving flexible and diverse camera motion control.

Teng Hu, Zhentao Yu, Zhengguang Zhou et al.

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.

Yizhe Tang, Zhimin Sun, Yuzhen Du et al.

Image inpainting aims to fill the missing region of an image. Recently, there has been a surge of interest in foreground-conditioned background inpainting, a sub-task that fills the background of an image while the foreground subject and associated text prompt are provided. Existing background inpainting methods typically strictly preserve the subject's original position from the source image, resulting in inconsistencies between the subject and the generated background. To address this challenge, we propose a new task, the "Text-Guided Subject-Position Variable Background Inpainting", which aims to dynamically adjust the subject position to achieve a harmonious relationship between the subject and the inpainted background, and propose the Adaptive Transformation Agent (A$^\text{T}$A) for this task. Firstly, we design a PosAgent Block that adaptively predicts an appropriate displacement based on given features to achieve variable subject-position. Secondly, we design the Reverse Displacement Transform (RDT) module, which arranges multiple PosAgent blocks in a reverse structure, to transform hierarchical feature maps from deep to shallow based on semantic information. Thirdly, we equip A$^\text{T}$A with a Position Switch Embedding to control whether the subject's position in the generated image is adaptively predicted or fixed. Extensive comparative experiments validate the effectiveness of our A$^\text{T}$A approach, which not only demonstrates superior inpainting capabilities in subject-position variable inpainting, but also ensures good performance on subject-position fixed inpainting.

Yinan Chen, Jiangning Zhang, Teng Hu et al.

Instruction-guided video editing has emerged as a rapidly advancing research direction, offering new opportunities for intuitive content transformation while also posing significant challenges for systematic evaluation. Existing video editing benchmarks fail to support the evaluation of instruction-guided video editing adequately and further suffer from limited source diversity, narrow task coverage and incomplete evaluation metrics. To address the above limitations, we introduce IVEBench, a modern benchmark suite specifically designed for instruction-guided video editing assessment. IVEBench comprises a diverse database of 600 high-quality source videos, spanning seven semantic dimensions, and covering video lengths ranging from 32 to 1,024 frames. It further includes 8 categories of editing tasks with 35 subcategories, whose prompts are generated and refined through large language models and expert review. Crucially, IVEBench establishes a three-dimensional evaluation protocol encompassing video quality, instruction compliance and video fidelity, integrating both traditional metrics and multimodal large language model-based assessments. Extensive experiments demonstrate the effectiveness of IVEBench in benchmarking state-of-the-art instruction-guided video editing methods, showing its ability to provide comprehensive and human-aligned evaluation outcomes.

Yuzhen Du, Teng Hu, Jiangning Zhang et al. · tencent-ai

Image restoration (IR) aims to recover high-quality images from degraded inputs, with recent deep learning advancements significantly enhancing performance. However, existing methods lack a unified training benchmark for iterations and configurations. We also identify a bias in image complexity distributions between commonly used IR training and testing datasets, resulting in suboptimal restoration outcomes. To address this, we introduce a large-scale IR dataset called ReSyn, which employs a novel image filtering method based on image complexity to ensure a balanced distribution and includes both real and AIGC synthetic images. We establish a unified training standard that specifies iterations and configurations for image restoration models, focusing on measuring model convergence and restoration capability. Additionally, we enhance transformer-based image restoration models using linear attention mechanisms by proposing RWKV-IR, which integrates linear complexity RWKV into the transformer structure, allowing for both global and local receptive fields. Instead of directly using Vision-RWKV, we replace the original Q-Shift in RWKV with a Depth-wise Convolution shift to better model local dependencies, combined with Bi-directional attention for comprehensive linear attention. We also introduce a Cross-Bi-WKV module that merges two Bi-WKV modules with different scanning orders for balanced horizontal and vertical attention. Extensive experiments validate the effectiveness of our RWKV-IR model.

Yige Chen, Teng Hu, Yizhe Tang et al.

With the help of Score Distillation Sampling (SDS) and the rapid development of neural 3D representations, some methods have been proposed to perform 3D editing such as adding additional geometries, or overwriting textures. However, generalized 3D non-rigid editing task, which requires changing both the structure (posture or composition) and appearance (texture) of the original object, remains to be challenging in 3D editing field. In this paper, we propose Plasticine3D, a novel text-guided fine-grained controlled 3D editing pipeline that can perform 3D non-rigid editing with large structure deformations. Our work divides the editing process into a geometry editing stage and a texture editing stage to achieve separate control of structure and appearance. In order to maintain the details of the original object from different viewpoints, we propose a Multi-View-Embedding (MVE) Optimization strategy to ensure that the guidance model learns the features of the original object from various viewpoints. For the purpose of fine-grained control, we propose Embedding-Fusion (EF) to blend the original characteristics with the editing objectives in the embedding space, and control the extent of editing by adjusting the fusion rate. Furthermore, in order to address the issue of gradual loss of details during the generation process under high editing intensity, as well as the problem of insignificant editing effects in some scenarios, we propose Score Projection Sampling (SPS) as a replacement of score distillation sampling, which introduces additional optimization phases for editing target enhancement and original detail maintenance, leading to better editing quality. Extensive experiments demonstrate the effectiveness of our method on 3D non-rigid editing tasks

Yuheng Chen, Teng Hu, Jiangning Zhang et al.

Recent advances in text-to-video diffusion models have enabled the generation of high-quality videos conditioned on textual descriptions. However, most existing text-to-video models rely solely on textual conditions, lacking general fine-grained controllability over video generation. To address this challenge, we propose InstanceV, a video generation framework that enables i) instance-level control and ii) global semantic consistency. Specifically, with the aid of proposed Instance-aware Masked Cross-Attention mechanism, InstanceV maximizes the utilization of additional instance-level grounding information to generate correctly attributed instances at designated spatial locations. To improve overall consistency, We introduce the Shared Timestep-Adaptive Prompt Enhancement module, which connects local instances with global semantics in a parameter-efficient manner. Furthermore, we incorporate Spatially-Aware Unconditional Guidance during both training and inference to alleviate the disappearance of small instances. Finally, we propose a new benchmark, named InstanceBench, which combines general video quality metrics with instance-aware metrics for more comprehensive evaluation on instance-level video generation. Extensive experiments demonstrate that InstanceV not only achieves remarkable instance-level controllability in video generation, but also outperforms existing state-of-the-art models in both general quality and instance-aware metrics across qualitative and quantitative evaluations.

Teng Hu, Jiangning Zhang, Zihan Su et al.

Recent advances in video generation have made it possible to produce visually compelling videos, with wide-ranging applications in content creation, entertainment, and virtual reality. However, most existing diffusion transformer based video generation models are limited to low-resolution outputs (<=720P) due to the quadratic computational complexity of the attention mechanism with respect to the output width and height. This computational bottleneck makes native high-resolution video generation (1080P/2K/4K) impractical for both training and inference. To address this challenge, we present UltraGen, a novel video generation framework that enables i) efficient and ii) end-to-end native high-resolution video synthesis. Specifically, UltraGen features a hierarchical dual-branch attention architecture based on global-local attention decomposition, which decouples full attention into a local attention branch for high-fidelity regional content and a global attention branch for overall semantic consistency. We further propose a spatially compressed global modeling strategy to efficiently learn global dependencies, and a hierarchical cross-window local attention mechanism to reduce computational costs while enhancing information flow across different local windows. Extensive experiments demonstrate that UltraGen can effectively scale pre-trained low-resolution video models to 1080P and even 4K resolution for the first time, outperforming existing state-of-the-art methods and super-resolution based two-stage pipelines in both qualitative and quantitative evaluations.

Ran Yi, Teng Hu, Zihan Su et al.

Autoregressive models have emerged as a powerful paradigm for visual content creation, but often overlook the intrinsic structural properties of visual data. Our prior work, IAR, initiated a direction to address this by reorganizing the visual codebook based on embedding similarity, thereby improving generation robustness. However, it is constrained by the rigidity of pre-trained codebooks and the inaccuracies of hard, uniform clustering. To overcome these limitations, we propose IAR2, an advanced autoregressive framework that enables a hierarchical semantic-detail synthesis process. At the core of IAR2 is a novel Semantic-Detail Associated Dual Codebook, which decouples image representations into a semantic codebook for global semantic information and a detail codebook for fine-grained refinements. It expands the quantization capacity from a linear to a polynomial scale, significantly enhancing expressiveness. To accommodate this dual representation, we propose a Semantic-Detail Autoregressive Prediction scheme coupled with a Local-Context Enhanced Autoregressive Head, which performs hierarchical prediction-first the semantic token, then the detail token-while leveraging a local context window to enhance spatial coherence. Furthermore, for conditional generation, we introduce a Progressive Attention-Guided Adaptive CFG mechanism that dynamically modulates the guidance scale for each token based on its relevance to the condition and its temporal position in the generation sequence, improving conditional alignment without sacrificing realism. Extensive experiments demonstrate that IAR2 sets a new state-of-the-art for autoregressive image generation, achieving a FID of 1.50 on ImageNet. Our model not only surpasses previous methods in performance but also demonstrates superior computational efficiency, highlighting the effectiveness of our structured, coarse-to-fine generation strategy.

Yizhe Tang, Yue Wang, Teng Hu et al.

Stroke-based Rendering (SBR) aims to decompose an input image into a sequence of parameterized strokes, which can be rendered into a painting that resembles the input image. Recently, Neural Painting methods that utilize deep learning and reinforcement learning models to predict the stroke sequences have been developed, but suffer from longer inference time or unstable training. To address these issues, we propose AttentionPainter, an efficient and adaptive model for single-step neural painting. First, we propose a novel scalable stroke predictor, which predicts a large number of stroke parameters within a single forward process, instead of the iterative prediction of previous Reinforcement Learning or auto-regressive methods, which makes AttentionPainter faster than previous neural painting methods. To further increase the training efficiency, we propose a Fast Stroke Stacking algorithm, which brings 13 times acceleration for training. Moreover, we propose Stroke-density Loss, which encourages the model to use small strokes for detailed information, to help improve the reconstruction quality. Finally, we propose a new stroke diffusion model for both conditional and unconditional stroke-based generation, which denoises in the stroke parameter space and facilitates stroke-based inpainting and editing applications helpful for human artists design. Extensive experiments show that AttentionPainter outperforms the state-of-the-art neural painting methods.