Multimodal generative models can process instructions in various modalities and demonstrate outstanding performance across a wide range of image generation tasks. However, their robustness in complex real-world scenarios remains limited due to insufficient generalized instruction alignment. We introduces \textbf{OmniGen2}, a unified multimodal generator designed to follow complex, fine-grained instructions. Our core contribution is a two-stage design that first builds a strong, world-knowledge-grounded foundation model and then aligns it using a progressive, multi-task instruction tuning strategy. The foundation model features a streamlined architecture with decoupled decoding for versatile multimodal generation and a novel positional encoding scheme to improve learning efficiency. We ground this model in real-world knowledge using large-scale data construction pipelines. Building on this foundation, we propose a progressive, reinforcement-based alignment process. This phase carefully schedules training tasks and reward signals to foster cross-task knowledge transfer, significantly improving the model's instruction-following capabilities. Our models demonstrate competitive performance on standard benchmarks and our dedicated in-context generation benchmark, \textbf{OmniContext}. We will release our models, code, benchmark, and training datasets to catalyze future research in building more capable and instruction aligned generative models.

Video understanding with multimodal large language models (MLLMs) remains challenging due to the long token sequences of videos, which contain extensive temporal dependencies and redundant frames.Existing approaches typically treat MLLMs as passive recognizers, processing entire videos or uniformly sampled frames without adaptive reasoning.Recent agent-based methods introduce external tools, yet still depend on manually designed workflows and perception-first strategies, resulting in inefficiency on long videos. We present EVA, an Efficient Reinforcement Learning framework for End-to-End Video Agent, which enables planning-before-perception through iterative summary–plan–action–reflection reasoning.EVA autonomously decides what to watch, when to watch, and how to watch, achieving query-driven and efficient video understanding.To train such agents, we design a simple yet effective three-stage learning pipeline—comprising supervised fine-tuning (SFT), Kahneman–Tversky Optimization (KTO), and Generalized Reward Policy Optimization (GRPO)—that bridges supervised imitation and reinforcement learning.We further construct high-quality datasets for each stage, supporting stable and reproducible training.We evaluate EVA on six video understanding benchmarks, demonstrating its comprehensive capabilities. Compared with existing baselines, EVA achieves a substantial improvement of 6--12\% over general MLLM baselines and a further 1--3\% gain over prior adaptive agent methods.

Multiple-choice question answering (MCQA) has been a popular format for evaluating and reinforcement fine-tuning (RFT) of modern multimodal language models. Its constrained output format allows for simplified, deterministic automatic verification.However, we find that the options may leak exploitable signals, which makes the accuracy metrics unreliable for indicating real capabilities and encourages explicit or implicit answer guessing behaviors during RFT. We propose ReVeL (\textbf{Re}write and \textbf{Ve}rify by \textbf{L}LM), a framework that rewrites multiple-choice questions into open-form questions while keeping answers verifiable whenever possible. The framework categorizes questions according to different answer types, apply different rewriting and verification schemes, respectively. When applied for RFT, we converted 20k MCQA examples and use GRPO to finetune Qwen2.5-VL models. Models trained on ReVeL-OpenQA match MCQA accuracy on multiple-choice benchmarks and improve OpenQA accuracy by about six percentage points, indicating better data efficiency and more robust reward signals than MCQA-based training. When used for evaluation, ReVeL also reveals up to 20 percentage points of score inflation in MCQA benchmarks (relative to OpenQA), improves judging accuracy, and reduces both cost and latency. We will release code and data publicly.

Text-to-Audio-Video (T2AV) generation aims to synthesize temporally coherent video and semantically synchronized audio from natural language, yet its evaluation remains fragmented, often relying on unimodal metrics or narrowly scoped benchmarks that fail to capture cross-modal alignment, instruction following, and perceptual realism under complex prompts. To address this limitation, we present T2AV-Compass, a unified benchmark for comprehensive evaluation of T2AV systems, consisting of 500 diverse and complex prompts constructed via a taxonomy-driven pipeline to ensure semantic richness and physical plausibility. Besides, T2AV-Compass introduces a dual-level evaluation framework that integrates objective signal-level metrics for video quality, audio quality, and cross-modal alignment with a subjective MLLM-as-a-Judge protocol for instruction following and realism assessment. Extensive evaluation of 15 representative T2AV systems reveals that even the strongest models fall substantially short of human-level realism and cross-modal consistency, with persistent failures in audio realism, fine-grained synchronization, instruction following, etc. These results indicate significant improvement room for future models and highlight the value of T2AV-Compass as a challenging and diagnostic testbed for advancing text-to-audio-video generation.

While Direct Preference Optimization (DPO) has become the de facto approach for aligning Large Vision-Language Models (LVLMs), it suffers from Likelihood Displacement, where the probability of both chosen and rejected responses collapses. This optimization flaw is especially detrimental in multimodal settings: the erosion of chosen likelihoods – a failure we term Visual Anchor Collapse – causes models to abandon visual evidence for strong language priors, precipitating significant hallucinations. To address this, we propose Asymmetric Constrained Preference Optimization (ACPO), a modality-agnostic alignment mechanism that applies dynamic, target-oriented scaling to preference optimization. ACPO derives a complexity-aware scaling coefficient applied exclusively to the rejected reward, asymmetrically suppressing the gradient flow on the rejected term while preserving the chosen distribution as a gradient-stable reference. While fundamentally a general-purpose objective, breaking this gradient symmetry is crucial for multimodal tasks, as it mitigates the suppression of visual tokens by language priors. Experiments on InternVL models demonstrate that ACPO effectively reverses the chosen-reward degradation of standard DPO. By halting Visual Anchor Collapse, ACPO generally outperforms baselines on hallucination benchmarks (HallusionBench, MM-IFEval) and general leaderboards (MMBench, MMStar, OCRBenchV2) while driving concurrent improvements in general capabilities.

We revisit video hallucination in multimodal large language models (Video-MLLMs) from a semantic aggregation perspective. While prior work attributes hallucinations to language priors, missing frames, or visual encoder biases, these explanations overlook errors arising during the aggregation of correct frame-level semantics into event-level interpretations. We term this phenomenon Semantic Aggregation Hallucination (SAH), which becomes increasingly prevalent in complex, multi-event video understanding tasks with rich temporal dependencies. To systematically study SAH, we introduce ELV-Halluc, the first benchmark designed for fine-grained evaluation of semantic aggregation errors. Our experiments reveal that SAH correlates with both semantic complexity and rapid semantic transitions. We further propose mitigation strategies: improved positional encoding preserves temporal structure, and reinforcement learning such as DPO enhances the model’s ability to distinguish semantics within and across events. Using a curated 8K adversarial video-text pair dataset, our approach achieves consistent gains across benchmarks, including a 27.7\% reduction in SAH rate on ELV-Halluc and Video-MME.

Temporal Action Detection (TAD) aims to locate the temporal boundaries and recognize categories of actions in videos. Existing single-stage methods, while emphasizing temporal interactions, often neglect the treatment of local features. Given the increasingly complex and overlapping actions in new datasets, relying solely on temporal interactions is insufficient. In this paper, we revisit the channel attention in TAD and propose a novel single-stage convolutional model called Highlighted Feature Interaction (HiFi). HiFi introduces a fundamental unit called the Expanded Excitation & Interaction Block (EEI), which evaluates the importance of each channel and scales the channels to suppress irrelevant and emphasize valuable channels. Consequently, HiFi can learn effective temporal representations at multi-scales through convolution and pooling operations. Extensive experiments on THUMOS-14, ActivityNet v1.3, HACS Segment, FineAction, Epic-Kitchen, and Ego4D-Moment Queries demonstrate HiFi outperforms the state-of-the-art methods and confirm the efficacy of weak temporal interactions and channel enhancement in TAD.

Character image animation, which synthesizes videos of reference characters driven by pose sequences, has advanced rapidly but remains largely limited to single-human settings. Existing methods struggle to generalize to multi-humanoid scenarios, which involve diverse humanoid forms, complex interactions, and frequent occlusions. We address this gap with two key innovations. First, we introduce unified motion representations that extract identity-agnostic 4D motions and explicitly bind them to corresponding characters, enabling generalization across diverse humanoid forms and seamless extension to multi-humanoid scenarios. Second, we propose a holistic 4D-anchored paradigm that constructs a shared 4D space to fuse motion representations with video latents, and further reinforces this process with hierarchical 4D-level supervision to better handle interactions and occlusions. We instantiate these ideas in MotionWeaver, an end-to-end framework for multi-humanoid image animation. To support this setting, we curate a 46-hour dataset of multi-human videos with rich interactions, and construct a 300-video benchmark featuring paired humanoid characters. Quantitative and qualitative experiments demonstrate that MotionWeaver not only achieves state-of-the-art results on our benchmark but also generalizes effectively across diverse humanoid forms, complex interactions, and challenging multi-humanoid scenarios.

While collaborative forecasting on distributed time series is highly desirable, directly pooling localized datasets is often impractical due to data sharing constraints. Federated learning offers a promising alternative, yet conventional federated learning algorithms require homogeneous model architectures, which are incompatible with the structural discrepancies, such as unaligned temporal resolutions and mismatched variable channels, commonly observed across decentralized nodes. To bridge this gap, we introduce PiXTime, a novel Transformer-based framework designed to natively accommodate and leverage structurally heterogeneous temporal data. At its core, PiXTime adopts a parameter-decoupling architecture, strategically partitioning the model into localized personalized modules and a globally aggregated shared backbone. Specifically, node-specific local modules act as dimensional adapters, projecting raw sequences of diverse lengths into a unified representation space. Concurrently, a globally synchronized VE Table injects consistent categorical identities into the feature space, allowing the shared backbone to collaboratively learn and generalize representations across inconsistent variable distributions. Comprehensive evaluations on multiple benchmarks demonstrate that PiXTime achieves state-of-the-art performance in heterogeneous federated environments, while maintaining robust superiority in standard homogeneous and centralized forecasting settings.

Current vision-language models (VLMs) typically stitch together separate image encoders and language decoders via multi-stage alignment, a modular framework that inevitably fragments pixel-level signals across frames and scatters early pixel-word interactions. In parallel, native VLMs, despite impressive performance on single images, remain largely unexplored in multi-image, video understanding, and spatial intelligence. Hence, we introduce NEO-ov, a native foundation model that learns cross-frame and pixel-word correspondence end-to-end, without any external encoders, auxiliary adapters, or post-hoc fusion. By eliminating module boundaries entirely, NEO-ov enables fine-grained and unified spatiotemporal modeling to emerge natively inside the model. Notably, NEO-ov largely narrows the gap to modular counterparts while excelling at fine-grained visual perception, validating that native "one-vision" architectures are not only feasible but competitive at scale. Beyond empirical performance, we unveil systematic architectural analyses and detailed training recipes to facilitate subsequent native multimodal modeling. Our code and models are publicly available at: https://github.com/EvolvingLMMs-Lab/NEO.

Autoregressive models with continuous tokens form a promising paradigm for visual generation, especially for text-to-image (T2I) synthesis, but they suffer from high computational cost. We study how to design compute-efficient linear attention within this framework. Specifically, we conduct a systematic empirical analysis of scaling behavior with respect to parameter counts under different design choices, focusing on (1) normalization paradigms in linear attention (division-based vs. subtraction-based) and (2) depthwise convolution for locality augmentation. Our results show that although subtraction-based normalization is effective for image classification, division-based normalization scales better for linear generative transformers. In addition, incorporating convolution for locality modeling plays a crucial role in autoregressive generation, consistent with findings in diffusion models. We further extend gating mechanisms, commonly used in causal linear attention, to the bidirectional setting and propose a KV gate. By introducing data-independent learnable parameters to the key and value states, the KV gate assigns token-wise memory weights, enabling flexible memory management similar to forget gates in language models. Based on these findings, we present LINA, a simple and compute-efficient T2I model built entirely on linear attention, capable of generating high-fidelity 1024x1024 images from user instructions. LINA achieves competitive performance on both class-conditional and T2I benchmarks, obtaining 2.18 FID on ImageNet (about 1.4B parameters) and 0.74 on GenEval (about 1.5B parameters). A single linear attention module reduces FLOPs by about 61 percent compared to softmax attention. Code and models are available at: https://github.com/techmonsterwang/LINA.

Recent large vision-language models (VLMs) remain fundamentally constrained by a persistent dichotomy: understanding and generation are treated as distinct problems, leading to fragmented architectures, cascaded pipelines, and misaligned representation spaces. We argue that this divide is not merely an engineering artifact, but a structural limitation that hinders the emergence of native multimodal intelligence. Hence, we introduce SenseNova-U1, a native unified multimodal paradigm built upon NEO-unify, in which understanding and generation evolve as synergistic views of a single underlying process. We launch two native unified variants, SenseNova-U1-8B-MoT and SenseNova-U1-A3B-MoT, built on dense (8B) and mixture-of-experts (30B-A3B) understanding baselines, respectively. Designed from first principles, they rival top-tier understanding-only VLMs across text understanding, vision-language perception, knowledge reasoning, agentic decision-making, and spatial intelligence. Meanwhile, they deliver strong semantic consistency and visual fidelity, excelling in conventional or knowledge-intensive any-to-image (X2I) synthesis, complex text-rich infographic generation, and interleaved vision-language generation, with or without think patterns. Beyond performance, we show detailed model design, data preprocessing, pre-/post-training, and inference strategies to support community research. Last but not least, preliminary evidence demonstrates that our models extend beyond perception and generation, performing strongly in vision-language-action (VLA) and world model (WM) scenarios. This points toward a broader roadmap where models do not translate between modalities, but think and act across them in a native manner. Multimodal AI is no longer about connecting separate systems, but about building a unified one and trusting the necessary capabilities to emerge from within.

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework that dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently. Extensive evaluations show that Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks. Agent Swarm also reduces latency by up to 4.5× over single-agent baselines. We release the post-trained Kimi K2.5 model checkpoint to facilitate future research and real-world applications of agentic intelligence.

Instruction-guided image editing has achieved remarkable progress, yet current models still face challenges with complex instructions and often require multiple samples to produce a desired result. Reinforcement Learning (RL) offers a promising solution, but its adoption in image editing has been severely hindered by the lack of a high-fidelity, efficient reward signal. In this work, we present a comprehensive methodology to overcome this barrier, centered on the development of a state-of-the-art, specialized reward model. We first introduce $\textbf{EditReward-Bench}$, a comprehensive benchmark to systematically evaluate reward models on editing quality. Guided by this benchmark, we develop $\textbf{EditScore}$, an efficient model to evaluate the quality of instruction-guided editing. Through meticulous data curation and filtering, EditScore effectively matches the performance of learning proprietary VLMs. Furthermore, coupled with an effective self-ensemble strategy tailored for the generative nature of EditScore, our largest variant even surpasses GPT-5 in the benchmark. We then demonstrate that a high-fidelity reward model is the key to unlocking online RL for image editing. Our experiments show that, while even the largest open-source VLMs fail to provide an effective learning signal, EditScore enables efficient and robust policy optimization. Applying our framework to a strong base model, OmniGen2, results in a final model that shows a substantial and consistent performance uplift. Overall, this work provides the first systematic path from benchmarking to reward modeling to RL training in image editing, showing that a high-fidelity, domain-specialized reward model is the key to unlocking the full potential of RL in this domain. Our code, models, and benchmark will be released publicly.

Model merging, typically on Instruct and Thinking models, has shown remarkable performance for efficient reasoning. In this paper, we systematically revisit the simplest merging method that interpolates two weights directly. Particularly, we observe that model interpolation follows a three-stage evolutionary paradigm with distinct behaviors on the reasoning trajectory. These dynamics provide a principled guide for navigating the performance-cost trade-off. Empirical results demonstrate that a strategically interpolated model surprisingly surpasses sophisticated model merging baselines on both efficiency and effectiveness. We further validate our findings with extensive ablation studies on model layers, modules, and decoding strategies. Ultimately, this work demystifies model interpolation and offers a practical framework for crafting models with precisely targeted reasoning capabilities.

Visual reasoning is a core component of human intelligence and a critical capability for advanced multimodal models. Yet current reasoning evaluations of multimodal large language models (MLLMs) often rely on text descriptions and allow language-based reasoning shortcuts, failing to measure genuine vision-centric reasoning. To address this, we introduce VisuLogic: a benchmark of 1,000 human-verified problems across six categories (e.g., quantitative shifts, spatial relations, attribute comparisons). These various types of questions can be evaluated to assess the visual reasoning capabilities of MLLMs from multiple perspectives. We evaluate leading MLLMs on this benchmark and analyze their results to identify common failure modes. Most models score below 30\% accuracy—only slightly above the 25\% random baseline and far below the 51.4\% achieved by humans—revealing significant gaps in visual reasoning.

Video multimodal large language models (Video-MLLMs) have achieved remarkable progress in video understanding. However, they remain vulnerable to hallucination-producing content inconsistent with or unrelated to video inputs. Previous video hallucination benchmarks primarily focus on short-videos. They attribute hallucinations to factors such as strong language priors, missing frames, or vision-language biases introduced by the visual encoder. While these causes indeed account for most hallucinations in short videos, they still oversimplify the cause of hallucinations. Sometimes, models generate incorrect outputs but with correct frame-level semantics. We refer to this type of hallucination as Semantic Aggregation Hallucination (SAH), which arises during the process of aggregating frame-level semantics into event-level semantic groups. Given that SAH becomes particularly critical in long videos due to increased semantic complexity across multiple events, it is essential to separate and thoroughly investigate the causes of this type of hallucination. To address the above issues, we introduce ELV-Halluc, the first benchmark dedicated to long-video hallucination, enabling a systematic investigation of SAH. Our experiments confirm the existence of SAH and show that it increases with semantic complexity. Additionally, we find that models are more prone to SAH on rapidly changing semantics. Moreover, we discuss potential approaches to mitigate SAH. We demonstrate that positional encoding strategy contributes to alleviating SAH, and further adopt DPO strategy to enhance the model's ability to distinguish semantics within and across events. To support this, we curate a dataset of 8K adversarial data pairs and achieve improvements on both ELV-Halluc and Video-MME, including a substantial 27.7

Recently, several spatial-temporal memory-based methods have verified that storing intermediate frames with masks as memory helps segment target objects in videos. However, they mainly focus on better matching between the current frame and memory frames without paying attention to the quality of the memory. Consequently, frames with poor segmentation masks may be memorized, leading to error accumulation problems. Besides, the linear increase of memory frames with the growth of frame numbers limits the ability of the models to handle long videos. To this end, we propose a Quality-aware Dynamic Memory Network (QDMN) to evaluate the segmentation quality of each frame, allowing the memory bank to selectively store accurately segmented frames and prevent error accumulation. Then, we combine the segmentation quality with temporal consistency to dynamically update the memory bank and make the models can handle videos of arbitrary length. The above operation ensures the reliability of memory frames and improves the quality of memory at the frame level. Moreover, we observe that the memory features extracted from reliable frames still contain noise and have limited representation capabilities. To address this problem, we propose to perform memory enhancement and anchoring on the basis of QDMN to improve the quality of memory from the feature level, resulting in a more robust and effective network QDMN++. Our method achieves state-of-the-art performance on all popular benchmarks. Moreover, extensive experiments demonstrate that the proposed memory screening mechanism can be applied to any memory-based methods as generic plugins.

In this paper, we investigate how to convert a pre-trained Diffusion Transformer (DiT) into a linear DiT, as its simplicity, parallelism, and efficiency for image generation. Through detailed exploration, we offer a suite of ready-to-use solutions, ranging from linear attention design to optimization strategies. Our core contributions include 5 practical guidelines: 1) Applying depth-wise convolution within simple linear attention is sufficient for image generation. 2) Using fewer heads in linear attention provides a free-lunch performance boost without increasing latency. 3) Inheriting weights from a fully converged, pre-trained DiT. 4) Loading all parameters except those related to linear attention. 5) Hybrid knowledge distillation: using a pre-trained teacher DiT to help the training of the student linear DiT, supervising not only the predicted noise but also the variance of the reverse diffusion process. These guidelines lead to our proposed Linear Diffusion Transformer (LiT), which serves as a safe and efficient alternative baseline for DiT with pure linear attention. In class-conditional 256×256 and 512×512 ImageNet generation, LiT can be quickly adapted from DiT using only 20% and 33% of DiT's training steps, respectively, while achieving comparable performance. LiT also rivals methods based on Mamba or Gated Linear Attention. Moreover, the same guidelines generalize to text-to-image generation: LiT can be swiftly converted from PixArt-Σ to generate high-quality images, maintaining comparable GenEval scores.

Understanding videos is one of the fundamental directions in computer vision research, with extensive efforts dedicated to exploring various architectures such as RNN, 3D CNN, and Transformers. The newly proposed architecture of state space model, e.g., Mamba, shows promising traits to extend its success in long sequence modeling to video modeling. To assess whether Mamba can be a viable alternative to Transformers in the video understanding domain, in this work, we conduct a comprehensive set of studies, probing different roles Mamba can play in modeling videos, while investigating diverse tasks where Mamba could exhibit superiority. We categorize Mamba into four roles for modeling videos, deriving a Video Mamba Suite composed of 14 models/modules, and evaluating them on 12 video understanding tasks. Our extensive experiments reveal the strong potential of Mamba on both video-only and video-language tasks while showing promising efficiency-performance trade-offs. We hope this work could provide valuable data points and insights for future research on video understanding. Code is public: https://github.com/OpenGVLab/video-mamba-suite.