Algorithm-driven services, including search, recommendation, and advertising, rely on multi-stage model pipelines. While research prioritizes model optimization, post-deployment monitoring focuses on operational reliability. Yet operationally healthy systems may still produce poor-quality results, exposing a gap in existing evaluation and monitoring practices. However, a systematic problem formulation and benchmark for this task remain lacking. To bridge this gap, we formulate System Quality Diagnosis (SQD) for algorithm-driven services, a new task that extends quality evaluation beyond identifying poor outcomes to attributing their causes across system components. Focusing on industrial search, we propose WISE, an agentic framework for end-to-end quality diagnosis that discovers user-facing issues and traces their origins to underlying algorithmic components. To improve diagnostic reliability, we introduce ReTrace for constructing evidence-consistent trajectory supervision and Adaptive Harness-Policy Optimization (AHPO) for failure-guided optimization of the agent harness and model policy. To enable systematic evaluation, we construct WISE-BENCH, a real-world multimodal search benchmark that combines user-facing evidence, internal pipeline traces, and expert diagnosis annotations. Extensive experiments show consistent gains across model scales, with WISE achieving 75.20% end-to-end accuracy on Qwen3.5-35B-A3B. In production, WISE supports quality auditing, online quality assurance, and model optimization, reducing diagnosis costs and improving downstream relevance models.
Training data attribution (TDA) identifies which training examples most influenced a model's prediction. Influence function methods are a theoretically grounded family of TDA methods and exploit gradients. To overcome the scalability challenge arising from gradient computation, the most popular strategy is random projection (e.g., TRAK, LoGRA). However, this still faces two bottlenecks when scaling to large training sets and high-quality attribution: \emph{(i)} storing and loading projected per-example gradients for all $N$ training examples, where query latency is dominated by I/O; and \emph{(ii)} forming the $D \times D$ inverse Hessian approximation, which costs $O(D^2)$ memory. Both bottlenecks scale with the projection dimension $D$, yet increasing $D$ is necessary for attribution quality -- creating a quality--scalability tradeoff. We introduce \textbf{LoRIF} (\textbf{Lo}w-\textbf{R}ank \textbf{I}nfluence \textbf{F}unctions), which exploits low-rank structures of gradient to address both bottlenecks. First, we store rank-$c$ factors of projected per-example gradients rather than full matrices, reducing storage and query-time I/O from $O(D)$ to $O(c\sqrt{D})$ per layer per sample. Second, we use truncated SVD with the Woodbury identity to approximate the inverse Hessian term in an $r$-dimensional subspace, reducing memory from $O(D^2)$ to $O(Dr)$. On models from 0.1B to 70B parameters trained on datasets with millions of examples, LoRIF achieves up to 20$\times$ storage reduction and query-time speedup compared to LoGRA, while matching or exceeding its attribution quality. LoRIF makes gradient-based TDA practical at frontier scale. Code is available at https://github.com/doub7e/LoRIF.
Semantic change detection (SCD) extends the binary change detection task to provide not only the change locations but also the detailed "from-to" categories in multi-temporal remote sensing data. Such detailed semantic insights into changes offer considerable advantages for a wide array of applications. However, since SCD involves the simultaneous optimization of multiple tasks, the model is prone to negative transfer due to task-specific learning difficulties and conflicting gradient flows. To address this issue, we propose Graph-Aggregated Prototype Learning for Remote Sensing Semantic Change Detection(GAPL-SCD). In this framework, a multi-task joint optimization method is designed to optimize the primary task of semantic segmentation and change detection, along with the auxiliary task of graph aggregation prototype learning. Adaptive weight allocation and gradient rotation methods are used to alleviate the conflict between training tasks and improve multi-task learning capabilities. Specifically, the graph aggregation prototype learning module constructs an interaction graph using high-level features. Prototypes serve as class proxies, enabling category-level domain alignment across time points and reducing interference from irrelevant changes. Additionally, the proposed self-query multi-level feature interaction and bi-temporal feature fusion modules further enhance multi-scale feature representation, improving performance in complex scenes. Experimental results on the SECOND and Landsat-SCD datasets demonstrate that our method achieves state-of-the-art performance, with significant improvements in accuracy and robustness for SCD task.
Anh Nguyen, Tianyi Xu, Phat Tran, Harry Yu, Huazheng Wang, Stefan Lee
Recent work has improved computer-use agents (CUAs) by combining retrieved task-relevant tutorials and look-ahead planning using world models with MLLM judges. However, the contribution of each component -- tutorials, the judge, and the world model -- to overall improvement has not been evaluated in isolation. To fill this gap, we first construct OSWorld-Tutorial and ScienceBoard-Tutorial: two datasets for CUAs with curated multimodal tutorials, covering everyday and scientific domains. We then propose RAG-PolicyJudge (RAG-PJ), a policy-judge framework for CUAs conditioned on task-specific tutorials: the policy proposes candidate actions, and the judge selects among them. To isolate the effect of world models given tutorials, we compare RAG-PJ with and without common world models: using either zero-shot text-based descriptions of next states or an oracle that returns ground-truth visual observations. We find that RAG-PJ gains little benefit from these rollouts when tutorials are present, even with perfect visual futures. This suggests prior world-model gains in tutorial-conditioned methods come mainly from tutorial understanding and the judge's ability to select the proposed actions, rather than from predicted rollouts. Error analysis reveals that the bottleneck is not rollout fidelity, but the judge's ability to reason over future states and recognize task completion. This highlights the need to improve not just next-state prediction in world models, but also reasoning over their predicted rollouts. Datasets and code are available at https://github.com/realjoenguyen/RAGPJ-official.
With the rise of Group Agent such as Claude Tag, agents increasingly serve as digital workers for long-term team collaboration. These agents operate over shared contexts that continuously accumulate through multi-user interactions. However, existing agent memory systems are mainly designed for individual users and struggle to capture group-level knowledge in shared contexts, including group relationships and collaboratively formed knowledge. Manual memory strategies further limit adaptation to diverse and evolving shared contexts. To this end, we propose \textbf{EvoGroup (EvoG)}, a self-evolving harness for group agents that starts from a minimal six-tool cold start and adapts through interaction with shared context. EvoG uses confidence and task outcomes to select informative trajectories, while also eliciting self-reflections on uncertainty in low-confidence cases. The analysis agent then examines selected trajectories through adaptive progressive disclosure, followed by bucketed analysis to synthesize high-level findings that guide the evolve agent in revising the Harness. Empirically, pass@1 on held-out split of EverMemBench improves from 71.07\% to 89.58\%, surpassing both self-evolving memory and human-designed harnesses, including Claude Code (85.83\%). To evaluate generalization, we freeze the evolved harness and evaluate it on the unseen GroupMemBench, where, with GPT-5.5, it demonstrates effective generalization and outperforms both memory baselines and human-designed harnesses.
Jinyuan Chang, Chenguang Duan, Yuling Jiao, Ruoxuan Li, Yi Xu, Jerry Zhijian Yang
Inference-time alignment for diffusion models aims to adapt a pre-trained reference diffusion model toward a target distribution without retraining the reference score network, thereby preserving the generative capacity of the reference model while enforcing desired properties at inference time. A central mechanism for achieving such alignment is guidance, which modifies the sampling dynamics through an additional drift term. In this work, we introduce variationally stable Doob's matching, a novel framework for provable guidance estimation grounded in Doob's $h$-transform. Our approach formulates guidance as the gradient of the logarithm of an underlying Doob's $h$-function and employs gradient-regularized regression to simultaneously estimate both the $h$-function and its gradient, resulting in a consistent estimator of the guidance. Theoretically, we establish non-asymptotic convergence rates for the estimated guidance. Moreover, we analyze the resulting controllable diffusion processes and prove non-asymptotic convergence guarantees for the generated distributions in the Wasserstein-2 distance. Finally, we show that variationally stable guidance estimators are adaptive to unknown low dimensionality, effectively mitigating the curse of dimensionality under low-dimensional linear subspace assumptions. Numerical experiments demonstrate the effectiveness of our proposed method and show the advantages compared to the baselines.
We present Reap, an open framework for learning and analyzing formal proof search in Lean4. Reap implements policy-value-guided Monte Carlo Tree Search directly in Lean, sharing proof-state transitions, AND-OR search, and replay verification between inference and training rollouts. Structured action logs expose how neural predictions and tactic execution interact during search. Using this framework, we train Reaper-1, a 1.7B-parameter policy-value model, through continued pre-training, supervised tactic learning, and iterative search-based reinforcement learning. Using a separate premise selector, the final system solves 79.9% of miniF2F-test at pass@32, establishing state-of-the-art performance among open provers of comparable size while remaining competitive with substantially larger models. Beyond improved solve rates, our results suggest that RL strengthens the prover's ability to organize proofs around useful intermediate results, favoring substantial advances over repeated local adjustments. This more directed proof construction is accompanied by increasingly elaborate tactic programs: fewer explicit search decisions are needed, while more work is performed within each decision. The resulting trade-off between search effort and tactic execution cost provides insight into how learning changes the prover's problem-solving strategy.
Social annotation platforms such as Perusall and Hypothesis are widely used to support student engagement with course readings, often through required commenting. Yet visible participation does not necessarily produce sustained or substantive peer exchange. Analyzing six weeks of Perusall activity from a graduate course (2,873 comments), we found that participation was often fragmented: 53.5% of comments received no reply, 38.3% were posted within the final 12 hours before the deadline, and over one-third of a double-coded sample went no further than repeating or agreeing with the reading or a peer. Interviews pointed to two linked difficulties: locating a discussion worth entering and contributing something that meaningfully advances it. We present Lookover, an AI-assisted annotation system that scaffolds both stages without writing for students. Discussion Overview helps students identify active themes and relevant threads, while Spark Thinking poses questions that encourage them to move beyond reaction and restatement toward questioning, extending, and substantively responding to peers. A preliminary study with five participants found Discussion Overview valuable for those who noticed it, while the usefulness of Spark Thinking depended on how much a participant already knew about the topic.
Jiuyi Xu, Xiao Hu, Meida Chen, Song Wang, Yang Ye, Yangming Shi
Post-training quantization (PTQ) reduces the memory footprint of vision-language-action (VLA) models, but lowering bit widths while maintaining task performance remains challenging. Existing work has used sensitivity estimates from calibration data to assign bit widths to different model components. However, using task success directly to guide mixed-precision allocation is underexplored because closed-loop evaluations are costly and the effect of quantizing one component depends on the precision of other components. To this end, we propose \textbf{QbD-VLA}, a task-guided mixed-precision quantization method that uses task success from short closed-loop evaluations to assign weight and activation bit widths while accounting for interactions between model components. Experiments covering four VLA models and all four LIBERO suites show strong task performance at average precisions of approximately 4 bits for weights and fewer than 6 bits for activations. On SIMPLER, the gain over uniform quantization is small. On a Franka Research 3 robot arm, simulation-selected allocations show no detected success-rate loss on X-VLA but fail on $π_{0.5}$. These findings support using task success to guide mixed-precision VLA quantization, while highlighting the need for real-world evaluation before deployment. Code is available at \href{https://github.com/jiuyixu25/QbD-VLA}{\texttt{github.com/jiuyixu25/QbD-VLA}}.
Robot imitation learning uses demonstrations from operators of varying skill, but selecting training data remains challenging. Existing methods score demonstrations using smoothness, diversity, or estimated influence on policy performance, but differences in selection and evaluation protocols make it difficult to determine whether they outperform simple baselines. We introduce \textbf{CuraEval}, a framework combining mathematical analysis of demonstration scores with controlled comparisons against fixed size-matched random subsets and full-data training on mixed-skill robomimic data using Diffusion Policy and BC-RNN. We show that the evaluated PSD score depends only on episode length and squared position magnitudes, and produces nearly the same ranking as episode length on \textit{square} ($ρ=0.999$). With Diffusion Policy, selecting the shortest 50\% of demonstrations improves \textit{square} success from 70.7\% under random selection to 76.5\%. Compared with random selection, our adapted geometric-entropy LOO selector reduces success at 25\% retention from 94.7\% to 74.0\% on \textit{can} and achieves 41.7\% on \textit{square}, below the mean success rates of all three independently sampled random subsets (59.8--62.5\%). On the same 3 \textit{square} Diffusion Policy training seeds, length-based selection at 75\% retention improves success over full-data training from 76.7\% to 86.7\% under the original protocol, but lowers it from 80.5\% to 79.5\% after checkpoint reporting, rollout count, and evaluation seed change jointly. No evaluated curated condition significantly outperforms full-data training after Holm correction, although confidence intervals still permit useful gains. Code is available at https://github.com/jiuyixu25/CuraEval.
Single-linkage clustering (SLC) is a fundamental method for hierarchical data analysis. In the distance setting, a $k$-clustering produced by SLC can be obtained by computing a minimum spanning tree (MST) and deleting its $k-1$ heaviest edges. This naturally induces a cost profile for the SLC hierarchy: for each $k\in[n]$, we define $\mathrm{cost}_k$ to be the weight of the resulting $k$-component spanning forest, equivalently, the minimum total weight of any spanning forest with exactly $k$ connected components. The corresponding \emph{SLC cost profile} is $(\mathrm{cost}_1,\ldots,\mathrm{cost}_n)$, and the scalar quantity $\mathrm{cost}(G)=\sum_{k=1}^{n}\mathrm{cost}_k$ is the area under this profile. We study the problem of approximating these quantities in sublinear time. We assume that the input is a weighted graph $G$ of average degree $d$ with edge weights in $\{1,\dots,W\}$, accessed through adjacency-list queries; missing edges are treated as having infinite distance. Our main result is a sampling-based algorithm that outputs a succinct sketch of the entire SLC cost profile in the distance setting. The algorithm runs in $\widetilde{O}(d\sqrt{W}/\varepsilon^3)$ time and returns a sketch from which one can derive estimates $(\widehat{\mathrm{cost}}_1,\ldots,\widehat{\mathrm{cost}}_n)$ satisfying $\sum_{k=1}^{n}\bigl|\widehat{\mathrm{cost}}_k-\mathrm{cost}_k\bigr| \le \varepsilon\,\mathrm{cost}(G)$. Thus, we obtain an $\ell_1$ approximation to the full profile whose error is at most an $\varepsilon$-fraction of the area under the true profile. In particular, this yields a $(1\pm\varepsilon)$-approximation to $\mathrm{cost}(G)$ within the same running time. We also prove a nearly matching lower bound of $Ω(d\sqrt{W}/\varepsilon^2)$ queries for estimating $\mathrm{cost}(G)$.
Linhan Wang, Zijian An, Mingyuan Zhang, Chen Dai, Yi Xu, Can Cui, Jiayan Wang, Zichong Yang 等 11 位作者
Video generative models provide rich physical priors for robot learning, yet existing world-action models (WAMs) face a fundamental trade-off: synchronous video generation at control rate is latency-prohibitive, while abandoning test-time visual imagination sacrifices task success. We show that visual imagination achieves both real-time inference and superior success rates when generated asynchronously off the critical path and consumed directly in latent space. We introduce GlanceWAM, which decouples imagination from control on a single shared video DiT backbone: an asynchronous proposer glances ahead on a slow clock to imagine a single lookahead frame seconds into the future in the background, while an action head decodes action chunks at control rate (48 ms) purely in latent space without blocking. Enabled by a non-interfering attention mask that isolates video representations and staleness-robust horizon training that accommodates asynchronous lookahead aging, GlanceWAM breaks the speed-success dilemma. Trained purely on demonstrations, it attains 72.2% on the 24-task RoboCasa kitchen benchmark (vs. 67.1% for synchronous Cosmos Policy) and 99.0% on LIBERO while cutting per-chunk control latency 24x relative to synchronous world-action models (48 ms on one A100). In single-arm and bimanual real-robot manipulation, it achieves higher average success than $π_{0.5}$ without any robot-data pretraining. Code is available at https://github.com/linhanwang/GlanceWAM and real-robot videos at https://linhanwang.github.io/glancewam/.
Modern automatic speech recognition (ASR) systems have achieved remarkable accuracy by scaling model depth and capacity, but at the cost of increased memory, energy, and computational demands. On edge devices where ASR is often most needed, such as watches and glasses, deploying such large models is infeasible due to extremely constrained resources. This raises a fundamental question: can we achieve high representational power in deep ASR models without scaling up parameterization? In this work, we revisit the role of depth and identify layer wise representational dynamics, in which most layers learn functionally similar transformations. Motivated by this insight, we propose a block-recurrent ASR architecture that replaces parameterized depth with a small set of recurrent blocks, each consisting of weight-shared layers, thereby preserving effective depth while drastically reducing model size. Through representation-guided grouping and knowledge distillation, block-recurrent models retain the accuracy of large reference models while using an order of magnitude fewer parameters. Across Open ASR Leaderboard benchmarks, a model with only two shared blocks recovers 97.4% of the accuracy of large models on average. The compact backbone further enables efficient on-device personalization through lightweight MoE-LoRA adaptation, allowing user-specific models to match or surpass the accuracy of general-purpose ASR systems with significantly lower inference cost. By reusing weights across recurrent steps, the proposed architecture reduces memory-access overhead and inference energy consumption. Together, these results show that high quality ASR can be achieved without large parameterization, and that structured recurrence provides a principled path towards ultra-compressed, energy-efficient, and personalized speech recognition on the edge.
Sean Bin Yang, Zongyi Xu, Jilin Hu, Ying Sun, Kristian Torp, Bin Yang, Torben Bach Pedersen, Christian S. Jensen
Urban region representation learning encodes heterogeneous urban signals, such as mobility flows, points of interest, and land-use patterns, into unified embeddings for downstream tasks, including check-in prediction, crime forecasting, and service demand estimation. Existing approaches rely on globally shared fusion and joint optimization, implicitly assuming uniform modality importance and naturally balanced multi-objective training. However, urban data is inherently heterogeneous. Our empirical analysis reveals two structural issues: (i) modality utility varies markedly across regions, causing static fusion to overemphasize signals from certain modalities; and (ii) heterogeneous objectives exhibit persistent gradient imbalance, causing representational bias and unstable training. We present PRISM, a self-adjusting framework for adaptive multi-view learning. PRISM adapts sparse Mixture-of-Experts (MoE) routing to learn region-specific view allocations, while a Proportional-Integral-Derivative (PID)-inspired regulation mechanism mitigates gradient conflicts during optimization. Experiments on three large-scale urban datasets show that PRISM achieves more competitive performance than strong multi-view baselines, while providing more stable optimization behavior and more balanced region representations.
Generating humorous memes is a challenging multimodal task that moves beyond direct image-to-caption supervision. It requires a nuanced reasoning over visual content, contextual cues, and subjective humor. To bridge this gap between visual perception and humorous punchline creation, we propose HUMOR}, a novel framework that guides VLMs through hierarchical reasoning and aligns them with group-wise human preferences. First, HUMOR employs a hierarchical, multi-path Chain-of-Thought (CoT): the model begins by identifying a template-level intent, then explores diverse reasoning paths under different contexts, and finally anchors onto a high-quality, context-specific path. This CoT supervision, which traces back from ground-truth captions, enhances reasoning diversity. We further analyze that this multi-path exploration with anchoring maintains a high expected humor quality, under the practical condition that high-quality paths retain significant probability mass. Second, to capture subjective humor, we train a pairwise reward model that operates within groups of memes sharing the same template. Following established theory, this approach ensures a consistent and robust proxy for human preference, even with subjective and noisy labels. The reward model then enables a group-wise reinforcement learning optimization, guaranteeing providing a theoretical guarantee for monotonic improvement within the trust region. Extensive experiments show that HUMOR empowers various VLMs with superior reasoning diversity, more reliable preference alignment, and higher overall meme quality. Beyond memes, our work presents a general training paradigm for open-ended, human-aligned multimodal generation, where success is guided by comparative judgment within coherent output group.
Yanlun Tu, Huacan Wang, Jin Xie, Ziyue Zhou, Jie Zhou, Ningyan Zhu, Ge Chen, Wangyi Chen 等 14 位作者
In industrial practice, AI-generated programmable logic controller (PLC) code must integrate with existing projects and implement the required behavior across scan cycles. Compilation alone does not establish runtime behavior, and subsequent edits can invalidate earlier verification results. We present SemaPLC, a project-grounded, verification-gated agent harness for PLC code generation. The harness combines project context and reusable PLC skills with specification review, compilation, and live-runtime validation. Its completion gate requires recorded checks to meet acceptance criteria for the current implementation. Edits invalidate previous verdicts, and failed checks guide bounded repair. We evaluate SemaPLC on 117 independent-POU tasks and 65 project-context tasks across seven language-model backbones. SemaPLC outperforms the evaluated baselines on both metrics for every backbone. Its mean verified-pass rate and dynamic score reach 72.6% and 52.2, exceeding the strongest baseline means by 8.7% points and 20.8 points, respectively. Verification-layer and failure analyses reveal how build recovery and output observability affect end-to-end results and identify behavioral errors that remain after compilation. We apply the harness in KUKA's iiQworks product environment for PLC development and discuss engineering lessons for its industrial integration and deployment. The industrial use indicates the agent assistance substantially improves PLC development efficiency by 45%.
Xinyi Xu, Bingguang Hao ByteDance Inc., Yongyi Xiong, Zimo Chen University of Science and Technology of China, Xinchen Liu, Hongxin Guo, Xuelong Wang, Silin Zhou 等 9 位作者
Self-deprecation is a prevalent communicative strategy in human society, often using image-text interplay to express emotions and intentions. Despite self-deprecation is widespread in real-world conversations, the ability of multimodal large language models (MLLMs) to understand it remains underexplored. To fill this gap, we introduce JanusMM, the first benchmark designed to evaluate MLLMs’ understanding of self-deprecation in real-world conversations. JanusMM contains 2,016 bilingual memes from three types of social interactions and provides a dual-task evaluation framework with six new metrics. The first task assesses MLLMs’ abilities in self-deprecation recognition and reasoning, while the second task evaluates the consistency of their understanding by simulating the perspectives of the initiator and responder. We evaluate ten frontier MLLMs and find that they exhibit weak recognition and reasoning abilities, with their understanding of self-deprecation remaining inconsistent across both perspectives.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Hao Yu, Tianyi Xu , McGill University, Michael A. Hedderich Ludwig-Maximilians-Universität München, Wassim Hamidouche Microsoft and INSA Rennes, Syed Waqas Zamir, David Ifeoluwa Adelani McGill University
Large language models (LLMs) are increasingly multilingual, yet open models continue to underperform relative to proprietary systems, with the gap most pronounced for African languages. Continued pre-training (CPT) offers a practical route to language adaptation, but improvements on demanding capabilities such as mathematical reasoning often remain limited. This limitation is driven in part by the uneven domain coverage and missing task-relevant knowledge that characterize many low-resource language corpora. We present AfriqueLLM, a suite of open LLMs adapted to 20 African languages through CPT on 26B tokens. We perform a comprehensive empirical study across five base models spanning sizes and architectures, including Llama 3.1, Gemma 3, and Qwen 3, and systematically analyze how CPT data composition shapes downstream performance. In particular, we vary mixtures that include math, code, and synthetic translated data, and evaluate the resulting models on a range of multilingual benchmarks. Our results identify data composition as the primary driver of CPT gains. Adding math, code, and synthetic translated data yields consistent improvements, including on reasoning-oriented evaluations. Within a fixed architecture, larger models typically improve performance, but architectural choices dominate scale when comparing across model families. Moreover, strong multilingual performance in the base model does not reliably predict post-CPT outcomes; robust architectures coupled with task-aligned data provide a more dependable recipe. Finally, our best models improve long-context performance, including document-level translation.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
AI 速览：Plan-X 用可学习的多模态语言模型作为语义规划器，先生成时空语义 token 序列作为“语义草图”，再指导扩散 Transformer 生成与指令对齐的视频。
Diffusion Transformers have demonstrated remarkable capabilities in visual synthesis, yet they often struggle with high-level semantic reasoning and long-horizon planning. This limitation frequently leads to visual hallucinations and mis-alignments with user instructions, especially in scenarios involving complex scene understanding, human-object interactions, multi-stage actions, and in-context motion reasoning. To address these challenges, we propose Plan-X, a framework that explicitly enforces high-level semantic planning to instruct video generation process. At its core lies a Semantic Planner, a learnable multimodal language model that reasons over the user's intent from both text prompts and visual context, and autoregressively generates a sequence of text-grounded spatio-temporal semantic tokens. These semantic tokens, complementary to high-level text prompt guidance, serve as structured "semantic sketches" over time for the video diffusion model, which has its strength at synthesizing high-fidelity visual details. Plan-X effectively integrates the strength of language models in multimodal in-context reasoning and planning, together with the strength of diffusion models in photorealistic video synthesis. Extensive experiments demonstrate that our framework substantially reduces visual hallucinations and enables fine-grained, instruction-aligned video generation consistent with multimodal context.