Heung-Yeung Shum

Hao Zhang, Feng Li, Shilong Liu et al. · tsinghua

We present DINO (\textbf{D}ETR with \textbf{I}mproved de\textbf{N}oising anch\textbf{O}r boxes), a state-of-the-art end-to-end object detector. % in this paper. DINO improves over previous DETR-like models in performance and efficiency by using a contrastive way for denoising training, a mixed query selection method for anchor initialization, and a look forward twice scheme for box prediction. DINO achieves $49.4$AP in $12$ epochs and $51.3$AP in $24$ epochs on COCO with a ResNet-50 backbone and multi-scale features, yielding a significant improvement of $\textbf{+6.0}$\textbf{AP} and $\textbf{+2.7}$\textbf{AP}, respectively, compared to DN-DETR, the previous best DETR-like model. DINO scales well in both model size and data size. Without bells and whistles, after pre-training on the Objects365 dataset with a SwinL backbone, DINO obtains the best results on both COCO \texttt{val2017} ($\textbf{63.2}$\textbf{AP}) and \texttt{test-dev} (\textbf{$\textbf{63.3}$AP}). Compared to other models on the leaderboard, DINO significantly reduces its model size and pre-training data size while achieving better results. Our code will be available at \url{https://github.com/IDEACVR/DINO}.

Feng Li, Hao Zhang, Huaizhe xu et al. · tsinghua

In this paper we present Mask DINO, a unified object detection and segmentation framework. Mask DINO extends DINO (DETR with Improved Denoising Anchor Boxes) by adding a mask prediction branch which supports all image segmentation tasks (instance, panoptic, and semantic). It makes use of the query embeddings from DINO to dot-product a high-resolution pixel embedding map to predict a set of binary masks. Some key components in DINO are extended for segmentation through a shared architecture and training process. Mask DINO is simple, efficient, and scalable, and it can benefit from joint large-scale detection and segmentation datasets. Our experiments show that Mask DINO significantly outperforms all existing specialized segmentation methods, both on a ResNet-50 backbone and a pre-trained model with SwinL backbone. Notably, Mask DINO establishes the best results to date on instance segmentation (54.5 AP on COCO), panoptic segmentation (59.4 PQ on COCO), and semantic segmentation (60.8 mIoU on ADE20K) among models under one billion parameters. Code is available at \url{https://github.com/IDEACVR/MaskDINO}.

Chun-Yu Sun, Yu-Qi Yang, Hao-Xiang Guo et al.

The lack of fine-grained 3D shape segmentation data is the main obstacle to developing learning-based 3D segmentation techniques. We propose an effective semi-supervised method for learning 3D segmentations from a few labeled 3D shapes and a large amount of unlabeled 3D data. For the unlabeled data, we present a novel multilevel consistency loss to enforce consistency of network predictions between perturbed copies of a 3D shape at multiple levels: point-level, part-level, and hierarchical level. For the labeled data, we develop a simple yet effective part substitution scheme to augment the labeled 3D shapes with more structural variations to enhance training. Our method has been extensively validated on the task of 3D object semantic segmentation on PartNet and ShapeNetPart, and indoor scene semantic segmentation on ScanNet. It exhibits superior performance to existing semi-supervised and unsupervised pre-training 3D approaches. Our code and trained models are publicly available at https://github.com/isunchy/semi_supervised_3d_segmentation.

Shunlin Lu, Ling-Hao Chen, Ailing Zeng et al. · tsinghua

This work targets a novel text-driven whole-body motion generation task, which takes a given textual description as input and aims at generating high-quality, diverse, and coherent facial expressions, hand gestures, and body motions simultaneously. Previous works on text-driven motion generation tasks mainly have two limitations: they ignore the key role of fine-grained hand and face controlling in vivid whole-body motion generation, and lack a good alignment between text and motion. To address such limitations, we propose a Text-aligned whOle-body Motion generATiOn framework, named HumanTOMATO, which is the first attempt to our knowledge towards applicable holistic motion generation in this research area. To tackle this challenging task, our solution includes two key designs: (1) a Holistic Hierarchical VQ-VAE (aka H$^2$VQ) and a Hierarchical-GPT for fine-grained body and hand motion reconstruction and generation with two structured codebooks; and (2) a pre-trained text-motion-alignment model to help generated motion align with the input textual description explicitly. Comprehensive experiments verify that our model has significant advantages in both the quality of generated motions and their alignment with text.

Saizhuo Wang, Hang Yuan, Leon Zhou et al.

One of the most important tasks in quantitative investment research is mining new alphas (effective trading signals or factors). Traditional alpha mining methods, either hand-crafted factor synthesizing or algorithmic factor mining (e.g., search with genetic programming), have inherent limitations, especially in implementing the ideas of quants. In this work, we propose a new alpha mining paradigm by introducing human-AI interaction, and a novel prompt engineering algorithmic framework to implement this paradigm by leveraging the power of large language models. Moreover, we develop Alpha-GPT, a new interactive alpha mining system framework that provides a heuristic way to ``understand'' the ideas of quant researchers and outputs creative, insightful, and effective alphas. We demonstrate the effectiveness and advantage of Alpha-GPT via a number of alpha mining experiments.

Xingyu Chen, Baoyuan Wang, Heung-Yeung Shum

We present HandAvatar, a novel representation for hand animation and rendering, which can generate smoothly compositional geometry and self-occlusion-aware texture. Specifically, we first develop a MANO-HD model as a high-resolution mesh topology to fit personalized hand shapes. Sequentially, we decompose hand geometry into per-bone rigid parts, and then re-compose paired geometry encodings to derive an across-part consistent occupancy field. As for texture modeling, we propose a self-occlusion-aware shading field (SelF). In SelF, drivable anchors are paved on the MANO-HD surface to record albedo information under a wide variety of hand poses. Moreover, directed soft occupancy is designed to describe the ray-to-surface relation, which is leveraged to generate an illumination field for the disentanglement of pose-independent albedo and pose-dependent illumination. Trained from monocular video data, our HandAvatar can perform free-pose hand animation and rendering while at the same time achieving superior appearance fidelity. We also demonstrate that HandAvatar provides a route for hand appearance editing. Project website: https://seanchenxy.github.io/HandAvatarWeb.

Sheng Yu, Zheng Yuan, Jun Xia et al. · tsinghua

Biomedical knowledge graphs (BioMedKGs) are essential infrastructures for biomedical and healthcare big data and artificial intelligence (AI), facilitating natural language processing, model development, and data exchange. For decades, these knowledge graphs have been developed via expert curation; however, this method can no longer keep up with today's AI development, and a transition to algorithmically generated BioMedKGs is necessary. In this work, we introduce the Biomedical Informatics Ontology System (BIOS), the first large-scale publicly available BioMedKG generated completely by machine learning algorithms. BIOS currently contains 4.1 million concepts, 7.4 million terms in two languages, and 7.3 million relation triplets. We present the methodology for developing BIOS, including the curation of raw biomedical terms, computational identification of synonymous terms and aggregation of these terms to create concept nodes, semantic type classification of the concepts, relation identification, and biomedical machine translation. We provide statistics on the current BIOS content and perform preliminary assessments of term quality, synonym grouping, and relation extraction. The results suggest that machine learning-based BioMedKG development is a viable alternative to traditional expert curation.

Zhirui Zhang, Hongbo Zhang, Haoxiang Fei et al.

Although large language models (LLMs) have demonstrated impressive coding capabilities, their ability to autonomously build production-scale software from explicit specifications remains an open question. We introduce SWE-AGI, an open-source benchmark for evaluating end-to-end, specification-driven construction of software systems written in MoonBit. SWE-AGI tasks require LLM-based agents to implement parsers, interpreters, binary decoders, and SAT solvers strictly from authoritative standards and RFCs under a fixed API scaffold. Each task involves implementing 1,000-10,000 lines of core logic, corresponding to weeks or months of engineering effort for an experienced human developer. By leveraging the nascent MoonBit ecosystem, SWE-AGI minimizes data leakage, forcing agents to rely on long-horizon architectural reasoning rather than code retrieval. Across frontier models, gpt-5.3-codex achieves the best overall performance (solving 19/22 tasks, 86.4%), outperforming claude-opus-4.6 (15/22, 68.2%), and kimi-2.5 exhibits the strongest performance among open-source models. Performance degrades sharply with increasing task difficulty, particularly on hard, specification-intensive systems. Behavioral analysis further reveals that as codebases scale, code reading, rather than writing, becomes the dominant bottleneck in AI-assisted development. Overall, while specification-driven autonomous software engineering is increasingly viable, substantial challenges remain before it can reliably support production-scale development.

Duomin Wang, Yu Deng, Zixin Yin et al.

We present a novel one-shot talking head synthesis method that achieves disentangled and fine-grained control over lip motion, eye gaze&blink, head pose, and emotional expression. We represent different motions via disentangled latent representations and leverage an image generator to synthesize talking heads from them. To effectively disentangle each motion factor, we propose a progressive disentangled representation learning strategy by separating the factors in a coarse-to-fine manner, where we first extract unified motion feature from the driving signal, and then isolate each fine-grained motion from the unified feature. We introduce motion-specific contrastive learning and regressing for non-emotional motions, and feature-level decorrelation and self-reconstruction for emotional expression, to fully utilize the inherent properties of each motion factor in unstructured video data to achieve disentanglement. Experiments show that our method provides high quality speech&lip-motion synchronization along with precise and disentangled control over multiple extra facial motions, which can hardly be achieved by previous methods.

Jiashuo Sun, Chengjin Xu, Lumingyuan Tang et al.

Although large language models (LLMs) have achieved significant success in various tasks, they often struggle with hallucination problems, especially in scenarios requiring deep and responsible reasoning. These issues could be partially addressed by introducing external knowledge graphs (KG) in LLM reasoning. In this paper, we propose a new LLM-KG integrating paradigm ``$\hbox{LLM}\otimes\hbox{KG}$'' which treats the LLM as an agent to interactively explore related entities and relations on KGs and perform reasoning based on the retrieved knowledge. We further implement this paradigm by introducing a new approach called Think-on-Graph (ToG), in which the LLM agent iteratively executes beam search on KG, discovers the most promising reasoning paths, and returns the most likely reasoning results. We use a number of well-designed experiments to examine and illustrate the following advantages of ToG: 1) compared with LLMs, ToG has better deep reasoning power; 2) ToG has the ability of knowledge traceability and knowledge correctability by leveraging LLMs reasoning and expert feedback; 3) ToG provides a flexible plug-and-play framework for different LLMs, KGs and prompting strategies without any additional training cost; 4) the performance of ToG with small LLM models could exceed large LLM such as GPT-4 in certain scenarios and this reduces the cost of LLM deployment and application. As a training-free method with lower computational cost and better generality, ToG achieves overall SOTA in 6 out of 9 datasets where most previous SOTAs rely on additional training.

Yi Ma, Doris Tsao, Heung-Yeung Shum

Ten years into the revival of deep networks and artificial intelligence, we propose a theoretical framework that sheds light on understanding deep networks within a bigger picture of Intelligence in general. We introduce two fundamental principles, Parsimony and Self-consistency, that address two fundamental questions regarding Intelligence: what to learn and how to learn, respectively. We believe the two principles are the cornerstones for the emergence of Intelligence, artificial or natural. While these two principles have rich classical roots, we argue that they can be stated anew in entirely measurable and computable ways. More specifically, the two principles lead to an effective and efficient computational framework, compressive closed-loop transcription, that unifies and explains the evolution of modern deep networks and many artificial intelligence practices. While we mainly use modeling of visual data as an example, we believe the two principles will unify understanding of broad families of autonomous intelligent systems and provide a framework for understanding the brain.

Jian Guo, Saizhuo Wang, Lionel M. Ni et al.

Quantitative investment (``quant'') is an interdisciplinary field combining financial engineering, computer science, mathematics, statistics, etc. Quant has become one of the mainstream investment methodologies over the past decades, and has experienced three generations: Quant 1.0, trading by mathematical modeling to discover mis-priced assets in markets; Quant 2.0, shifting quant research pipeline from small ``strategy workshops'' to large ``alpha factories''; Quant 3.0, applying deep learning techniques to discover complex nonlinear pricing rules. Despite its advantage in prediction, deep learning relies on extremely large data volume and labor-intensive tuning of ``black-box'' neural network models. To address these limitations, in this paper, we introduce Quant 4.0 and provide an engineering perspective for next-generation quant. Quant 4.0 has three key differentiating components. First, automated AI changes quant pipeline from traditional hand-craft modeling to the state-of-the-art automated modeling, practicing the philosophy of ``algorithm produces algorithm, model builds model, and eventually AI creates AI''. Second, explainable AI develops new techniques to better understand and interpret investment decisions made by machine learning black-boxes, and explains complicated and hidden risk exposures. Third, knowledge-driven AI is a supplement to data-driven AI such as deep learning and it incorporates prior knowledge into modeling to improve investment decision, in particular for quantitative value investing. Moreover, we discuss how to build a system that practices the Quant 4.0 concept. Finally, we propose ten challenging research problems for quant technology, and discuss potential solutions, research directions, and future trends.

Jingcheng Hu, Yinmin Zhang, Shijie Shang et al.

We introduce Parallel Coordinated Reasoning (PaCoRe), a training-and-inference framework designed to overcome a central limitation of contemporary language models: their inability to scale test-time compute (TTC) far beyond sequential reasoning under a fixed context window. PaCoRe departs from the traditional sequential paradigm by driving TTC through massive parallel exploration coordinated via a message-passing architecture in multiple rounds. Each round launches many parallel reasoning trajectories, compacts their findings into context-bounded messages, and synthesizes these messages to guide the next round and ultimately produce the final answer. Trained end-to-end with large-scale, outcome-based reinforcement learning, the model masters the synthesis abilities required by PaCoRe and scales to multi-million-token effective TTC without exceeding context limits. The approach yields strong improvements across diverse domains, and notably pushes reasoning beyond frontier systems in mathematics: an 8B model reaches 94.5% on HMMT 2025, surpassing GPT-5's 93.2% by scaling effective TTC to roughly two million tokens. We open-source model checkpoints, training data, and the full inference pipeline to accelerate follow-up work.

Shengming Yin, Zekai Zhang, Zecheng Tang et al.

Recent visual generative models often struggle with consistency during image editing due to the entangled nature of raster images, where all visual content is fused into a single canvas. In contrast, professional design tools employ layered representations, allowing isolated edits while preserving consistency. Motivated by this, we propose \textbf{Qwen-Image-Layered}, an end-to-end diffusion model that decomposes a single RGB image into multiple semantically disentangled RGBA layers, enabling \textbf{inherent editability}, where each RGBA layer can be independently manipulated without affecting other content. To support variable-length decomposition, we introduce three key components: (1) an RGBA-VAE to unify the latent representations of RGB and RGBA images; (2) a VLD-MMDiT (Variable Layers Decomposition MMDiT) architecture capable of decomposing a variable number of image layers; and (3) a Multi-stage Training strategy to adapt a pretrained image generation model into a multilayer image decomposer. Furthermore, to address the scarcity of high-quality multilayer training images, we build a pipeline to extract and annotate multilayer images from Photoshop documents (PSD). Experiments demonstrate that our method significantly surpasses existing approaches in decomposition quality and establishes a new paradigm for consistent image editing. Our code and models are released on \href{https://github.com/QwenLM/Qwen-Image-Layered}{https://github.com/QwenLM/Qwen-Image-Layered}

Xili Dai, Ke Chen, Shengbang Tong et al.

Autoencoding has achieved great empirical success as a framework for learning generative models for natural images. Autoencoders often use generic deep networks as the encoder or decoder, which are difficult to interpret, and the learned representations lack clear structure. In this work, we make the explicit assumption that the image distribution is generated from a multi-stage sparse deconvolution. The corresponding inverse map, which we use as an encoder, is a multi-stage convolution sparse coding (CSC), with each stage obtained from unrolling an optimization algorithm for solving the corresponding (convexified) sparse coding program. To avoid computational difficulties in minimizing distributional distance between the real and generated images, we utilize the recent closed-loop transcription (CTRL) framework that optimizes the rate reduction of the learned sparse representations. Conceptually, our method has high-level connections to score-matching methods such as diffusion models. Empirically, our framework demonstrates competitive performance on large-scale datasets, such as ImageNet-1K, compared to existing autoencoding and generative methods under fair conditions. Even with simpler networks and fewer computational resources, our method demonstrates high visual quality in regenerated images. More surprisingly, the learned autoencoder performs well on unseen datasets. Our method enjoys several side benefits, including more structured and interpretable representations, more stable convergence, and scalability to large datasets. Our method is arguably the first to demonstrate that a concatenation of multiple convolution sparse coding/decoding layers leads to an interpretable and effective autoencoder for modeling the distribution of large-scale natural image datasets.

Yu Deng, Baoyuan Wang, Heung-Yeung Shum

A key challenge for novel view synthesis of monocular portrait images is 3D consistency under continuous pose variations. Most existing methods rely on 2D generative models which often leads to obvious 3D inconsistency artifacts. We present a 3D-consistent novel view synthesis approach for monocular portrait images based on a recent proposed 3D-aware GAN, namely Generative Radiance Manifolds (GRAM), which has shown strong 3D consistency at multiview image generation of virtual subjects via the radiance manifolds representation. However, simply learning an encoder to map a real image into the latent space of GRAM can only reconstruct coarse radiance manifolds without faithful fine details, while improving the reconstruction fidelity via instance-specific optimization is time-consuming. We introduce a novel detail manifolds reconstructor to learn 3D-consistent fine details on the radiance manifolds from monocular images, and combine them with the coarse radiance manifolds for high-fidelity reconstruction. The 3D priors derived from the coarse radiance manifolds are used to regulate the learned details to ensure reasonable synthesized results at novel views. Trained on in-the-wild 2D images, our method achieves high-fidelity and 3D-consistent portrait synthesis largely outperforming the prior art.

Yukai Shi, Jianan Wang, He Cao et al.

In this paper, we present TOSS, which introduces text to the task of novel view synthesis (NVS) from just a single RGB image. While Zero-1-to-3 has demonstrated impressive zero-shot open-set NVS capability, it treats NVS as a pure image-to-image translation problem. This approach suffers from the challengingly under-constrained nature of single-view NVS: the process lacks means of explicit user control and often results in implausible NVS generations. To address this limitation, TOSS uses text as high-level semantic information to constrain the NVS solution space. TOSS fine-tunes text-to-image Stable Diffusion pre-trained on large-scale text-image pairs and introduces modules specifically tailored to image and camera pose conditioning, as well as dedicated training for pose correctness and preservation of fine details. Comprehensive experiments are conducted with results showing that our proposed TOSS outperforms Zero-1-to-3 with more plausible, controllable and multiview-consistent NVS results. We further support these results with comprehensive ablations that underscore the effectiveness and potential of the introduced semantic guidance and architecture design.

Xiaohang Ren, Xingyu Chen, Pengfei Yao et al.

The SOTA face swap models still suffer the problem of either target identity (i.e., shape) being leaked or the target non-identity attributes (i.e., background, hair) failing to be fully preserved in the final results. We show that this insufficient disentanglement is caused by two flawed designs that were commonly adopted in prior models: (1) counting on only one compressed encoder to represent both the semantic-level non-identity facial attributes(i.e., pose) and the pixel-level non-facial region details, which is contradictory to satisfy at the same time; (2) highly relying on long skip-connections between the encoder and the final generator, leaking a certain amount of target face identity into the result. To fix them, we introduce a new face swap framework called 'WSC-swap' that gets rid of skip connections and uses two target encoders to respectively capture the pixel-level non-facial region attributes and the semantic non-identity attributes in the face region. To further reinforce the disentanglement learning for the target encoder, we employ both identity removal loss via adversarial training (i.e., GAN) and the non-identity preservation loss via prior 3DMM models like [11]. Extensive experiments on both FaceForensics++ and CelebA-HQ show that our results significantly outperform previous works on a rich set of metrics, including one novel metric for measuring identity consistency that was completely neglected before.

Ailin Huang, Ang Li, Aobo Kong et al.

We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: sharp reasoning and fast, reliable execution. Step 3.5 Flash pairs a 196B-parameter foundation with 11B active parameters for efficient inference. It is optimized with interleaved 3:1 sliding-window/full attention and Multi-Token Prediction (MTP-3) to reduce the latency and cost of multi-round agentic interactions. To reach frontier-level intelligence, we design a scalable reinforcement learning framework that combines verifiable signals with preference feedback, while remaining stable under large-scale off-policy training, enabling consistent self-improvement across mathematics, code, and tool use. Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and 51.0% on Terminal-Bench 2.0, comparable to frontier models such as GPT-5.2 xHigh and Gemini 3.0 Pro. By redefining the efficiency frontier, Step 3.5 Flash provides a high-density foundation for deploying sophisticated agents in real-world industrial environments.

Jingcheng Hu, Yinmin Zhang, Qi Han et al.

We introduce Open-Reasoner-Zero, the first open source implementation of large-scale reasoning-oriented RL training on the base model focusing on scalability, simplicity and accessibility. Through extensive experiments, we demonstrate that a minimalist approach, vanilla PPO with GAE ($λ=1$, $γ=1$) and straightforward rule-based rewards, without any KL regularization, is sufficient to scale up both benchmark performance and response length, replicating the scaling phenomenon observed in DeepSeek-R1-Zero. Using the same base model, Qwen2.5-32B base, as DeepSeek-R1-Zero-Qwen-32B, our implementation achieves superior performance across AIME2024, MATH500, and GPQA Diamond, while demonstrating remarkable efficiency, requiring only 1/10 of the training steps compared to the DeepSeek-R1-Zero pipeline. Moreover, our analysis not only covers training dynamics and ablation for critical design choices, but also quantitatively shows how the learned critic in Reasoner-Zero training effectively identifies and devalues repetitive response patterns, yielding more robust advantage estimations and enhancing training stability. Embracing the principles of open-source, we release our source code, training data, and various model weights, fostering reproducibility and encouraging further exploration of the properties of related models.

Lei Zhang, Heung-Yeung Shum

This paper revisits the principle of uniform convergence in statistical learning, discusses how it acts as the foundation behind machine learning, and attempts to gain a better understanding of the essential problem that current deep learning algorithms are solving. Using computer vision as an example domain in machine learning, the discussion shows that recent research trends in leveraging increasingly large-scale data to perform pre-training for representation learning are largely to reduce the discrepancy between a practically tractable empirical loss and its ultimately desired but intractable expected loss. Furthermore, this paper suggests a few future research directions, predicts the continued increase of data, and argues that more fundamental research is needed on robustness, interpretability, and reasoning capabilities of machine learning by incorporating structure and knowledge.

Jian Guo, Heung-Yeung Shum

Traditional quantitative investment research is encountering diminishing returns alongside rising labor and time costs. To overcome these challenges, we introduce the Large Investment Model (LIM), a novel research paradigm designed to enhance both performance and efficiency at scale. LIM employs end-to-end learning and universal modeling to create an upstream foundation model capable of autonomously learning comprehensive signal patterns from diverse financial data spanning multiple exchanges, instruments, and frequencies. These "global patterns" are subsequently transferred to downstream strategy modeling, optimizing performance for specific tasks. We detail the system architecture design of LIM, address the technical challenges inherent in this approach, and outline potential directions for future research. The advantages of LIM are demonstrated through a series of numerical experiments on cross-instrument prediction for commodity futures trading, leveraging insights from stock markets.

Guoqing Ma, Haoyang Huang, Kun Yan et al.

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.

Ailin Huang, Boyong Wu, Bruce Wang et al.

Real-time speech interaction, serving as a fundamental interface for human-machine collaboration, holds immense potential. However, current open-source models face limitations such as high costs in voice data collection, weakness in dynamic control, and limited intelligence. To address these challenges, this paper introduces Step-Audio, the first production-ready open-source solution. Key contributions include: 1) a 130B-parameter unified speech-text multi-modal model that achieves unified understanding and generation, with the Step-Audio-Chat version open-sourced; 2) a generative speech data engine that establishes an affordable voice cloning framework and produces the open-sourced lightweight Step-Audio-TTS-3B model through distillation; 3) an instruction-driven fine control system enabling dynamic adjustments across dialects, emotions, singing, and RAP; 4) an enhanced cognitive architecture augmented with tool calling and role-playing abilities to manage complex tasks effectively. Based on our new StepEval-Audio-360 evaluation benchmark, Step-Audio achieves state-of-the-art performance in human evaluations, especially in terms of instruction following. On open-source benchmarks like LLaMA Question, shows 9.3% average performance improvement, demonstrating our commitment to advancing the development of open-source multi-modal language technologies. Our code and models are available at https://github.com/stepfun-ai/Step-Audio.

Boyong Wu, Chao Yan, Chen Hu et al.

This paper presents Step-Audio 2, an end-to-end multi-modal large language model designed for industry-strength audio understanding and speech conversation. By integrating a latent audio encoder and reasoning-centric reinforcement learning (RL), Step-Audio 2 achieves promising performance in automatic speech recognition (ASR) and audio understanding. To facilitate genuine end-to-end speech conversation, Step-Audio 2 incorporates the generation of discrete audio tokens into language modeling, significantly enhancing its responsiveness to paralinguistic information such as speaking styles and emotions. To effectively leverage the rich textual and acoustic knowledge in real-world data, Step-Audio 2 integrates retrieval-augmented generation (RAG) and is able to call external tools such as web search to mitigate hallucination and audio search to switch timbres. Trained on millions of hours of speech and audio data, Step-Audio 2 delivers intelligence and expressiveness across diverse conversational scenarios. Evaluation results demonstrate that Step-Audio 2 achieves state-of-the-art performance on various audio understanding and conversational benchmarks compared to other open-source and commercial solutions. Please visit https://github.com/stepfun-ai/Step-Audio2 for more information.

Haoyang Huang, Guoqing Ma, Nan Duan et al.

We present Step-Video-TI2V, a state-of-the-art text-driven image-to-video generation model with 30B parameters, capable of generating videos up to 102 frames based on both text and image inputs. We build Step-Video-TI2V-Eval as a new benchmark for the text-driven image-to-video task and compare Step-Video-TI2V with open-source and commercial TI2V engines using this dataset. Experimental results demonstrate the state-of-the-art performance of Step-Video-TI2V in the image-to-video generation task. Both Step-Video-TI2V and Step-Video-TI2V-Eval are available at https://github.com/stepfun-ai/Step-Video-TI2V.

Zeming Li, Xiangyue Liu, Xiangyu Zhang et al.

Diffusion models have emerged as powerful generative frameworks, creating data samples by progressively denoising an initial random state. Traditionally, this initial state is sampled from a simple, fixed distribution like isotropic Gaussian, inherently lacking structure and a direct mechanism for external control. While recent efforts have explored ways to introduce controllability into the diffusion process, particularly at the initialization stage, they often rely on deterministic or heuristic approaches. These methods can be suboptimal, lack expressiveness, and are difficult to scale or integrate into more sophisticated optimization frameworks. In this paper, we introduce NoiseAR, a novel method for AutoRegressive Initial Noise Prior for Diffusion Models. Instead of a static, unstructured source, NoiseAR learns to generate a dynamic and controllable prior distribution for the initial noise. We formulate the generation of the initial noise prior's parameters as an autoregressive probabilistic modeling task over spatial patches or tokens. This approach enables NoiseAR to capture complex spatial dependencies and introduce learned structure into the initial state. Crucially, NoiseAR is designed to be conditional, allowing text prompts to directly influence the learned prior, thereby achieving fine-grained control over the diffusion initialization. Our experiments demonstrate that NoiseAR can generate initial noise priors that lead to improved sample quality and enhanced consistency with conditional inputs, offering a powerful, learned alternative to traditional random initialization. A key advantage of NoiseAR is its probabilistic formulation, which naturally supports seamless integration into probabilistic frameworks like Markov Decision Processes and Reinforcement Learning. Our code will be available at https://github.com/HKUST-SAIL/NoiseAR/

Yue Ma, Yingqing He, Hongfa Wang et al.

Despite recent advances in image-to-video generation, better controllability and local animation are less explored. Most existing image-to-video methods are not locally aware and tend to move the entire scene. However, human artists may need to control the movement of different objects or regions. Additionally, current I2V methods require users not only to describe the target motion but also to provide redundant detailed descriptions of frame contents. These two issues hinder the practical utilization of current I2V tools. In this paper, we propose a practical framework, named Follow-Your-Click, to achieve image animation with a simple user click (for specifying what to move) and a short motion prompt (for specifying how to move). Technically, we propose the first-frame masking strategy, which significantly improves the video generation quality, and a motion-augmented module equipped with a short motion prompt dataset to improve the short prompt following abilities of our model. To further control the motion speed, we propose flow-based motion magnitude control to control the speed of target movement more precisely. Our framework has simpler yet precise user control and better generation performance than previous methods. Extensive experiments compared with 7 baselines, including both commercial tools and research methods on 8 metrics, suggest the superiority of our approach. Project Page: https://follow-your-click.github.io/

Youkang Kong, Yang Liu, Yue Dong et al.

3D shapes from scanning, reconstruction, or AI-generated content often lack simple quad mesh layouts -- critical for efficient editing and modeling. Existing quad-remeshing techniques typically produce complex layouts with irregular loops, leading to tedious manual cleanup and extensive algorithm tuning. We introduce SQuadGen, a diffusion-based generative framework that leverages Chart Distance Fields (CDF) to synthesize simple quad layouts on 3D shapes. Our approach addresses two key challenges: (1) the discrete nature of mesh connectivity, which hinders learning, and (2) the scarcity of large-scale datasets with simple quad meshes. To overcome the first, we propose CDF, a continuous surface-based representation enabling effective learning and synthesis of quad layouts. To address the second, we define loop-aware simplicity metrics and construct a large-scale dataset of high-quality quad layouts recovered from public 3D repositories through a robust quad-recovery pipeline. Extensive evaluations across diverse 3D inputs show that SQuadGen consistently outperforms existing methods, producing robust, artist-friendly simple quad layouts.

Deyu Zhou, Quan Sun, Yuang Peng et al. · tsinghua

We introduce MAGI, a hybrid video generation framework that combines masked modeling for intra-frame generation with causal modeling for next-frame generation. Our key innovation, Complete Teacher Forcing (CTF), conditions masked frames on complete observation frames rather than masked ones (namely Masked Teacher Forcing, MTF), enabling a smooth transition from token-level (patch-level) to frame-level autoregressive generation. CTF significantly outperforms MTF, achieving a +23% improvement in FVD scores on first-frame conditioned video prediction. To address issues like exposure bias, we employ targeted training strategies, setting a new benchmark in autoregressive video generation. Experiments show that MAGI can generate long, coherent video sequences exceeding 100 frames, even when trained on as few as 16 frames, highlighting its potential for scalable, high-quality video generation.

StepFun, Bin Wang, Bojun Wang et al.

Large language models (LLMs) face low hardware efficiency during decoding, especially for long-context reasoning tasks. This paper introduces Step-3, a 321B-parameter VLM with hardware-aware model-system co-design optimized for minimizing decoding costs. Step-3 innovates in two key dimensions: (1) A novel Multi-Matrix Factorization Attention (MFA) mechanism that significantly reduces both KV cache size and computation while maintaining high attention expressiveness, and (2) Attention-FFN Disaggregation (AFD), a distributed inference system that decouples attention and Feed-Forward Network (FFN) layers into specialized subsystems. This co-design achieves unprecedented cost efficiency: Step-3 significantly reduces theoretical decoding costs compared with models like DeepSeek-V3 and Qwen3 MoE 235B, with the gains widening at longer context. Step-3 achieves low cost while activating 38B parameters per token (more than DeepSeek-V3 and Qwen3 MoE 235B), demonstrating that hardware-aligned attention arithmetic intensity, MoE sparsity, and AFD are critical to cost-effectiveness. We perform a head-to-head comparison with DeepSeek-V3 in its favorable scenarios. Our implementation on Hopper GPUs achieves a decoding throughput of up to 4,039 tokens per second per GPU under 50ms TPOT SLA (4K context, FP8, no MTP). It is higher than DeepSeek-V3's 2,324 in the same setup and sets a new Pareto frontier for LLM decoding.

Ailin Huang, Bingxin Li, Bruce Wang et al.

Large Audio-Language Models (LALMs) have significantly advanced intelligent human-computer interaction, yet their reliance on text-based outputs limits their ability to generate natural speech responses directly, hindering seamless audio interactions. To address this, we introduce Step-Audio-AQAA, a fully end-to-end LALM designed for Audio Query-Audio Answer (AQAA) tasks. The model integrates a dual-codebook audio tokenizer for linguistic and semantic feature extraction, a 130-billion-parameter backbone LLM and a neural vocoder for high-fidelity speech synthesis. Our post-training approach employs interleaved token-output of text and audio to enhance semantic coherence and combines Direct Preference Optimization (DPO) with model merge to improve performance. Evaluations on the StepEval-Audio-360 benchmark demonstrate that Step-Audio-AQAA excels especially in speech control, outperforming the state-of-art LALMs in key areas. This work contributes a promising solution for end-to-end LALMs and highlights the critical role of token-based vocoder in enhancing overall performance for AQAA tasks.

Yue Ma, Zexuan Yan, Hongyu Liu et al. · tencent-ai

We present Follow-Your-Emoji-Faster, an efficient diffusion-based framework for freestyle portrait animation driven by facial landmarks. The main challenges in this task are preserving the identity of the reference portrait, accurately transferring target expressions, and maintaining long-term temporal consistency while ensuring generation efficiency. To address identity preservation and accurate expression retargeting, we enhance Stable Diffusion with two key components: a expression-aware landmarks as explicit motion signals, which improve motion alignment, support exaggerated expressions, and reduce identity leakage; and a fine-grained facial loss that leverages both expression and facial masks to better capture subtle expressions and faithfully preserve the reference appearance. With these components, our model supports controllable and expressive animation across diverse portrait types, including real faces, cartoons, sculptures, and animals. However, diffusion-based frameworks typically struggle to efficiently generate long-term stable animation results, which remains a core challenge in this task. To address this, we propose a progressive generation strategy for stable long-term animation, and introduce a Taylor-interpolated cache, achieving a 2.6X lossless acceleration. These two strategies ensure that our method produces high-quality results efficiently, making it user-friendly and accessible. Finally, we introduce EmojiBench++, a more comprehensive benchmark comprising diverse portraits, driving videos, and landmark sequences. Extensive evaluations on EmojiBench++ demonstrate that Follow-Your-Emoji-Faster achieves superior performance in both animation quality and controllability. The code, training dataset and benchmark will be found in https://follow-your-emoji.github.io/.

Jingcheng Hu, Houyi Li, Yinmin Zhang et al.

We propose novel attention architectures, Multi-matrix Factorization Attention (MFA) and MFA-Key-Reuse (MFA-KR). Existing variants for standard Multi-Head Attention (MHA), including SOTA methods like MLA, fail to maintain as strong performance under stringent Key-Value cache (KV cache) constraints. MFA enhances model capacity by efficiently scaling up both the number and dimension of attention heads through low-rank matrix factorization in the Query-Key (QK) circuit. Extending MFA, MFA-KR further reduces memory requirements by repurposing the key cache as value through value projection re-parameterization. MFA's design enables strong model capacity when working under tight KV cache budget, while MFA-KR is suitable for even harsher KV cache limits with minor performance trade-off. Notably, in our extensive and large-scale experiments, the proposed architecture outperforms MLA and performs comparably to MHA, while reducing KV cache usage by up to 56% and 93.7%, respectively.

Zixin Yin, Xili Dai, Ling-Hao Chen et al.

Text-guided color editing in images and videos is a fundamental yet unsolved problem, requiring fine-grained manipulation of color attributes, including albedo, light source color, and ambient lighting, while preserving physical consistency in geometry, material properties, and light-matter interactions. Existing training-free methods offer broad applicability across editing tasks but struggle with precise color control and often introduce visual inconsistency in both edited and non-edited regions. In this work, we present ColorCtrl, a training-free color editing method that leverages the attention mechanisms of modern Multi-Modal Diffusion Transformers (MM-DiT). By disentangling structure and color through targeted manipulation of attention maps and value tokens, our method enables accurate and consistent color editing, along with word-level control of attribute intensity. Our method modifies only the intended regions specified by the prompt, leaving unrelated areas untouched. Extensive experiments on both SD3 and FLUX.1-dev demonstrate that ColorCtrl outperforms existing training-free approaches and achieves state-of-the-art performances in both edit quality and consistency. Furthermore, our method surpasses strong commercial models such as FLUX.1 Kontext Max and GPT-4o Image Generation in terms of consistency. When extended to video models like CogVideoX, our approach exhibits greater advantages, particularly in maintaining temporal coherence and editing stability. Finally, our method also generalizes to instruction-based editing diffusion models such as Step1X-Edit and FLUX.1 Kontext dev, further demonstrating its versatility.

Zixin Yin, Xili Dai, Duomin Wang et al.

The reliance on implicit point matching via attention has become a core bottleneck in drag-based editing, resulting in a fundamental compromise on weakened inversion strength and costly test-time optimization (TTO). This compromise severely limits the generative capabilities of diffusion models, suppressing high-fidelity inpainting and text-guided creation. In this paper, we introduce LazyDrag, the first drag-based image editing method for Multi-Modal Diffusion Transformers, which directly eliminates the reliance on implicit point matching. In concrete terms, our method generates an explicit correspondence map from user drag inputs as a reliable reference to boost the attention control. This reliable reference opens the potential for a stable full-strength inversion process, which is the first in the drag-based editing task. It obviates the necessity for TTO and unlocks the generative capability of models. Therefore, LazyDrag naturally unifies precise geometric control with text guidance, enabling complex edits that were previously out of reach: opening the mouth of a dog and inpainting its interior, generating new objects like a ``tennis ball'', or for ambiguous drags, making context-aware changes like moving a hand into a pocket. Additionally, LazyDrag supports multi-round workflows with simultaneous move and scale operations. Evaluated on the DragBench, our method outperforms baselines in drag accuracy and perceptual quality, as validated by VIEScore and human evaluation. LazyDrag not only establishes new state-of-the-art performance, but also paves a new way to editing paradigms.

Jingyun Xue, Hongfa Wang, Qi Tian et al.

Controllable character image animation has a wide range of applications. Although existing studies have consistently improved performance, challenges persist in the field of character image animation, particularly concerning stability in complex backgrounds and tasks involving multiple characters. To address these challenges, we propose a novel multi-condition guided framework for character image animation, employing several well-designed input modules to enhance the implicit decoupling capability of the model. First, the optical flow guider calculates the background optical flow map as guidance information, which enables the model to implicitly learn to decouple the background motion into background constants and background momentum during training, and generate a stable background by setting zero background momentum during inference. Second, the depth order guider calculates the order map of the characters, which transforms the depth information into the positional information of multiple characters. This facilitates the implicit learning of decoupling different characters, especially in accurately separating the occluded body parts of multiple characters. Third, the reference pose map is input to enhance the ability to decouple character texture and pose information in the reference image. Furthermore, to fill the gap of fair evaluation of multi-character image animation, we propose a new benchmark comprising about 4,000 frames. Extensive qualitative and quantitative evaluations demonstrate that our method excels in generating high-quality character animations, especially in scenarios of complex backgrounds and multiple characters.

Yue Ma, Hongyu Liu, Hongfa Wang et al.

We present Follow-Your-Emoji, a diffusion-based framework for portrait animation, which animates a reference portrait with target landmark sequences. The main challenge of portrait animation is to preserve the identity of the reference portrait and transfer the target expression to this portrait while maintaining temporal consistency and fidelity. To address these challenges, Follow-Your-Emoji equipped the powerful Stable Diffusion model with two well-designed technologies. Specifically, we first adopt a new explicit motion signal, namely expression-aware landmark, to guide the animation process. We discover this landmark can not only ensure the accurate motion alignment between the reference portrait and target motion during inference but also increase the ability to portray exaggerated expressions (i.e., large pupil movements) and avoid identity leakage. Then, we propose a facial fine-grained loss to improve the model's ability of subtle expression perception and reference portrait appearance reconstruction by using both expression and facial masks. Accordingly, our method demonstrates significant performance in controlling the expression of freestyle portraits, including real humans, cartoons, sculptures, and even animals. By leveraging a simple and effective progressive generation strategy, we extend our model to stable long-term animation, thus increasing its potential application value. To address the lack of a benchmark for this field, we introduce EmojiBench, a comprehensive benchmark comprising diverse portrait images, driving videos, and landmarks. We show extensive evaluations on EmojiBench to verify the superiority of Follow-Your-Emoji.

Xin-Yang Zheng, Hao Pan, Peng-Shuai Wang et al.

Although the recent rapid evolution of 3D generative neural networks greatly improves 3D shape generation, it is still not convenient for ordinary users to create 3D shapes and control the local geometry of generated shapes. To address these challenges, we propose a diffusion-based 3D generation framework -- locally attentional SDF diffusion, to model plausible 3D shapes, via 2D sketch image input. Our method is built on a two-stage diffusion model. The first stage, named occupancy-diffusion, aims to generate a low-resolution occupancy field to approximate the shape shell. The second stage, named SDF-diffusion, synthesizes a high-resolution signed distance field within the occupied voxels determined by the first stage to extract fine geometry. Our model is empowered by a novel view-aware local attention mechanism for image-conditioned shape generation, which takes advantage of 2D image patch features to guide 3D voxel feature learning, greatly improving local controllability and model generalizability. Through extensive experiments in sketch-conditioned and category-conditioned 3D shape generation tasks, we validate and demonstrate the ability of our method to provide plausible and diverse 3D shapes, as well as its superior controllability and generalizability over existing work. Our code and trained models are available at https://zhengxinyang.github.io/projects/LAS-Diffusion.html

Jianfeng Gao, Baolin Peng, Chunyuan Li et al.

This article presents a hybrid approach based on a Grounded Text Generation (GTG) model to building robust task bots at scale. GTG is a hybrid model which uses a large-scale Transformer neural network as its backbone, combined with symbol-manipulation modules for knowledge base inference and prior knowledge encoding, to generate responses grounded in dialog belief state and real-world knowledge for task completion. GTG is pre-trained on large amounts of raw text and human conversational data, and can be fine-tuned to complete a wide range of tasks. The hybrid approach and its variants are being developed simultaneously by multiple research teams. The primary results reported on task-oriented dialog benchmarks are very promising, demonstrating the big potential of this approach. This article provides an overview of this progress and discusses related methods and technologies that can be incorporated for building robust conversational AI systems.

Li Zhou, Jianfeng Gao, Di Li et al.

This paper describes the development of Microsoft XiaoIce, the most popular social chatbot in the world. XiaoIce is uniquely designed as an AI companion with an emotional connection to satisfy the human need for communication, affection, and social belonging. We take into account both intelligent quotient (IQ) and emotional quotient (EQ) in system design, cast human-machine social chat as decision-making over Markov Decision Processes (MDPs), and optimize XiaoIce for long-term user engagement, measured in expected Conversation-turns Per Session (CPS). We detail the system architecture and key components including dialogue manager, core chat, skills, and an empathetic computing module. We show how XiaoIce dynamically recognizes human feelings and states, understands user intent, and responds to user needs throughout long conversations. Since her launch in 2014, XiaoIce has communicated with over 660 million active users and succeeded in establishing long-term relationships with many of them. Analysis of large scale online logs shows that XiaoIce has achieved an average CPS of 23, which is significantly higher than that of other chatbots and even human conversations.

Heung-Yeung Shum, Xiaodong He, Di Li

Conversational systems have come a long way since their inception in the 1960s. After decades of research and development, we've seen progress from Eliza and Parry in the 60's and 70's, to task-completion systems as in the DARPA Communicator program in the 2000s, to intelligent personal assistants such as Siri in the 2010s, to today's social chatbots like XiaoIce. Social chatbots' appeal lies not only in their ability to respond to users' diverse requests, but also in being able to establish an emotional connection with users. The latter is done by satisfying users' need for communication, affection, as well as social belonging. To further the advancement and adoption of social chatbots, their design must focus on user engagement and take both intellectual quotient (IQ) and emotional quotient (EQ) into account. Users should want to engage with a social chatbot; as such, we define the success metric for social chatbots as conversation-turns per session (CPS). Using XiaoIce as an illustrative example, we discuss key technologies in building social chatbots from core chat to visual awareness to skills. We also show how XiaoIce can dynamically recognize emotion and engage the user throughout long conversations with appropriate interpersonal responses. As we become the first generation of humans ever living with AI, we have a responsibility to design social chatbots to be both useful and empathetic, so they will become ubiquitous and help society as a whole.