Lijun Yu

Lijun Yu, Yong Cheng, Kihyuk Sohn et al. · cmu, deepmind

We introduce the MAsked Generative VIdeo Transformer, MAGVIT, to tackle various video synthesis tasks with a single model. We introduce a 3D tokenizer to quantize a video into spatial-temporal visual tokens and propose an embedding method for masked video token modeling to facilitate multi-task learning. We conduct extensive experiments to demonstrate the quality, efficiency, and flexibility of MAGVIT. Our experiments show that (i) MAGVIT performs favorably against state-of-the-art approaches and establishes the best-published FVD on three video generation benchmarks, including the challenging Kinetics-600. (ii) MAGVIT outperforms existing methods in inference time by two orders of magnitude against diffusion models and by 60x against autoregressive models. (iii) A single MAGVIT model supports ten diverse generation tasks and generalizes across videos from different visual domains. The source code and trained models will be released to the public at https://magvit.cs.cmu.edu.

Lijun Yu, José Lezama, Nitesh B. Gundavarapu et al. · cmu, deepmind

While Large Language Models (LLMs) are the dominant models for generative tasks in language, they do not perform as well as diffusion models on image and video generation. To effectively use LLMs for visual generation, one crucial component is the visual tokenizer that maps pixel-space inputs to discrete tokens appropriate for LLM learning. In this paper, we introduce MAGVIT-v2, a video tokenizer designed to generate concise and expressive tokens for both videos and images using a common token vocabulary. Equipped with this new tokenizer, we show that LLMs outperform diffusion models on standard image and video generation benchmarks including ImageNet and Kinetics. In addition, we demonstrate that our tokenizer surpasses the previously top-performing video tokenizer on two more tasks: (1) video compression comparable to the next-generation video codec (VCC) according to human evaluations, and (2) learning effective representations for action recognition tasks.

Lijun Yu, Yong Cheng, Zhiruo Wang et al. · cmu, deepmind

In this work, we introduce Semantic Pyramid AutoEncoder (SPAE) for enabling frozen LLMs to perform both understanding and generation tasks involving non-linguistic modalities such as images or videos. SPAE converts between raw pixels and interpretable lexical tokens (or words) extracted from the LLM's vocabulary. The resulting tokens capture both the semantic meaning and the fine-grained details needed for visual reconstruction, effectively translating the visual content into a language comprehensible to the LLM, and empowering it to perform a wide array of multimodal tasks. Our approach is validated through in-context learning experiments with frozen PaLM 2 and GPT 3.5 on a diverse set of image understanding and generation tasks. Our method marks the first successful attempt to enable a frozen LLM to generate image content while surpassing state-of-the-art performance in image understanding tasks, under the same setting, by over 25%.

Xiaopeng Yuan, Haibo Jin, Ye Yu et al.

Recent work moves intermediate reasoning from natural-language traces into latent or cache-level representations to reduce token overhead and avoid a discrete communication bottleneck. However, this shift also removes a key advantage of textual reasoning: intermediate states are no longer inspectable, making it difficult to determine whether a latent state still preserves the constraints of the original query. As a result, latent reasoning typically operates in an open loop, where a latent state is produced and consumed without an input-anchored fidelity check. We propose ReLAT (Reconstruction-Guided Latent Reasoning At Test Time), a self-supervised test-time training method that closes this loop using the query itself as the reference. Our key observation is that if a latent state faithfully represents a query, the query should be recoverable from it; if the query cannot be recovered, the latent state has lost task-relevant information. ReLAT operationalizes this principle by constructing a differentiable Question -> Latent Thought -> Question cycle and optimizing query reconstruction loss through the latent thought before answer generation. This anchors opaque latent computation to the problem specification it is supposed to represent. Across mathematical reasoning, knowledge QA, and code generation benchmarks on the Qwen family, ReLAT consistently improves over single-model inference, text-based collaboration, open-loop latent collaboration, and alternative test-time training objectives. On Qwen3-8B, ReLAT raises AIME 2024 accuracy from 56.7% to 73.3%, a 16.6-point gain over the strongest open-loop latent baseline.

Lijun Yu, Jin Miao, Xiaoyu Sun et al. · cmu

Document understanding tasks, in particular, Visually-rich Document Entity Retrieval (VDER), have gained significant attention in recent years thanks to their broad applications in enterprise AI. However, publicly available data have been scarce for these tasks due to strict privacy constraints and high annotation costs. To make things worse, the non-overlapping entity spaces from different datasets hinder the knowledge transfer between document types. In this paper, we propose a method to collect massive-scale and weakly labeled data from the web to benefit the training of VDER models. The collected dataset, named DocumentNet, does not depend on specific document types or entity sets, making it universally applicable to all VDER tasks. The current DocumentNet consists of 30M documents spanning nearly 400 document types organized in a four-level ontology. Experiments on a set of broadly adopted VDER tasks show significant improvements when DocumentNet is incorporated into the pre-training for both classic and few-shot learning settings. With the recent emergence of large language models (LLMs), DocumentNet provides a large data source to extend their multi-modal capabilities for VDER.

Mark Weber, Lijun Yu, Qihang Yu et al.

Masked transformer models for class-conditional image generation have become a compelling alternative to diffusion models. Typically comprising two stages - an initial VQGAN model for transitioning between latent space and image space, and a subsequent Transformer model for image generation within latent space - these frameworks offer promising avenues for image synthesis. In this study, we present two primary contributions: Firstly, an empirical and systematic examination of VQGANs, leading to a modernized VQGAN. Secondly, a novel embedding-free generation network operating directly on bit tokens - a binary quantized representation of tokens with rich semantics. The first contribution furnishes a transparent, reproducible, and high-performing VQGAN model, enhancing accessibility and matching the performance of current state-of-the-art methods while revealing previously undisclosed details. The second contribution demonstrates that embedding-free image generation using bit tokens achieves a new state-of-the-art FID of 1.52 on the ImageNet 256x256 benchmark, with a compact generator model of mere 305M parameters. The code for this project is available on https://github.com/markweberdev/maskbit.

Haoran Sun, Lijun Yu, Bo Dai et al.

Score-based modeling through stochastic differential equations (SDEs) has provided a new perspective on diffusion models, and demonstrated superior performance on continuous data. However, the gradient of the log-likelihood function, i.e., the score function, is not properly defined for discrete spaces. This makes it non-trivial to adapt \textcolor{\cdiff}{the score-based modeling} to categorical data. In this paper, we extend diffusion models to discrete variables by introducing a stochastic jump process where the reverse process denoises via a continuous-time Markov chain. This formulation admits an analytical simulation during backward sampling. To learn the reverse process, we extend score matching to general categorical data and show that an unbiased estimator can be obtained via simple matching of the conditional marginal distributions. We demonstrate the effectiveness of the proposed method on a set of synthetic and real-world music and image benchmarks.

Yuqing Wang, Chuofan Ma, Zhijie Lin et al.

Visual generation with discrete tokens has gained significant attention as it enables a unified token prediction paradigm shared with language models, promising seamless multimodal architectures. However, current discrete generation methods remain limited to low-dimensional latent tokens (typically 8-32 dims), sacrificing the semantic richness essential for understanding. While high-dimensional pretrained representations (768-1024 dims) could bridge this gap, their discrete generation poses fundamental challenges. In this paper, we present Cubic Discrete Diffusion (CubiD), the first discrete generation model for high-dimensional representations. CubiD performs fine-grained masking throughout the high-dimensional discrete representation -- any dimension at any position can be masked and predicted from partial observations. This enables the model to learn rich correlations both within and across spatial positions, with the number of generation steps fixed at $T$ regardless of feature dimensionality, where $T \ll hwd$. On ImageNet-256, CubiD achieves state-of-the-art discrete generation with strong scaling behavior from 900M to 3.7B parameters. Crucially, we validate that these discretized tokens preserve original representation capabilities, demonstrating that the same discrete tokens can effectively serve both understanding and generation tasks. We hope this work will inspire future research toward unified multimodal architectures. Code is available at: https://github.com/YuqingWang1029/CubiD.

Chong Bao, Shichen Liu, Lijun Yu et al.

Digital humans are fundamental to immersive interaction, yet creating a unified model for holistic modalities, including text, audio, motion, and visual content, remains an open challenge. In this paper, we present Archon, a fully pretrained, human-centric unified multimodal model for holistic avatar generation. Archon unifies seven modalities with modality-specific tokenizers, and a native autoregressive unified multimodal model pretrained on synchronized modalities and 72 diverse tasks to model holistic joint distributions. To address the token explosion challenge in high-fidelity talking videos, we introduce a memory-efficient semantic video reparameterization, achieving 4x token reduction while preserving fine-grained dynamics, coupled with a semantic-driven video diffusion decoder. We further propose a "Thinking in Modality" that decomposes ambiguous cross-modal tasks into stepwise thinking in an alternative chain of modality, progressively enhancing fidelity and controllability. Extensive experiments demonstrate that Archon achieves superior or comparable performance across diverse digital human generation tasks, validating the effectiveness of our unified framework. Project page: https://zju3dv.github.io/archon/.

Xinchen Yan, Chen Liang, Lijun Yu et al.

This paper investigates the scaling properties of autoregressive next-pixel prediction, a simple, end-to-end yet under-explored framework for unified vision models. Starting with images at resolutions of 32x32, we train a family of Transformers using IsoFlops profiles across compute budgets up to 7e19 FLOPs and evaluate three distinct target metrics: next-pixel prediction objective, ImageNet classification accuracy, and generation quality measured by Fr'echet Distance. First, optimal scaling strategy is critically task-dependent. At a fixed 32x32 resolution alone, the optimal scaling properties for image classification and image generation diverge, where generation optimal setup requires the data size grow three to five times faster than for the classification optimal setup. Second, as image resolution increases, the optimal scaling strategy indicates that the model size must grow much faster than data size. Surprisingly, by projecting our findings, we discover that the primary bottleneck is compute rather than the amount of training data. As compute continues to grow four to five times annually, we forecast the feasibility of pixel-by-pixel modeling of images within the next five years.

Kai Hu, Weichen Yu, Yining Li et al. · cmu

Recent research indicates that large language models (LLMs) are susceptible to jailbreaking attacks that can generate harmful content. This paper introduces a novel token-level attack method, Adaptive Dense-to-Sparse Constrained Optimization (ADC), which has been shown to successfully jailbreak multiple open-source LLMs. Drawing inspiration from the difficulties of discrete token optimization, our method relaxes the discrete jailbreak optimization into a continuous optimization process while gradually increasing the sparsity of the optimizing vectors. This technique effectively bridges the gap between discrete and continuous space optimization. Experimental results demonstrate that our method is more effective and efficient than state-of-the-art token-level methods. On Harmbench, our approach achieves the highest attack success rate on seven out of eight LLMs compared to the latest jailbreak methods. Trigger Warning: This paper contains model behavior that can be offensive in nature.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Lingxiao Zhao, Xueying Ding, Lijun Yu et al.

Discrete diffusion models have seen a surge of attention with applications on naturally discrete data such as language and graphs. Although discrete-time discrete diffusion has been established for a while, only recently Campbell et al. (2022) introduced the first framework for continuous-time discrete diffusion. However, their training and sampling processes differ significantly from the discrete-time version, necessitating nontrivial approximations for tractability. In this paper, we first present a series of mathematical simplifications of the variational lower bound that enable more accurate and easy-to-optimize training for discrete diffusion. In addition, we derive a simple formulation for backward denoising that enables exact and accelerated sampling, and importantly, an elegant unification of discrete-time and continuous-time discrete diffusion. Thanks to simpler analytical formulations, both forward and now also backward probabilities can flexibly accommodate any noise distribution, including different noise distributions for multi-element objects. Experiments show that our proposed USD3 (for Unified Simplified Discrete Denoising Diffusion) outperform all SOTA baselines on established datasets. We open-source our unified code at https://github.com/LingxiaoShawn/USD3.

Dan Kondratyuk, Lijun Yu, Xiuye Gu et al. · cmu, deepmind

We present VideoPoet, a language model capable of synthesizing high-quality video, with matching audio, from a large variety of conditioning signals. VideoPoet employs a decoder-only transformer architecture that processes multimodal inputs -- including images, videos, text, and audio. The training protocol follows that of Large Language Models (LLMs), consisting of two stages: pretraining and task-specific adaptation. During pretraining, VideoPoet incorporates a mixture of multimodal generative objectives within an autoregressive Transformer framework. The pretrained LLM serves as a foundation that can be adapted for a range of video generation tasks. We present empirical results demonstrating the model's state-of-the-art capabilities in zero-shot video generation, specifically highlighting VideoPoet's ability to generate high-fidelity motions. Project page: http://sites.research.google/videopoet/

Agrim Gupta, Lijun Yu, Kihyuk Sohn et al. · cmu, deepmind

We present W.A.L.T, a transformer-based approach for photorealistic video generation via diffusion modeling. Our approach has two key design decisions. First, we use a causal encoder to jointly compress images and videos within a unified latent space, enabling training and generation across modalities. Second, for memory and training efficiency, we use a window attention architecture tailored for joint spatial and spatiotemporal generative modeling. Taken together these design decisions enable us to achieve state-of-the-art performance on established video (UCF-101 and Kinetics-600) and image (ImageNet) generation benchmarks without using classifier free guidance. Finally, we also train a cascade of three models for the task of text-to-video generation consisting of a base latent video diffusion model, and two video super-resolution diffusion models to generate videos of $512 \times 896$ resolution at $8$ frames per second.

Ruibin Yuan, Hanfeng Lin, Shuyue Guo et al.

We tackle the task of long-form music generation--particularly the challenging \textbf{lyrics-to-song} problem--by introducing YuE, a family of open foundation models based on the LLaMA2 architecture. Specifically, YuE scales to trillions of tokens and generates up to five minutes of music while maintaining lyrical alignment, coherent musical structure, and engaging vocal melodies with appropriate accompaniment. It achieves this through (1) track-decoupled next-token prediction to overcome dense mixture signals, (2) structural progressive conditioning for long-context lyrical alignment, and (3) a multitask, multiphase pre-training recipe to converge and generalize. In addition, we redesign the in-context learning technique for music generation, enabling versatile style transfer (e.g., converting Japanese city pop into an English rap while preserving the original accompaniment) and bidirectional generation. Through extensive evaluation, we demonstrate that YuE matches or even surpasses some of the proprietary systems in musicality and vocal agility. In addition, fine-tuning YuE enables additional controls and enhanced support for tail languages. Furthermore, beyond generation, we show that YuE's learned representations can perform well on music understanding tasks, where the results of YuE match or exceed state-of-the-art methods on the MARBLE benchmark. Keywords: lyrics2song, song generation, long-form, foundation model, music generation

Kaiwen Zha, Lijun Yu, Alireza Fathi et al.

Image tokenization, the process of transforming raw image pixels into a compact low-dimensional latent representation, has proven crucial for scalable and efficient image generation. However, mainstream image tokenization methods generally have limited compression rates, making high-resolution image generation computationally expensive. To address this challenge, we propose to leverage language for efficient image tokenization, and we call our method Text-Conditioned Image Tokenization (TexTok). TexTok is a simple yet effective tokenization framework that leverages language to provide a compact, high-level semantic representation. By conditioning the tokenization process on descriptive text captions, TexTok simplifies semantic learning, allowing more learning capacity and token space to be allocated to capture fine-grained visual details, leading to enhanced reconstruction quality and higher compression rates. Compared to the conventional tokenizer without text conditioning, TexTok achieves average reconstruction FID improvements of 29.2% and 48.1% on ImageNet-256 and -512 benchmarks respectively, across varying numbers of tokens. These tokenization improvements consistently translate to 16.3% and 34.3% average improvements in generation FID. By simply replacing the tokenizer in Diffusion Transformer (DiT) with TexTok, our system can achieve a 93.5x inference speedup while still outperforming the original DiT using only 32 tokens on ImageNet-512. TexTok with a vanilla DiT generator achieves state-of-the-art FID scores of 1.46 and 1.62 on ImageNet-256 and -512 respectively. Furthermore, we demonstrate TexTok's superiority on the text-to-image generation task, effectively utilizing the off-the-shelf text captions in tokenization. Project page is at: https://kaiwenzha.github.io/textok/.

Gwanghyun Kim, Alonso Martinez, Yu-Chuan Su et al. · cmu

Training diffusion models for audiovisual sequences allows for a range of generation tasks by learning conditional distributions of various input-output combinations of the two modalities. Nevertheless, this strategy often requires training a separate model for each task which is expensive. Here, we propose a novel training approach to effectively learn arbitrary conditional distributions in the audiovisual space.Our key contribution lies in how we parameterize the diffusion timestep in the forward diffusion process. Instead of the standard fixed diffusion timestep, we propose applying variable diffusion timesteps across the temporal dimension and across modalities of the inputs. This formulation offers flexibility to introduce variable noise levels for various portions of the input, hence the term mixture of noise levels. We propose a transformer-based audiovisual latent diffusion model and show that it can be trained in a task-agnostic fashion using our approach to enable a variety of audiovisual generation tasks at inference time. Experiments demonstrate the versatility of our method in tackling cross-modal and multimodal interpolation tasks in the audiovisual space. Notably, our proposed approach surpasses baselines in generating temporally and perceptually consistent samples conditioned on the input. Project page: avdit2024.github.io

Xiaopeng Yuan, Xingjian Zhang, Ke Xu et al.

Large language models (LLMs) are increasingly used for tasks that require complex reasoning. Most benchmarks focus on final outcomes but overlook the intermediate reasoning steps - such as planning, revision, and decision making under resource constraints. We argue that measuring these internal processes is essential for understanding model behavior and improving reliability. We propose using strategic games as a natural evaluation environment: closed, rule-based systems with clear states, limited resources, and automatic feedback. We introduce a framework that evaluates LLMs along three core dimensions: planning, revision, and resource-constrained decision making. To operationalize this, we define metrics beyond win rate, including overcorrection risk rate, correction success rate, improvement slope, and over-budget ratio. In 4320 adversarial rounds across 12 leading models, ChatGPT-o3-mini achieves the top composite score, with a win rate of 74.7 percent, a correction success rate of 78.6 percent, and an improvement slope of 0.041. By contrast, Qwen-Plus, despite an overcorrection risk rate of 81.6 percent, wins only 25.6 percent of its matches - primarily due to excessive resource use. We also observe a negative correlation between overcorrection risk rate and correction success rate (Pearson r = -0.51, p = 0.093), suggesting that more frequent edits do not always improve outcomes. Our findings highlight the value of assessing not only what LLMs decide but how they arrive at those decisions

Jiannan Xiang, Yun Zhu, Lei Shu et al.

Developing and testing user interfaces (UIs) and training AI agents to interact with them are challenging due to the dynamic and diverse nature of real-world mobile environments. Existing methods often rely on cumbersome physical devices or limited static analysis of screenshots, which hinders scalable testing and the development of intelligent UI agents. We introduce UISim, a novel image-based UI simulator that offers a dynamic and interactive platform for exploring mobile phone environments purely from screen images. Our system employs a two-stage method: given an initial phone screen image and a user action, it first predicts the abstract layout of the next UI state, then synthesizes a new, visually consistent image based on this predicted layout. This approach enables the realistic simulation of UI transitions. UISim provides immediate practical benefits for UI testing, rapid prototyping, and synthetic data generation. Furthermore, its interactive capabilities pave the way for advanced applications, such as UI navigation task planning for AI agents. Our experimental results show that UISim outperforms end-to-end UI generation baselines in generating realistic and coherent subsequent UI states, highlighting its fidelity and potential to streamline UI development and enhance AI agent training.

Lijun Yu, Yijun Qian, Wenhe Liu et al.

Activity detection is one of the attractive computer vision tasks to exploit the video streams captured by widely installed cameras. Although achieving impressive performance, conventional activity detection algorithms are usually designed under certain constraints, such as using trimmed and/or object-centered video clips as inputs. Therefore, they failed to deal with the multi-scale multi-instance cases in real-world unconstrained video streams, which are untrimmed and have large field-of-views. Real-time requirements for streaming analysis also mark brute force expansion of them unfeasible. To overcome these issues, we propose Argus++, a robust real-time activity detection system for analyzing unconstrained video streams. The design of Argus++ introduces overlapping spatio-temporal cubes as an intermediate concept of activity proposals to ensure coverage and completeness of activity detection through over-sampling. The overall system is optimized for real-time processing on standalone consumer-level hardware. Extensive experiments on different surveillance and driving scenarios demonstrated its superior performance in a series of activity detection benchmarks, including CVPR ActivityNet ActEV 2021, NIST ActEV SDL UF/KF, TRECVID ActEV 2020/2021, and ICCV ROAD 2021.

Lijun Yu, Peng Chen, Wenhe Liu et al.

We focus on the problem of detecting traffic events in a surveillance scenario, including the detection of both vehicle actions and traffic collisions. Existing event detection systems are mostly learning-based and have achieved convincing performance when a large amount of training data is available. However, in real-world scenarios, collecting sufficient labeled training data is expensive and sometimes impossible (e.g. for traffic collision detection). Moreover, the conventional 2D representation of surveillance views is easily affected by occlusions and different camera views in nature. To deal with the aforementioned problems, in this paper, we propose a training-free monocular 3D event detection system for traffic surveillance. Our system firstly projects the vehicles into the 3D Euclidean space and estimates their kinematic states. Then we develop multiple simple yet effective ways to identify the events based on the kinematic patterns, which need no further training. Consequently, our system is robust to the occlusions and the viewpoint changes. Exclusive experiments report the superior result of our method on large-scale real-world surveillance datasets, which validates the effectiveness of our proposed system.

Lijun Yu, Dawei Zhang, Xiangqun Chen et al.

We propose a traffic danger recognition model that works with arbitrary traffic surveillance cameras to identify and predict car crashes. There are too many cameras to monitor manually. Therefore, we developed a model to predict and identify car crashes from surveillance cameras based on a 3D reconstruction of the road plane and prediction of trajectories. For normal traffic, it supports real-time proactive safety checks of speeds and distances between vehicles to provide insights about possible high-risk areas. We achieve good prediction and recognition of car crashes without using any labeled training data of crashes. Experiments on the BrnoCompSpeed dataset show that our model can accurately monitor the road, with mean errors of 1.80% for distance measurement, 2.77 km/h for speed measurement, 0.24 m for car position prediction, and 2.53 km/h for speed prediction.

Lijun Yu, Dawei Zhang, Xiangqun Chen et al.

Multiplayer Online Battle Arena (MOBA) is currently one of the most popular genres of digital games around the world. The domain of knowledge contained in these complicated games is large. It is hard for humans and algorithms to evaluate the real-time game situation or predict the game result. In this paper, we introduce MOBA-Slice, a time slice based evaluation framework of relative advantage between teams in MOBA games. MOBA-Slice is a quantitative evaluation method based on learning, similar to the value network of AlphaGo. It establishes a foundation for further MOBA related research including AI development. In MOBA-Slice, with an analysis of the deciding factors of MOBA game results, we design a neural network model to fit our discounted evaluation function. Then we apply MOBA-Slice to Defense of the Ancients 2 (DotA2), a typical and popular MOBA game. Experiments on a large number of match replays show that our model works well on arbitrary matches. MOBA-Slice not only has an accuracy 3.7% higher than DotA Plus Assistant at result prediction, but also supports the prediction of the remaining time of the game, and then realizes the evaluation of relative advantage between teams.