Flow matching has recently emerged as a strong paradigm for state-of-the-art text-to-image (T2I) generation, enabling high-quality generation with a small number of sampling steps. As these models are increasingly integrated into real-world applications, ensuring safe and non-sensitive content generation has become a critical requirement. However, adapting safety and concept removal methods to this new generation framework remains an open challenge. Specifically, prior methods largely rely on iterative trajectory steering across a number of denoising steps or on CLIP-centric prompt embedding manipulation. These design assumptions pose fundamental bottlenecks for safety in flow matching-based T2I generation, where limited sampling steps constrain iterative correction and modern context-aware text encoders diminish the effectiveness of embedding-level interventions. In this paper, we propose VESFlow, a training-free safety method tailored to flow matching with extremely few sampling steps. Leveraging the fact that flow matching models learn the marginal velocity, we directly edit the velocity field via a safe-conditional posterior. VESFlow steers the trajectory toward safe outputs while leaving the conditioning prompt unchanged. Building on the observation that VESFlow leaves outputs unchanged under benign prompts, we further introduce a risk score-based filtering that bypasses velocity editing to reduce computational cost while preserving benign prompt generation. Based on this filtering, we propose VESFlow+, a stronger variant of VESFlow that not only edits the velocity toward the safe direction, but also pushes it away from the unsafe direction. Experimental results show that VESFlow+ removes the target concept, reducing the attack success rate by NudeNet to 6.3% on Ring-A-Bell and 6.8% on MMA-Diffusion on the 4-step MeanFlow model, while preserving fidelity on benign prompts.
展开完整摘要收起摘要↓
Flow matching has recently emerged as a strong paradigm for state-of-the-art text-to-image (T2I) generation, enabling high-quality generation with a small number of sampling steps. As these models are increasingly integrated into real-world applications, ensuring safe and non-sensitive content generation has become a critical requirement. However, adapting safety and concept removal methods to this new generation framework remains an open challenge. Specifically, prior methods largely rely on iterative trajectory steering across a number of denoising steps or on CLIP-centric prompt embedding manipulation. These design assumptions pose fundamental bottlenecks for safety in flow matching-based T2I generation, where limited sampling steps constrain iterative correction and modern context-aware text encoders diminish the effectiveness of embedding-level interventions. In this paper, we propose VESFlow, a training-free safety method tailored to flow matching with extremely few sampling steps. Leveraging the fact that flow matching models learn the marginal velocity, we directly edit the velocity field via a safe-conditional posterior. VESFlow steers the trajectory toward safe outputs while leaving the conditioning prompt unchanged. Building on the observation that VESFlow leaves outputs unchanged under benign prompts, we further introduce a risk score-based filtering that bypasses velocity editing to reduce computational cost while preserving benign prompt generation. Based on this filtering, we propose VESFlow+, a stronger variant of VESFlow that not only edits the velocity toward the safe direction, but also pushes it away from the unsafe direction. Experimental results show that VESFlow+ removes the target concept, reducing the attack success rate by NudeNet to 6.3% on Ring-A-Bell and 6.8% on MMA-Diffusion on the 4-step MeanFlow model, while preserving fidelity on benign prompts.
Physical simulations for intricate geometries with path-dependent constitutive models face difficulties due to the enormous computational cost they require. Recently, the emergence of generative AI models, which succeed in image and video synthesis tasks, has provided a promise to further improve simulations. Although U-Net-based denoising diffusion probabilistic models (DDPMs) have been adopted for elastic stress field generation, they typically require hundreds of sampling steps, and applications of generative models to path-dependent, e.g. plastic, stress fields remain very limited. In this work, we propose a novel flow matching (FM) model based on a transformer backbone for high-resolution path-dependent stress field generation with stochastic loading-unloading paths and geometry. The proposed model operates within the latent space of a variational autoencoder (VAE) and formulates the simulation of plastic fields as a video synthesis task, directly generating the stress fields across all time steps. Meanwhile, we design a non-Gaussian source distribution for flow matching, such that crossings among conditional transport paths are reduced during training. This enables our model to generate satisfactory samples in one step without relying on distillation. In addition, we introduce token-level loading embeddings and two auxiliary networks to further enhance the model performance in path-dependent simulation. The results demonstrate that, even with a limited training dataset, our model can accurately generate high-resolution path-dependent fields. It is much more computationally efficient than finite element analysis, providing a speedup of 6 to 7 times over FEM on CPUs and approximately two orders of magnitude speedup on consumer-grade GPUs.
展开完整摘要收起摘要↓
Physical simulations for intricate geometries with path-dependent constitutive models face difficulties due to the enormous computational cost they require. Recently, the emergence of generative AI models, which succeed in image and video synthesis tasks, has provided a promise to further improve simulations. Although U-Net-based denoising diffusion probabilistic models (DDPMs) have been adopted for elastic stress field generation, they typically require hundreds of sampling steps, and applications of generative models to path-dependent, e.g. plastic, stress fields remain very limited. In this work, we propose a novel flow matching (FM) model based on a transformer backbone for high-resolution path-dependent stress field generation with stochastic loading-unloading paths and geometry. The proposed model operates within the latent space of a variational autoencoder (VAE) and formulates the simulation of plastic fields as a video synthesis task, directly generating the stress fields across all time steps. Meanwhile, we design a non-Gaussian source distribution for flow matching, such that crossings among conditional transport paths are reduced during training. This enables our model to generate satisfactory samples in one step without relying on distillation. In addition, we introduce token-level loading embeddings and two auxiliary networks to further enhance the model performance in path-dependent simulation. The results demonstrate that, even with a limited training dataset, our model can accurately generate high-resolution path-dependent fields. It is much more computationally efficient than finite element analysis, providing a speedup of 6 to 7 times over FEM on CPUs and approximately two orders of magnitude speedup on consumer-grade GPUs.
作者Zi-Yuan Hu, Zicong Tang, Shijia Huang, Yanyang Li, Michael R. Lyu, Liwei Wang
Recent advancements in chain-of-thought (CoT) reasoning have shown promise in enhancing video understanding and reasoning capabilities of multimodal large language models (MLLMs). However, existing CoT-based MLLMs require labor-intensive CoT annotations and incur substantial training and inference overhead. While visual latent reasoning has emerged as a more efficient alternative, existing methods primarily focus on image tasks and heavily rely on additional supervision signals for visual latent generation (e.g., CoT traces, auxiliary images, or fine-grained annotations), limiting their scalability and transferability to video tasks. To bridge this gap, we introduce VideoLatent, a novel MLLM equipped with a latent injection module tailored for video understanding and reasoning. Specifically, VideoLatent learns to perform visual latent reasoning using a new latent self-forcing training paradigm, which comprises latent alignment and latent diversity objectives, and relies solely on standard video-question-answer triplets. Extensive experiments across 14 benchmarks demonstrate that our model consistently outperforms existing standard and latent MLLMs on general video understanding and complex video reasoning. Compared with Video-R1, our VideoLatent achieves superior computational efficiency, reducing training/inference overhead by $\sim$6$\times$/$\sim$68$\times$. Moreover, experiments demonstrate that our method has strong generalizability to different MLLM backbones and different model scales.
展开完整摘要收起摘要↓
Recent advancements in chain-of-thought (CoT) reasoning have shown promise in enhancing video understanding and reasoning capabilities of multimodal large language models (MLLMs). However, existing CoT-based MLLMs require labor-intensive CoT annotations and incur substantial training and inference overhead. While visual latent reasoning has emerged as a more efficient alternative, existing methods primarily focus on image tasks and heavily rely on additional supervision signals for visual latent generation (e.g., CoT traces, auxiliary images, or fine-grained annotations), limiting their scalability and transferability to video tasks. To bridge this gap, we introduce VideoLatent, a novel MLLM equipped with a latent injection module tailored for video understanding and reasoning. Specifically, VideoLatent learns to perform visual latent reasoning using a new latent self-forcing training paradigm, which comprises latent alignment and latent diversity objectives, and relies solely on standard video-question-answer triplets. Extensive experiments across 14 benchmarks demonstrate that our model consistently outperforms existing standard and latent MLLMs on general video understanding and complex video reasoning. Compared with Video-R1, our VideoLatent achieves superior computational efficiency, reducing training/inference overhead by $\sim$6$\times$/$\sim$68$\times$. Moreover, experiments demonstrate that our method has strong generalizability to different MLLM backbones and different model scales.
作者Yutian Cheng, Xiaojian Ma, Xianhao Wang, Min Yang, Rongpeng Su, Hangxin Liu, Xi Chen, Shuai Li, Qing Li
Neural world models coupled with model predictive control (MPC) replan at every environment step to bound accumulated prediction error, but this incurs substantial computational overhead. Reusing a cached plan reduces this overhead, yet its effectiveness depends on how prediction mismatch propagates through the local dynamics. We analyze this trade-off with a perturbation-based dynamic-regret framework and show that stale-plan penalties scale with the reuse tolerance, the accumulated mismatch since the last replanning step, and the local dynamics sensitivity. Based on this structure, we propose AdaReP, a training-free wrapper that adapts the replanning tolerance online using the current deviation from the cached rollout and a local sensitivity estimate, without modifying the learned world model or planner. Across image-space planning, latent-space control, and real-world robotic manipulation, AdaReP substantially reduces planner-side computation while maintaining comparable task performance, including over 80% fewer queries on a 50-trial physical robot study.
展开完整摘要收起摘要↓
Neural world models coupled with model predictive control (MPC) replan at every environment step to bound accumulated prediction error, but this incurs substantial computational overhead. Reusing a cached plan reduces this overhead, yet its effectiveness depends on how prediction mismatch propagates through the local dynamics. We analyze this trade-off with a perturbation-based dynamic-regret framework and show that stale-plan penalties scale with the reuse tolerance, the accumulated mismatch since the last replanning step, and the local dynamics sensitivity. Based on this structure, we propose AdaReP, a training-free wrapper that adapts the replanning tolerance online using the current deviation from the cached rollout and a local sensitivity estimate, without modifying the learned world model or planner. Across image-space planning, latent-space control, and real-world robotic manipulation, AdaReP substantially reduces planner-side computation while maintaining comparable task performance, including over 80% fewer queries on a 50-trial physical robot study.
作者Ruiliang Zhou, Xuecheng Wu, Kang He, Guangyun Han, Bin Liu, Qinqin Chen, Wende Xu, Qingjie Zhao, Chengru Song
While Diffusion Transformers (DiTs) have revolutionized high-fidelity video generation, their reliance on 3D full attention creates a quadratic computational bottleneck. Existing sparse methods face a dilemma: dynamic pruning suffers from prohibitive runtime overhead and memory fragmentation, while static heuristics fail to capture fine-grained dependencies. In this work, we propose ScalingAttention, a training-free framework grounded in a key inductive bias: while individual activations are input-dependent, the high-mass attention regions for each head rapidly converge to a stable, prompt-agnostic Intrinsic Sparse Topology. This topology is weight-encoded, scale-invariant, and efficient to extract. ScalingAttention decouples topology discovery from sparsity control via: (1) WEST (Weight-Encoded Sparse Topology), which extracts a robust block-sparse prior mask offline to eliminate runtime search; (2) FAST (Fidelity-Aware Sensitivity Tuning), which adaptively tunes head-wise sparsity based on diffusion fidelity requirements. To ensure practical acceleration, we co-design a hardware-aligned bit-wise block-sparse kernel. Experiments on Wan2.1 show up to 1.90X end-to-end speedup with superior fidelity, establishing a new Pareto frontier over state-of-the-art baselines.
展开完整摘要收起摘要↓
While Diffusion Transformers (DiTs) have revolutionized high-fidelity video generation, their reliance on 3D full attention creates a quadratic computational bottleneck. Existing sparse methods face a dilemma: dynamic pruning suffers from prohibitive runtime overhead and memory fragmentation, while static heuristics fail to capture fine-grained dependencies. In this work, we propose ScalingAttention, a training-free framework grounded in a key inductive bias: while individual activations are input-dependent, the high-mass attention regions for each head rapidly converge to a stable, prompt-agnostic Intrinsic Sparse Topology. This topology is weight-encoded, scale-invariant, and efficient to extract. ScalingAttention decouples topology discovery from sparsity control via: (1) WEST (Weight-Encoded Sparse Topology), which extracts a robust block-sparse prior mask offline to eliminate runtime search; (2) FAST (Fidelity-Aware Sensitivity Tuning), which adaptively tunes head-wise sparsity based on diffusion fidelity requirements. To ensure practical acceleration, we co-design a hardware-aligned bit-wise block-sparse kernel. Experiments on Wan2.1 show up to 1.90X end-to-end speedup with superior fidelity, establishing a new Pareto frontier over state-of-the-art baselines.
作者Kai Zeng, Moran Li, Zhengwei Wang, Yingchen Yu, Yiheng Lin, Ruichuan An, Ming Lu, Qi She, Wentao Zhang
Visual text editing aims to precisely modify text in images and videos while preserving stylistic consistency and visual realism. Despite significant advances in the image domain, video text editing remains largely unexplored: it is a localized task demanding stroke-level precision within small text regions, which compounds the challenges of cross-frame accuracy, temporal coherence, and stylistic fidelity. We introduce SteerVTE, a unified framework that \underline{steer}s a frozen video diffusion model to perform precise \underline{V}ideo \underline{T}ext \underline{E}diting through style and glyph control. Built on a frozen diffusion transformer, SteerVTE attaches a lightweight text context adapter with two complementary modules: a style encoder capturing the original text's visual attributes, and dual-granularity glyph encoders encoding the target text at both the line and character levels. To overcome the inherently weak text rendering priors of video foundation models, we further propose a glyph-aware spatial-focal loss and a three-stage progressive training curriculum that scales from image to video data. To support large-scale training, we also develop an automatic synthesis pipeline and construct SteerVTE-1M, a dataset of one million triplets spanning diverse scenes, fonts, and stylistic effects. Extensive experiments demonstrate that SteerVTE substantially outperforms existing video editing baselines across text accuracy, style consistency, and temporal coherence.
展开完整摘要收起摘要↓
Visual text editing aims to precisely modify text in images and videos while preserving stylistic consistency and visual realism. Despite significant advances in the image domain, video text editing remains largely unexplored: it is a localized task demanding stroke-level precision within small text regions, which compounds the challenges of cross-frame accuracy, temporal coherence, and stylistic fidelity. We introduce SteerVTE, a unified framework that \underline{steer}s a frozen video diffusion model to perform precise \underline{V}ideo \underline{T}ext \underline{E}diting through style and glyph control. Built on a frozen diffusion transformer, SteerVTE attaches a lightweight text context adapter with two complementary modules: a style encoder capturing the original text's visual attributes, and dual-granularity glyph encoders encoding the target text at both the line and character levels. To overcome the inherently weak text rendering priors of video foundation models, we further propose a glyph-aware spatial-focal loss and a three-stage progressive training curriculum that scales from image to video data. To support large-scale training, we also develop an automatic synthesis pipeline and construct SteerVTE-1M, a dataset of one million triplets spanning diverse scenes, fonts, and stylistic effects. Extensive experiments demonstrate that SteerVTE substantially outperforms existing video editing baselines across text accuracy, style consistency, and temporal coherence.
Diffusion policies enable multimodal robot behavior but offer limited ability to choose among behavior modes at inference time, even though such control is desirable in human-robot settings. Prior solutions to this lack of control have utilized Signal Temporal Logic (STL) to express human intentions and provide corresponding guidance for diffusion policy inference. However, these approaches can only guide diffusion policies that jointly generate future actions and states, increasing both complexity and runtime. We propose a novel guidance method for action-only diffusion policies that uses a separate learned world model to enable differentiable evaluation of STL robustness, with its gradient then injected into the diffusion process. This steers behavior toward constraint satisfaction without retraining, improving constraint adherence while preserving task performance. On the Can Transport task from Robomimic, our method maintains 100% task success while reducing constraint violations from over 80% for baseline methods to 4%. We also discuss extensions toward improved robustness and more complex constraints.
展开完整摘要收起摘要↓
Diffusion policies enable multimodal robot behavior but offer limited ability to choose among behavior modes at inference time, even though such control is desirable in human-robot settings. Prior solutions to this lack of control have utilized Signal Temporal Logic (STL) to express human intentions and provide corresponding guidance for diffusion policy inference. However, these approaches can only guide diffusion policies that jointly generate future actions and states, increasing both complexity and runtime. We propose a novel guidance method for action-only diffusion policies that uses a separate learned world model to enable differentiable evaluation of STL robustness, with its gradient then injected into the diffusion process. This steers behavior toward constraint satisfaction without retraining, improving constraint adherence while preserving task performance. On the Can Transport task from Robomimic, our method maintains 100% task success while reducing constraint violations from over 80% for baseline methods to 4%. We also discuss extensions toward improved robustness and more complex constraints.
Maintaining physical consistency in video generators and world models increasingly relies on vision-language models (VLMs) as automated judges that provide reward signals, ranking decisions, and data-filtering criteria. Yet VLMs differ substantially in training data and architecture, encoding physical phenomena through distinct internal representations. A single global evaluation schema therefore gives every VLM the same axes of competence, regardless of what each can actually perceive. We propose JudgeFit, an iterative refinement procedure that discovers a per-VLM evaluation taxonomy. An initial taxonomy is constructed by prompting the target VLM to enumerate physics errors on a small set of videos and clustering the resulting descriptions. The taxonomy is then refined through a diagnostic step: we calibrate the VLM's per-dimension scores to human physical-commonsense ratings, diagnose which dimensions it scores unreliably or redundantly, and prompt an LLM to repair them, iterating until convergence. We further instantiate this procedure as a benchmark and apply it to 16 VLMs spanning eight model families. The refined taxonomy outperforms the global-schema baseline on held-out videos for every VLM tested, with a mean relative improvement of approximately 32%. Beyond aggregate accuracy, the per-VLM profiles expose model-specific blind spots that overall rankings cannot anticipate, with reliability patterns differing markedly across model families.
展开完整摘要收起摘要↓
Maintaining physical consistency in video generators and world models increasingly relies on vision-language models (VLMs) as automated judges that provide reward signals, ranking decisions, and data-filtering criteria. Yet VLMs differ substantially in training data and architecture, encoding physical phenomena through distinct internal representations. A single global evaluation schema therefore gives every VLM the same axes of competence, regardless of what each can actually perceive. We propose JudgeFit, an iterative refinement procedure that discovers a per-VLM evaluation taxonomy. An initial taxonomy is constructed by prompting the target VLM to enumerate physics errors on a small set of videos and clustering the resulting descriptions. The taxonomy is then refined through a diagnostic step: we calibrate the VLM's per-dimension scores to human physical-commonsense ratings, diagnose which dimensions it scores unreliably or redundantly, and prompt an LLM to repair them, iterating until convergence. We further instantiate this procedure as a benchmark and apply it to 16 VLMs spanning eight model families. The refined taxonomy outperforms the global-schema baseline on held-out videos for every VLM tested, with a mean relative improvement of approximately 32%. Beyond aggregate accuracy, the per-VLM profiles expose model-specific blind spots that overall rankings cannot anticipate, with reliability patterns differing markedly across model families.
Automated Reward Design (ARD) aims to replace manual reward engineering in reinforcement learning with language-driven reward function synthesis. However, existing approaches based on large language models (LLMs) remain inherently correlation-driven, relying on iterative environmental feedback to refine reward hypotheses for each specific task. This paradigm not only results in inefficient reasoning but also makes LLMs susceptible to semantically plausible yet causally spurious reward components, leading to ineffective optimization. To address these limitations, we propose the Causal Reward World Model (CRWM), which explicitly models the causal topological relationships between candidate reward components and task-targeted physical variables through offline pre-training on multi-task interaction data. Based on a coarse-to-fine pre-training strategy, we introduce a joint optimization module that integrates Explicit Mechanism Decoupling with Confidence-Aware Soft Fusion to refine coarse structural priors using micro-level trajectories, thereby constructing a robust and interpretable causal skeleton. During inference, LLMs leverage CRWM as a task-irrelevant causal prior to constrain the reward generation, enabling zero-shot reward function design. Our work opens up a new white-box paradigm for the ARD problem. Extensive experiments on complex continuous control benchmarks demonstrate that CRWM generates executable reward functions without feedback-driven reward refinement, significantly reducing the design latency for acquiring new robotic skills while matching or surpassing state-of-the-art performance, and further exhibits strong generalization capabilities across unseen tasks and diverse robotic embodiments.
展开完整摘要收起摘要↓
Automated Reward Design (ARD) aims to replace manual reward engineering in reinforcement learning with language-driven reward function synthesis. However, existing approaches based on large language models (LLMs) remain inherently correlation-driven, relying on iterative environmental feedback to refine reward hypotheses for each specific task. This paradigm not only results in inefficient reasoning but also makes LLMs susceptible to semantically plausible yet causally spurious reward components, leading to ineffective optimization. To address these limitations, we propose the Causal Reward World Model (CRWM), which explicitly models the causal topological relationships between candidate reward components and task-targeted physical variables through offline pre-training on multi-task interaction data. Based on a coarse-to-fine pre-training strategy, we introduce a joint optimization module that integrates Explicit Mechanism Decoupling with Confidence-Aware Soft Fusion to refine coarse structural priors using micro-level trajectories, thereby constructing a robust and interpretable causal skeleton. During inference, LLMs leverage CRWM as a task-irrelevant causal prior to constrain the reward generation, enabling zero-shot reward function design. Our work opens up a new white-box paradigm for the ARD problem. Extensive experiments on complex continuous control benchmarks demonstrate that CRWM generates executable reward functions without feedback-driven reward refinement, significantly reducing the design latency for acquiring new robotic skills while matching or surpassing state-of-the-art performance, and further exhibits strong generalization capabilities across unseen tasks and diverse robotic embodiments.
作者Hongkai Zheng, Ta-Ying Cheng, Benjamin Klein, Yisong Yue, Zhuoning Yuan
Video diffusion models have enabled remarkable progress in video generation and editing. However, content preservation remains a core challenge: existing methods regenerate every pixel and often alter elements that should remain unchanged, such as characters or background scenes. We introduce Vera, a layered diffusion framework for content-preserving video editing. Instead of regenerating the entire video, Vera generates an edit layer along with an alpha matte for compositing with the source video, separating creative editing from content preservation by design. To encourage coherent composition with the source video, we extend the text-to-video DiT into a Mixture-of-Transformers (MoT) architecture, with separate DiTs for each layer that interact through joint self-attention. To support the training of Vera, we further construct a high-quality layered dataset with accurate alpha mattes, diverse scenes and dynamics, and visual effects. Across our quantitative benchmark and human preference study, Vera outperforms leading open-source video editing models in content preservation while remaining competitive in edit quality, using 486K frames of layered training data.
展开完整摘要收起摘要↓
Video diffusion models have enabled remarkable progress in video generation and editing. However, content preservation remains a core challenge: existing methods regenerate every pixel and often alter elements that should remain unchanged, such as characters or background scenes. We introduce Vera, a layered diffusion framework for content-preserving video editing. Instead of regenerating the entire video, Vera generates an edit layer along with an alpha matte for compositing with the source video, separating creative editing from content preservation by design. To encourage coherent composition with the source video, we extend the text-to-video DiT into a Mixture-of-Transformers (MoT) architecture, with separate DiTs for each layer that interact through joint self-attention. To support the training of Vera, we further construct a high-quality layered dataset with accurate alpha mattes, diverse scenes and dynamics, and visual effects. Across our quantitative benchmark and human preference study, Vera outperforms leading open-source video editing models in content preservation while remaining competitive in edit quality, using 486K frames of layered training data.
Developing generalist embodied agents requires interactive environments providing visually realistic feedback and accurate action-conditioned dynamics. Interactive world models address this by simulating such complex dynamics. However, purely data-driven methods struggle to ensure precise control alignment and physically plausible visual feedback due to a lack of explicit structural constraints. To address this, we propose IOI, a hybrid interactive world model integrating analytical kinematic priors with learned physical dynamics. Unlike data-driven approaches prone to spatiotemporal drift, IOI introduces explicit kinematic guidance, computing forward kinematics from action sequences for accurate motion trajectories. These trajectories are rendered into synchronized front, side, and top orthographic projections, eliminating the need for extrinsic camera calibration. A Multi-view Kinematic Aggregation and Injection module fuses these geometric cues and injects them into the video generator, providing geometry-consistent guidance. Conditioning video generation on these deterministic trajectories establishes a synergy between the analytical simulator and the world model. Decoupling deterministic motion into the kinematic prior frees the generator to model stochastic physical interactions. Experiments on the RoboTwin benchmark validate IOI across kinematic fidelity, out-of-distribution (OOD) generalization, and policy evaluation. IOI achieves state-of-the-art simulation performance and robust zero-shot generalization to unseen OOD tasks. Furthermore, IOI serves as a reliable policy evaluator, yielding success rates closely aligning with ground-truth physics simulators. On real-world platforms, policies trained on IOI-synthesized data match those trained on teleoperation demonstrations, solidifying its practical value for embodied policy learning.
展开完整摘要收起摘要↓
Developing generalist embodied agents requires interactive environments providing visually realistic feedback and accurate action-conditioned dynamics. Interactive world models address this by simulating such complex dynamics. However, purely data-driven methods struggle to ensure precise control alignment and physically plausible visual feedback due to a lack of explicit structural constraints. To address this, we propose IOI, a hybrid interactive world model integrating analytical kinematic priors with learned physical dynamics. Unlike data-driven approaches prone to spatiotemporal drift, IOI introduces explicit kinematic guidance, computing forward kinematics from action sequences for accurate motion trajectories. These trajectories are rendered into synchronized front, side, and top orthographic projections, eliminating the need for extrinsic camera calibration. A Multi-view Kinematic Aggregation and Injection module fuses these geometric cues and injects them into the video generator, providing geometry-consistent guidance. Conditioning video generation on these deterministic trajectories establishes a synergy between the analytical simulator and the world model. Decoupling deterministic motion into the kinematic prior frees the generator to model stochastic physical interactions. Experiments on the RoboTwin benchmark validate IOI across kinematic fidelity, out-of-distribution (OOD) generalization, and policy evaluation. IOI achieves state-of-the-art simulation performance and robust zero-shot generalization to unseen OOD tasks. Furthermore, IOI serves as a reliable policy evaluator, yielding success rates closely aligning with ground-truth physics simulators. On real-world platforms, policies trained on IOI-synthesized data match those trained on teleoperation demonstrations, solidifying its practical value for embodied policy learning.
Recent diffusion-based models have enabled realistic audio-driven avatar generation in real-time streaming. However, existing approaches struggle to maintain visual temporal consistency and fail to explicitly perceive user intent in complex interactive streaming scenarios. To address these challenges, we propose InteractiveAvatar, a real-time infinite-streaming video generation framework that supports visually consistent avatar video generation and intent-aware interactions. With autoregressive distillation, InteractiveAvatar achieves real-time str-eaming generation of human avatars over arbitrarily long durations. For visual consistency, we introduce a Long-Short Visual Memory (LSVM) mechanism that flexibly compresses historical visual information into compact tokens, preserving both short-range coherence and long-term consistency. To generate avatars with speeches and actions aligned with user intent, we propose a Reasoning-Reaction Module (RRM), which incorporates a State-Cycling strategy and a Cache-Switching mechanism. Extensive experimental results over diverse scenarios demonstrate that our method achieves state-of-the-art visual consistency in long-duration generation, while enabling complex user-avatar interaction in real time.
展开完整摘要收起摘要↓
Recent diffusion-based models have enabled realistic audio-driven avatar generation in real-time streaming. However, existing approaches struggle to maintain visual temporal consistency and fail to explicitly perceive user intent in complex interactive streaming scenarios. To address these challenges, we propose InteractiveAvatar, a real-time infinite-streaming video generation framework that supports visually consistent avatar video generation and intent-aware interactions. With autoregressive distillation, InteractiveAvatar achieves real-time str-eaming generation of human avatars over arbitrarily long durations. For visual consistency, we introduce a Long-Short Visual Memory (LSVM) mechanism that flexibly compresses historical visual information into compact tokens, preserving both short-range coherence and long-term consistency. To generate avatars with speeches and actions aligned with user intent, we propose a Reasoning-Reaction Module (RRM), which incorporates a State-Cycling strategy and a Cache-Switching mechanism. Extensive experimental results over diverse scenarios demonstrate that our method achieves state-of-the-art visual consistency in long-duration generation, while enabling complex user-avatar interaction in real time.
Recent years have witnessed remarkable progress in image generation and editing, particularly regarding instruction following and visual fidelity. However, when handling ambiguous intentions, logical reasoning, and Out-of-Distribution (OOD) knowledge, existing image models often yield sub-optimal results due to a lack of deep reasoning capabilities and real-time external information. Although emerging unified understanding-and-generation models attempt to bridge this gap, they remain constrained by their intrinsic parameter scales and static knowledge gaps. Inspired by agentic paradigms, we propose RS-Gen: a plug-and-play, training-free, multi-stage image agentic framework. RS-Gen innovatively introduces a "Questioning-and-Solving" closed-loop mechanism to accurately identify logical issues and knowledge gaps, autonomously planning actions to bridge information deficits and execute deep logical reasoning. Extensive experiments demonstrate that RS-Gen significantly expands the capability boundaries of foundational image generation and editing models. Specifically, on the WISE Verified and RISEBench benchmarks, RS-Gen yields substantial absolute performance gains of 0.313 for Qwen-Image and 19.70 for Qwen-Image-Edit-2511, respectively, successfully elevating both to the state-of-the-art (SOTA) level among open-source models.
展开完整摘要收起摘要↓
Recent years have witnessed remarkable progress in image generation and editing, particularly regarding instruction following and visual fidelity. However, when handling ambiguous intentions, logical reasoning, and Out-of-Distribution (OOD) knowledge, existing image models often yield sub-optimal results due to a lack of deep reasoning capabilities and real-time external information. Although emerging unified understanding-and-generation models attempt to bridge this gap, they remain constrained by their intrinsic parameter scales and static knowledge gaps. Inspired by agentic paradigms, we propose RS-Gen: a plug-and-play, training-free, multi-stage image agentic framework. RS-Gen innovatively introduces a "Questioning-and-Solving" closed-loop mechanism to accurately identify logical issues and knowledge gaps, autonomously planning actions to bridge information deficits and execute deep logical reasoning. Extensive experiments demonstrate that RS-Gen significantly expands the capability boundaries of foundational image generation and editing models. Specifically, on the WISE Verified and RISEBench benchmarks, RS-Gen yields substantial absolute performance gains of 0.313 for Qwen-Image and 19.70 for Qwen-Image-Edit-2511, respectively, successfully elevating both to the state-of-the-art (SOTA) level among open-source models.
作者Sara Dorfman, Maya Vishnevsky, Omer Dahary, Or Patashnik, Daniel Cohen-Or
Modern text-to-image models excel in visual fidelity and prompt adherence. However, this strict adherence comes at the cost of diversity: generated samples tend to collapse into a single visual interpretation. Existing methods to improve diversity produce outputs driven by incidental variations rather than meaningful design choices. This motivates a new variant of the diversity task where structure is enforced on the generated samples. We introduce a method for controlled diversity that enables Semantic Browsing, where users can navigate structured image galleries and experience creative exploration through a systematic traversal of meaningful, interpretable axes of variation. Achieving this level of semantic control requires a deep understanding of the scene. We exploit the fact that recent text-to-image models are trained on elaborated captions, effectively decoupling semantic decision-making from pixel generation. This enables a paradigm shift: instead of relying on stochastic variation within the text-to-image model, we induce diversity directly at the text level. By leveraging rich textual representations, we allow a Vision Language Model (VLM) to operate on the full scene context. To overcome the generic outputs typical of standard VLMs, we employ an agentic workflow that explicitly enforces structured variation attuned to the original prompt. We demonstrate that our method produces diverse and navigable design spaces where every variation corresponds to a specific, user-understandable semantic decision.
展开完整摘要收起摘要↓
Modern text-to-image models excel in visual fidelity and prompt adherence. However, this strict adherence comes at the cost of diversity: generated samples tend to collapse into a single visual interpretation. Existing methods to improve diversity produce outputs driven by incidental variations rather than meaningful design choices. This motivates a new variant of the diversity task where structure is enforced on the generated samples. We introduce a method for controlled diversity that enables Semantic Browsing, where users can navigate structured image galleries and experience creative exploration through a systematic traversal of meaningful, interpretable axes of variation. Achieving this level of semantic control requires a deep understanding of the scene. We exploit the fact that recent text-to-image models are trained on elaborated captions, effectively decoupling semantic decision-making from pixel generation. This enables a paradigm shift: instead of relying on stochastic variation within the text-to-image model, we induce diversity directly at the text level. By leveraging rich textual representations, we allow a Vision Language Model (VLM) to operate on the full scene context. To overcome the generic outputs typical of standard VLMs, we employ an agentic workflow that explicitly enforces structured variation attuned to the original prompt. We demonstrate that our method produces diverse and navigable design spaces where every variation corresponds to a specific, user-understandable semantic decision.
Modern video diffusion models achieve higher generation quality through scaling, but this also increases inference cost. Although many acceleration methods have been proposed, a central challenge is that the most effective acceleration strategy is highly instance-specific: a recipe that works well for one combination of model, hardware, and inference configuration often does not transfer to another. Different models vary in architecture, numerical sensitivity, and attention concentration patterns. Inference settings differ in spatial and temporal resolution and video duration, while hardware platforms differ in memory hierarchy, supported numerical formats, and kernel throughput. These factors create a large tuning space, making manual performance engineering costly. We present Sol Video Inference Engine, an agentic, native, training-free acceleration framework for video diffusion models. It organizes five broadly applicable techniques, cache, sparse attention, token pruning, quantization, and kernel fusion, into an agentic acceleration stack for instance-specific optimization. For a concrete deployment target defined by a model, hardware platform, and serving configuration, parallel skill agents optimize the implementation of each technique, an agent integrator composes them into a global acceleration stack, and a human validator provides feedback on generation quality. We instantiate this workflow on three video models with different sizes and architectures: 64B Cosmos3-Super, 22B LTX-2.3, and 2B SANA-Video. With little human effort, the full stack achieves more than 2x end-to-end acceleration while maintaining near-lossless VBench quality, demonstrating the effectiveness of the agent framework for video diffusion acceleration.
展开完整摘要收起摘要↓
Modern video diffusion models achieve higher generation quality through scaling, but this also increases inference cost. Although many acceleration methods have been proposed, a central challenge is that the most effective acceleration strategy is highly instance-specific: a recipe that works well for one combination of model, hardware, and inference configuration often does not transfer to another. Different models vary in architecture, numerical sensitivity, and attention concentration patterns. Inference settings differ in spatial and temporal resolution and video duration, while hardware platforms differ in memory hierarchy, supported numerical formats, and kernel throughput. These factors create a large tuning space, making manual performance engineering costly. We present Sol Video Inference Engine, an agentic, native, training-free acceleration framework for video diffusion models. It organizes five broadly applicable techniques, cache, sparse attention, token pruning, quantization, and kernel fusion, into an agentic acceleration stack for instance-specific optimization. For a concrete deployment target defined by a model, hardware platform, and serving configuration, parallel skill agents optimize the implementation of each technique, an agent integrator composes them into a global acceleration stack, and a human validator provides feedback on generation quality. We instantiate this workflow on three video models with different sizes and architectures: 64B Cosmos3-Super, 22B LTX-2.3, and 2B SANA-Video. With little human effort, the full stack achieves more than 2x end-to-end acceleration while maintaining near-lossless VBench quality, demonstrating the effectiveness of the agent framework for video diffusion acceleration.
Training image generation foundation models consumes substantial resources. Previous methods have attempted to leverage semantic guidance to accelerate the training process, yet their experiments were only conducted on simple datasets such as ImageNet, at low resolutions, and with small-scale models. In this paper, we propose SeFi-Image, a text-to-image foundation model built upon semantic-first diffusion, a novel latent diffusion modeling paradigm. We instantiate SeFi-Image at three model scales, 1B, 2B, and 5B parameters, enabling systematic study of scaling behavior and flexible deployment under varying compute budgets. Notably, our largest 5B model was trained with merely 125K A800 GPU hours, corresponding to roughly 10-20% of the training compute used by Z-Image. However, it achieves results comparable to or even superior to Qwen-Image and Z-Image. Despite this modest training compute, SeFi-Image achieves strong performance on a wide range of benchmarks, including GenEval, DPG, LongTextBench, OneIG, and CVTG-2K. Moreover, we provide DMD2-distilled few-step turbo variants for each model scale to accommodate diverse hardware constraints and latency requirements. We publicly release our code, weights and hope this work offers the community useful insights into semantic-guided diffusion modeling for T2I generation, while also providing practical and readily deployable model options.
展开完整摘要收起摘要↓
Training image generation foundation models consumes substantial resources. Previous methods have attempted to leverage semantic guidance to accelerate the training process, yet their experiments were only conducted on simple datasets such as ImageNet, at low resolutions, and with small-scale models. In this paper, we propose SeFi-Image, a text-to-image foundation model built upon semantic-first diffusion, a novel latent diffusion modeling paradigm. We instantiate SeFi-Image at three model scales, 1B, 2B, and 5B parameters, enabling systematic study of scaling behavior and flexible deployment under varying compute budgets. Notably, our largest 5B model was trained with merely 125K A800 GPU hours, corresponding to roughly 10-20% of the training compute used by Z-Image. However, it achieves results comparable to or even superior to Qwen-Image and Z-Image. Despite this modest training compute, SeFi-Image achieves strong performance on a wide range of benchmarks, including GenEval, DPG, LongTextBench, OneIG, and CVTG-2K. Moreover, we provide DMD2-distilled few-step turbo variants for each model scale to accommodate diverse hardware constraints and latency requirements. We publicly release our code, weights and hope this work offers the community useful insights into semantic-guided diffusion modeling for T2I generation, while also providing practical and readily deployable model options.
Although Large Language Models (LLMs) excel in many tasks, their assessment in Portuguese has received less attention, particularly for open-ended, discursive tasks that demand deeper reasoning and generation capabilities. While the original BLUEX benchmark addressed the scarcity of Portuguese evaluation datasets through multiple-choice questions from Brazilian university entrance exams, it did not cover the more challenging second-phase examinations, which require free-form written responses. In this work, we introduce BLUEX v2, a benchmark derived from the second-phase entrance exams of Brazil's two leading universities: UNICAMP (Comvest) and USP (Fuvest), spanning exam years 2022-2025. Our dataset comprises 395 questions unfolding into 919 graded subquestions, with 55.7% of questions containing associated images. Each question is annotated with subject area, official reference answers, LLM-generated rubric criteria, and six cognitive capability tags. We evaluate 21 state-of-the-art LLMs using an LLM-as-a-judge protocol. Results reveal a 4.92-point performance spread across models (4.18-9.10 on a 0-10 scale), with Mathematical Reasoning and Image Understanding emerging as the hardest capability dimensions. The dataset, evaluation code, and model outputs are publicly available at https://anonymous.4open.science/r/BLUEXv2.
展开完整摘要收起摘要↓
Although Large Language Models (LLMs) excel in many tasks, their assessment in Portuguese has received less attention, particularly for open-ended, discursive tasks that demand deeper reasoning and generation capabilities. While the original BLUEX benchmark addressed the scarcity of Portuguese evaluation datasets through multiple-choice questions from Brazilian university entrance exams, it did not cover the more challenging second-phase examinations, which require free-form written responses. In this work, we introduce BLUEX v2, a benchmark derived from the second-phase entrance exams of Brazil's two leading universities: UNICAMP (Comvest) and USP (Fuvest), spanning exam years 2022-2025. Our dataset comprises 395 questions unfolding into 919 graded subquestions, with 55.7% of questions containing associated images. Each question is annotated with subject area, official reference answers, LLM-generated rubric criteria, and six cognitive capability tags. We evaluate 21 state-of-the-art LLMs using an LLM-as-a-judge protocol. Results reveal a 4.92-point performance spread across models (4.18-9.10 on a 0-10 scale), with Mathematical Reasoning and Image Understanding emerging as the hardest capability dimensions. The dataset, evaluation code, and model outputs are publicly available at https://anonymous.4open.science/r/BLUEXv2.
Autoregressive visual models (AVMs) based on next-scale prediction have emerged as a prominent paradigm for image and video synthesis. However, decomposing the generation process into discrete scales with varying granularities in AVM makes semantic errors difficult to identify and correct, thereby undermining the quality of the final output. Prior efforts to enhance AVM can be categorized into training-based and training-free approaches. Although training-based efforts to enhance AVM generation quality come at substantial computational cost, existing training-free methods neglect intermediate generation states, leaving semantic errors undiagnosed and allowing them to accumulate into the final output. In this paper, we focus on training-free paradigms and propose Gazer, a framework that integrates multimodal large language model feedback into the AVM sampling loop for in-generation semantic correction. Concretely, Gazer operates via two cooperating stages: the Reflective Diagnosis stage diagnoses semantic errors from intermediate states, while the Semantic Correction stage rewinds and rectifies the generation trajectory to realign with the target prompt. Experiments on compositional image and video benchmarks demonstrate that Gazer improves semantic alignment and compositional accuracy across multiple AVMs without additional training.
展开完整摘要收起摘要↓
Autoregressive visual models (AVMs) based on next-scale prediction have emerged as a prominent paradigm for image and video synthesis. However, decomposing the generation process into discrete scales with varying granularities in AVM makes semantic errors difficult to identify and correct, thereby undermining the quality of the final output. Prior efforts to enhance AVM can be categorized into training-based and training-free approaches. Although training-based efforts to enhance AVM generation quality come at substantial computational cost, existing training-free methods neglect intermediate generation states, leaving semantic errors undiagnosed and allowing them to accumulate into the final output. In this paper, we focus on training-free paradigms and propose Gazer, a framework that integrates multimodal large language model feedback into the AVM sampling loop for in-generation semantic correction. Concretely, Gazer operates via two cooperating stages: the Reflective Diagnosis stage diagnoses semantic errors from intermediate states, while the Semantic Correction stage rewinds and rectifies the generation trajectory to realign with the target prompt. Experiments on compositional image and video benchmarks demonstrate that Gazer improves semantic alignment and compositional accuracy across multiple AVMs without additional training.
作者Kunwar Maheep Singh, Christian Theobalt, Rishabh Dabral
We present Generative Relightable Avatars (GRA), a person-specific method for photorealistic free-view rendering and environment-map relighting of full-body humans. We postulate that modeling fine-grained appearance details is inherently a one-to-many problem that can benefit from a generative formulation. In contrast to fully regressive relightable avatar methods, GRA follows a hybrid approach that combines controllable, physics-grounded relighting with probabilistic refinement. Starting from a tracked animated mesh, we optimize material parameters in UV-space and render a coarse relit appearance under a target HDR environment map. Next, we refine the textures with a feed-forward model to capture pose-dependent texture dynamics and illumination effects beyond simplified reflectance assumptions. Finally, a fine-tuned video-to-video diffusion model transforms the physically grounded renderings into temporally coherent, high-detail videos while preserving 3D control, with an error-recycling strategy for generating long videos. Experimental evaluations demonstrate our method's improved perceptual quality over prior relightable avatar baselines. Project Page: https://vcai.mpi-inf.mpg.de/projects/GRA/
展开完整摘要收起摘要↓
We present Generative Relightable Avatars (GRA), a person-specific method for photorealistic free-view rendering and environment-map relighting of full-body humans. We postulate that modeling fine-grained appearance details is inherently a one-to-many problem that can benefit from a generative formulation. In contrast to fully regressive relightable avatar methods, GRA follows a hybrid approach that combines controllable, physics-grounded relighting with probabilistic refinement. Starting from a tracked animated mesh, we optimize material parameters in UV-space and render a coarse relit appearance under a target HDR environment map. Next, we refine the textures with a feed-forward model to capture pose-dependent texture dynamics and illumination effects beyond simplified reflectance assumptions. Finally, a fine-tuned video-to-video diffusion model transforms the physically grounded renderings into temporally coherent, high-detail videos while preserving 3D control, with an error-recycling strategy for generating long videos. Experimental evaluations demonstrate our method's improved perceptual quality over prior relightable avatar baselines. Project Page: https://vcai.mpi-inf.mpg.de/projects/GRA/
Deep generative models reproduce the observational distribution of their training data, inheriting any spurious associations it contains. A common source is an unobserved confounder that shapes both an attribute the user wants to control at sampling time and an attribute expected to vary in response. Existing causal generative approaches resolve the resulting ambiguity by imposing structural assumptions strong enough to single out one interventional distribution; in image domains, such assumptions are rarely warranted, and the data is generally consistent with a set of distinct causal mechanisms — a feasible region of interventional distributions. We propose CauVaDE (Causal Variational Deep Embedding), built on a canonical augmented SCM in which the unobserved confounder collapses, without loss of generality, into a discrete latent cluster of bounded support while continuous variation is absorbed into independent noises. We prove that this canonical class is dense, in both observational and interventional Wasserstein distance, in the class of augmented SCMs compatible with a given causal diagram, and instantiate it as a mixture variational autoencoder whose cluster variable plays the role of the canonical confounder. An entropy regularizer with weight $γ$ on the cluster posterior then traces a family of candidate causal effects that fit the observational data to comparable likelihood while spanning the feasible region. Experiments on image data benchmarks show that CauVaDE produces diverse interventional samples and improves FID against an unconfounded reference.
展开完整摘要收起摘要↓
Deep generative models reproduce the observational distribution of their training data, inheriting any spurious associations it contains. A common source is an unobserved confounder that shapes both an attribute the user wants to control at sampling time and an attribute expected to vary in response. Existing causal generative approaches resolve the resulting ambiguity by imposing structural assumptions strong enough to single out one interventional distribution; in image domains, such assumptions are rarely warranted, and the data is generally consistent with a set of distinct causal mechanisms — a feasible region of interventional distributions. We propose CauVaDE (Causal Variational Deep Embedding), built on a canonical augmented SCM in which the unobserved confounder collapses, without loss of generality, into a discrete latent cluster of bounded support while continuous variation is absorbed into independent noises. We prove that this canonical class is dense, in both observational and interventional Wasserstein distance, in the class of augmented SCMs compatible with a given causal diagram, and instantiate it as a mixture variational autoencoder whose cluster variable plays the role of the canonical confounder. An entropy regularizer with weight $γ$ on the cluster posterior then traces a family of candidate causal effects that fit the observational data to comparable likelihood while spanning the feasible region. Experiments on image data benchmarks show that CauVaDE produces diverse interventional samples and improves FID against an unconfounded reference.
作者Cong Wan, Zeyu Guo, Zijian Cai, Jiangyang Li, SongLin Dong, Lin Peng, Xiangyang Luo, Zhiheng Ma, Yihong Gong
Massive unstructured multimodal streams suffer from high "data entropy," impeding both efficient human knowledge acquisition and high-quality AI post-training. Existing passive annotation paradigms, heavily reliant on heuristic rules or general VLMs, are costly, monotonous, and fail to unlock the deep procedural logic embedded in raw data. We elevate data processing to a learnable capability, proposing a paradigm shift towards Agentic Data Tailoring, which actively refining and structuring data to align with diverse user and downstream intents. To overcome the data scarcity bottleneck in training such high-order capabilities, we design a two-stage pipeline grounding generative semantic synthesis in deterministic Factual Anchors, yielding a large-scale dataset spanning five core physical and digital domains. Building upon this, $\text{DataClaw}_0$-9B model synergizes Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO), achieving robust alignment with complex refinement and tailoring intents. To systematically quantify this capability, we construct $\text{DataClaw}_0$-val, the first benchmark dedicated to data refinement. Crucially, we adopt downstream post-training as the ultimate validation touchstone. Evaluations on video generation, real-world VQA, and GUI navigation confirm that $\text{DataClaw}_0$ delivers high-information-density tailored data, facilitating efficient model adaptation to new tasks under limited training data regimes. Project page: https://czjdsg.github.io/MakeAnyData
展开完整摘要收起摘要↓
Massive unstructured multimodal streams suffer from high "data entropy," impeding both efficient human knowledge acquisition and high-quality AI post-training. Existing passive annotation paradigms, heavily reliant on heuristic rules or general VLMs, are costly, monotonous, and fail to unlock the deep procedural logic embedded in raw data. We elevate data processing to a learnable capability, proposing a paradigm shift towards Agentic Data Tailoring, which actively refining and structuring data to align with diverse user and downstream intents. To overcome the data scarcity bottleneck in training such high-order capabilities, we design a two-stage pipeline grounding generative semantic synthesis in deterministic Factual Anchors, yielding a large-scale dataset spanning five core physical and digital domains. Building upon this, $\text{DataClaw}_0$-9B model synergizes Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO), achieving robust alignment with complex refinement and tailoring intents. To systematically quantify this capability, we construct $\text{DataClaw}_0$-val, the first benchmark dedicated to data refinement. Crucially, we adopt downstream post-training as the ultimate validation touchstone. Evaluations on video generation, real-world VQA, and GUI navigation confirm that $\text{DataClaw}_0$ delivers high-information-density tailored data, facilitating efficient model adaptation to new tasks under limited training data regimes. Project page: https://czjdsg.github.io/MakeAnyData
Generating a coherent multi-shot video requires structured cross-shot memory. Subject appearance, scene context, and speaker identity must persist across cuts. Existing approaches either train end-to-end over fixed-length sequences and cannot scale, generate shot-by-shot with memory banks that grow linearly, or orchestrate pretrained generators under an LLM planner without a multi-shot-aware backbone. We present UnityShots, a memory-driven multi-shot audio-video generation system built on LTX-2.3, trained on annotated cinematic and music-video shots. The video stream maintains two fixed-size slots, a long-term memory (LTM) slot anchored to the opening shot and a short-term memory (STM) slot holding the immediately preceding tail, both updated at every cut by a boundary-conditioned gate that fuses visual cut probability and beat-tracker signals. The audio stream injects a reference speaker token at every shot to preserve vocal timbre without a sliding audio bank. A discrete cut-type prior, learned through AdaLN, becomes an inference-time control knob over transition strength. We release a benchmark of $200$ multi-cultural multi-shot sequences spanning six ethnic regions and ten or more languages, with per-shot reference identities, reference audio, and per-boundary transition labels. Evaluated across I2V, T2V, and R2V conditioning modes, UnityShots leads open-source baselines on every cross-shot coherence metric and matches the strongest closed-source system on the multi-shot axes.
展开完整摘要收起摘要↓
Generating a coherent multi-shot video requires structured cross-shot memory. Subject appearance, scene context, and speaker identity must persist across cuts. Existing approaches either train end-to-end over fixed-length sequences and cannot scale, generate shot-by-shot with memory banks that grow linearly, or orchestrate pretrained generators under an LLM planner without a multi-shot-aware backbone. We present UnityShots, a memory-driven multi-shot audio-video generation system built on LTX-2.3, trained on annotated cinematic and music-video shots. The video stream maintains two fixed-size slots, a long-term memory (LTM) slot anchored to the opening shot and a short-term memory (STM) slot holding the immediately preceding tail, both updated at every cut by a boundary-conditioned gate that fuses visual cut probability and beat-tracker signals. The audio stream injects a reference speaker token at every shot to preserve vocal timbre without a sliding audio bank. A discrete cut-type prior, learned through AdaLN, becomes an inference-time control knob over transition strength. We release a benchmark of $200$ multi-cultural multi-shot sequences spanning six ethnic regions and ten or more languages, with per-shot reference identities, reference audio, and per-boundary transition labels. Evaluated across I2V, T2V, and R2V conditioning modes, UnityShots leads open-source baselines on every cross-shot coherence metric and matches the strongest closed-source system on the multi-shot axes.
作者Tianyou Wang, Anson Lei, Joe Watson, Ingmar Posner
Imitation learning has emerged as a powerful paradigm for learning visuomotor policies, but its generalisation and stability are limited by the scale and quality of demonstration data needed. A promising direction is to leverage more abundant but heterogeneous data sources, which differ in action space and often lack action labels altogether. Existing co-training approaches that combine heterogeneous data sources rely on heuristic and hand-engineered alignment techniques. In contrast, we argue that action representations should be grounded in prediction: actions that produce the same effect on the environment should share the same representation, regardless of their sources. To this end, we instantiate this principle by using a grounded latent-action world model (GLAM), a pair of generative models with a shared latent action space across data sources that is grounded by predicting future observations consistently across sources. This latent action space is used to train downstream behavioural cloning (BC) policies which map observations to latent actions and decode them back to robot actions, providing a paradigm for learning from heterogeneous data. Empirically, we demonstrate that GLAM successfully learns an aligned latent action space that facilitates action transfer across data sources with and without action labels. Across five manipulation tasks in simulation and in the real world, GLAM-aligned policies significantly outperform BC baselines and prior latent-action methods, achieving an average of +48% improvement in task success rate with the same data-scarce setting. Videos and code are available at https://viccccciv.github.io/glam/.
展开完整摘要收起摘要↓
Imitation learning has emerged as a powerful paradigm for learning visuomotor policies, but its generalisation and stability are limited by the scale and quality of demonstration data needed. A promising direction is to leverage more abundant but heterogeneous data sources, which differ in action space and often lack action labels altogether. Existing co-training approaches that combine heterogeneous data sources rely on heuristic and hand-engineered alignment techniques. In contrast, we argue that action representations should be grounded in prediction: actions that produce the same effect on the environment should share the same representation, regardless of their sources. To this end, we instantiate this principle by using a grounded latent-action world model (GLAM), a pair of generative models with a shared latent action space across data sources that is grounded by predicting future observations consistently across sources. This latent action space is used to train downstream behavioural cloning (BC) policies which map observations to latent actions and decode them back to robot actions, providing a paradigm for learning from heterogeneous data. Empirically, we demonstrate that GLAM successfully learns an aligned latent action space that facilitates action transfer across data sources with and without action labels. Across five manipulation tasks in simulation and in the real world, GLAM-aligned policies significantly outperform BC baselines and prior latent-action methods, achieving an average of +48% improvement in task success rate with the same data-scarce setting. Videos and code are available at https://viccccciv.github.io/glam/.
Recently, world models have emerged as a promising paradigm for building intelligent agents by learning predictive models that estimate future environment states conditioned on observations and actions. In particular, JEPA-style latent world models provide an efficient alternative to pixel space prediction by learning action-conditioned dynamics in compact representation spaces. However, existing latent world models typically rely on one-step prediction and must be recursively rolled out for long-horizon planning, which leads to compounding errors and a mismatch between training objectives and downstream planning tasks. To address this limitation, we propose Variable-length Latent World Models (VLWMs), a framework that learns to predict future latent states conditioned on action sequences of variable lengths. Instead of training only on one-step transitions, VLWMs directly model temporally extended dynamics, allowing the same predictor to evaluate action plans over different horizons. We further introduce a curriculum training strategy that progressively expands the action horizon, stabilizing optimization from short-range dynamics to long-range prediction. At test time, we design planning methods tailored to VLWMs to better exploit their variable-length predictive capabilities. Experiments on long-horizon control tasks show that VLWMs significantly improve latent space world models, achieving 13% average improvement over the state-of-the-art LeWM across different datasets, with especially large gains on tasks requiring extended planning. These results suggest that VLWM provides a simple yet effective paradigm for improving long-horizon prediction and planning in latent world models.
展开完整摘要收起摘要↓
Recently, world models have emerged as a promising paradigm for building intelligent agents by learning predictive models that estimate future environment states conditioned on observations and actions. In particular, JEPA-style latent world models provide an efficient alternative to pixel space prediction by learning action-conditioned dynamics in compact representation spaces. However, existing latent world models typically rely on one-step prediction and must be recursively rolled out for long-horizon planning, which leads to compounding errors and a mismatch between training objectives and downstream planning tasks. To address this limitation, we propose Variable-length Latent World Models (VLWMs), a framework that learns to predict future latent states conditioned on action sequences of variable lengths. Instead of training only on one-step transitions, VLWMs directly model temporally extended dynamics, allowing the same predictor to evaluate action plans over different horizons. We further introduce a curriculum training strategy that progressively expands the action horizon, stabilizing optimization from short-range dynamics to long-range prediction. At test time, we design planning methods tailored to VLWMs to better exploit their variable-length predictive capabilities. Experiments on long-horizon control tasks show that VLWMs significantly improve latent space world models, achieving 13% average improvement over the state-of-the-art LeWM across different datasets, with especially large gains on tasks requiring extended planning. These results suggest that VLWM provides a simple yet effective paradigm for improving long-horizon prediction and planning in latent world models.
作者Yuanhao Chiang, Hongbo Duan, Chunru Yang, Jiahua Pei, Yi Liu, Xueqian Wang
Autoregressive text-to-image (T2I) generation has recently advanced rapidly, yet aligning generated images with human preferences remains challenging. GRPO-style online reinforcement learning provides an effective framework; however, existing methods typically treat reference-policy divergence as fixed, despite its direct impact on policy optimization. We study this overlooked factor within a unified f-divergence framework, encompassing forward KL, reverse KL, and JS divergence, for GRPO-style autoregressive T2I alignment. Our systematic theoretical analysis reveals that different divergences reshape token-level updates in distinct ways. In particular, under the sampled-token shaping form used, JS regularization achieves a favorable trade-off by mitigating uniform bias relative to the reference policy while still discouraging large deviations. Extensive experiments on LlamaGen and Janus-7B show that JS divergence achieves the strongest or highly competitive optimization performance on most evaluation metrics while maintaining favorable generation diversity. The code is available at https://github.com/tuoyou-hao/BPD-GRPO.
展开完整摘要收起摘要↓
Autoregressive text-to-image (T2I) generation has recently advanced rapidly, yet aligning generated images with human preferences remains challenging. GRPO-style online reinforcement learning provides an effective framework; however, existing methods typically treat reference-policy divergence as fixed, despite its direct impact on policy optimization. We study this overlooked factor within a unified f-divergence framework, encompassing forward KL, reverse KL, and JS divergence, for GRPO-style autoregressive T2I alignment. Our systematic theoretical analysis reveals that different divergences reshape token-level updates in distinct ways. In particular, under the sampled-token shaping form used, JS regularization achieves a favorable trade-off by mitigating uniform bias relative to the reference policy while still discouraging large deviations. Extensive experiments on LlamaGen and Janus-7B show that JS divergence achieves the strongest or highly competitive optimization performance on most evaluation metrics while maintaining favorable generation diversity. The code is available at https://github.com/tuoyou-hao/BPD-GRPO.
Single domain generalization (SDG) aims to learn a robust model, which could perform well on many unseen domains while there is only one single domain available for training. One of the promising directions for achieving single-domain generalization is to generate out-of-domain (OOD) training data through data augmentation or image generation. Given the rapid advancements in AI-generated content (AIGC), this paper is the first to propose leveraging powerful pre-trained text-to-image (T2I) foundation models to create the training data. However, manually designing textual prompts to generate images for all possible domains is often impractical, and some domain characteristics may be too abstract to describe with words. To address these challenges, we propose a novel Progressive Adversarial Prompt Tuning (PAPT) framework for pre-trained diffusion models. Instead of relying on static textual domains, our approach learns two sets of abstract prompts as conditions for the diffusion model: one that captures domain-invariant category information and another that models domain-specific styles. This adversarial learning mechanism enables the T2I model to generate images in various domain styles while preserving key categorical features. Extensive experiments demonstrate the effectiveness of the proposed method, achieving superior performances to state-of-the-art single-domain generalization approaches.
展开完整摘要收起摘要↓
Single domain generalization (SDG) aims to learn a robust model, which could perform well on many unseen domains while there is only one single domain available for training. One of the promising directions for achieving single-domain generalization is to generate out-of-domain (OOD) training data through data augmentation or image generation. Given the rapid advancements in AI-generated content (AIGC), this paper is the first to propose leveraging powerful pre-trained text-to-image (T2I) foundation models to create the training data. However, manually designing textual prompts to generate images for all possible domains is often impractical, and some domain characteristics may be too abstract to describe with words. To address these challenges, we propose a novel Progressive Adversarial Prompt Tuning (PAPT) framework for pre-trained diffusion models. Instead of relying on static textual domains, our approach learns two sets of abstract prompts as conditions for the diffusion model: one that captures domain-invariant category information and another that models domain-specific styles. This adversarial learning mechanism enables the T2I model to generate images in various domain styles while preserving key categorical features. Extensive experiments demonstrate the effectiveness of the proposed method, achieving superior performances to state-of-the-art single-domain generalization approaches.
作者Pradhaan S Bhat, Naveen Chandra R, Rishubh Parihar, Vaibhav Vavilala, R. Venkatesh Babu, D. A. Forsyth, Anand Bhattad
Text and 2D-conditioning interfaces provide weak, ambiguous control over spatial transformations in image editing — particularly under large object motions and camera changes. Prior work has used 3D primitives such as boxes, but only as loose conditioning signals indicating approximate object location rather than specifying the transformation. We instead use 3D boxes as structured specifications: the user provides the input and output boxes of the edit, casting editing as a well-posed geometry problem. This ``thinking in boxes'' interface, where each box face is color-coded to convey 3D orientation, gives precise control over translation, rotation, scaling, and viewpoint changes in real images while preserving scene and object identity, and recovering previously unseen object regions. To ground transformations in scene appearance, we introduce a depth-aligned planar floor as a global reference frame, shaded with depth-aware cues. Conditioned on this structure, an image generator produces consistent results under large transformations. Trained in two stages — on synthetic multi-object scenes and a small set of real-world videos from Objectron — the system generalizes to complex, in-the-wild real images. Our method operates directly on real photographs and substantially outperforms recent state-of-the-art methods on large 3D edits.
展开完整摘要收起摘要↓
Text and 2D-conditioning interfaces provide weak, ambiguous control over spatial transformations in image editing — particularly under large object motions and camera changes. Prior work has used 3D primitives such as boxes, but only as loose conditioning signals indicating approximate object location rather than specifying the transformation. We instead use 3D boxes as structured specifications: the user provides the input and output boxes of the edit, casting editing as a well-posed geometry problem. This ``thinking in boxes'' interface, where each box face is color-coded to convey 3D orientation, gives precise control over translation, rotation, scaling, and viewpoint changes in real images while preserving scene and object identity, and recovering previously unseen object regions. To ground transformations in scene appearance, we introduce a depth-aligned planar floor as a global reference frame, shaded with depth-aware cues. Conditioned on this structure, an image generator produces consistent results under large transformations. Trained in two stages — on synthetic multi-object scenes and a small set of real-world videos from Objectron — the system generalizes to complex, in-the-wild real images. Our method operates directly on real photographs and substantially outperforms recent state-of-the-art methods on large 3D edits.
作者Zhuo Cao, Lena Krieger, Fernanda Nader, Xuan Zhao, Hanno Scharr, Ira Assent
Data editing with generative methods typically requires differentiable objectives and gradient-based search. However, these assumptions break down in flow-based settings, where edits are performed through forward and backward integration and often involve non-differentiable or black-box objectives. We introduce residual-space evolutionary optimization, a model-agnostic framework that addresses this gap by combining flow-based generative editing with evolutionary algorithms. Building on the observation that conditional flow matching (CFM) can disentangle condition-controlled factors from instance-specific residuals, our framework directly operates in residual space and separates two complementary search regimes: self-pollination performs local exploitation through feature-preserving residual refinement, and cross-pollination promotes broader exploration by recombining residuals across heterogeneous samples. As a proof of concept, we validate on MorphoMNIST, a benchmark dataset for counterfactual generation, and on crystal data, demonstrating that this exploration--exploitation decomposition provides a useful mechanism for balancing target alignment, instance preservation, and diversity, and extends beyond images to real-world scientific domains.
展开完整摘要收起摘要↓
Data editing with generative methods typically requires differentiable objectives and gradient-based search. However, these assumptions break down in flow-based settings, where edits are performed through forward and backward integration and often involve non-differentiable or black-box objectives. We introduce residual-space evolutionary optimization, a model-agnostic framework that addresses this gap by combining flow-based generative editing with evolutionary algorithms. Building on the observation that conditional flow matching (CFM) can disentangle condition-controlled factors from instance-specific residuals, our framework directly operates in residual space and separates two complementary search regimes: self-pollination performs local exploitation through feature-preserving residual refinement, and cross-pollination promotes broader exploration by recombining residuals across heterogeneous samples. As a proof of concept, we validate on MorphoMNIST, a benchmark dataset for counterfactual generation, and on crystal data, demonstrating that this exploration--exploitation decomposition provides a useful mechanism for balancing target alignment, instance preservation, and diversity, and extends beyond images to real-world scientific domains.
A companion study established a de-biased, cross-model VLM-as-3D-judge that reliably ranks single-image-to-3D mesh quality where cheap geometry and CLIP proxies fall short. This paper asks: can that judge's preferences specialize a strong open generator, TRELLIS, on one asset class (furniture), cheaply and without human labels? Taking the judge from ranking to optimization is where the work lives. Pushing a VLM judge into the training and evaluation loop exposes failure modes ranking never triggered, so our contribution is an optimization-grade hardening of the judge: a training judge (Qwen2.5-VL-7B) held distinct from an evaluation judge (InternVL3-8B) to break circularity; position-bias correction; and fixes for three failure modes (image overload, geometry-hiding splat renders, and reference-free judging that rewards clean-but-wrong outputs), with calibration evidence (clear-gap win-rate 0.83-1.0; base-vs-base ~0.5). Using this protocol as an independent evaluator, and working only from public models and data with lightweight parameter-efficient adaptation, we find our methods match the strong base rather than exceed it. Independent base samples carry essentially no learnable preference (0.94 order-flip rate), so signal must be engineered by quality-contrastive construction. Across six adaptation methods, two input regimes, and a severity sweep, the most targeted - conditioner repair under severe degradation - reaches parity (0.50) with the base, while no method clears the >=65% win-rate target. The result is mechanistic: clean inputs saturate the judge, flow-DIT fine-tuning washes out through the sampler, and conditioning repair is the locus that moves geometry. Win-rates are directional at n=8 objects. Matching a strong public-data base with cheap adaptation is itself informative: exceeding it needs more than lightweight PEFT on public data, and the judge protocol is reusable.
展开完整摘要收起摘要↓
A companion study established a de-biased, cross-model VLM-as-3D-judge that reliably ranks single-image-to-3D mesh quality where cheap geometry and CLIP proxies fall short. This paper asks: can that judge's preferences specialize a strong open generator, TRELLIS, on one asset class (furniture), cheaply and without human labels? Taking the judge from ranking to optimization is where the work lives. Pushing a VLM judge into the training and evaluation loop exposes failure modes ranking never triggered, so our contribution is an optimization-grade hardening of the judge: a training judge (Qwen2.5-VL-7B) held distinct from an evaluation judge (InternVL3-8B) to break circularity; position-bias correction; and fixes for three failure modes (image overload, geometry-hiding splat renders, and reference-free judging that rewards clean-but-wrong outputs), with calibration evidence (clear-gap win-rate 0.83-1.0; base-vs-base ~0.5). Using this protocol as an independent evaluator, and working only from public models and data with lightweight parameter-efficient adaptation, we find our methods match the strong base rather than exceed it. Independent base samples carry essentially no learnable preference (0.94 order-flip rate), so signal must be engineered by quality-contrastive construction. Across six adaptation methods, two input regimes, and a severity sweep, the most targeted - conditioner repair under severe degradation - reaches parity (0.50) with the base, while no method clears the >=65% win-rate target. The result is mechanistic: clean inputs saturate the judge, flow-DIT fine-tuning washes out through the sampler, and conditioning repair is the locus that moves geometry. Win-rates are directional at n=8 objects. Matching a strong public-data base with cheap adaptation is itself informative: exceeding it needs more than lightweight PEFT on public data, and the judge protocol is reusable.
Traditional animation production relies heavily on manual drawing and iterative refinement, particularly for key-pose design, in-betweening, and character coloring. While existing animation and video generation methods have made notable progress, they typically depend on RGB boundary frames, dense frame-wise conditions, or complete sketch sequences, limiting their applicability under low-cost input conditions. We present SketchKeyAnime, a video diffusion framework for generating structurally controllable, appearance-consistent, and temporally coherent animations from sparse key-sketch inputs. Given a single reference RGB image and a few temporally indexed key sketches, SketchKeyAnime introduces a dual-branch conditioning mechanism to encode local geometric constraints alongside semantic-temporal context. It leverages Sketch Cross Attention to fuse reference image and sketch conditions with learnable gating, and incorporates an Adaptive Weighted Loss to strengthen supervision on key-sketch frames and line-art regions. Experimental results on the Aesthetic subset of Sakuga-42M show that our approach consistently outperforms representative animation interpolation and sketch-guided generation baselines. Compared to the best-performing baseline, SketchKeyAnime reduces EDMD by 31.9% and FVD by 9.5%, demonstrating superior sketch fidelity and temporal coherence, while achieving the best overall performance across most quantitative metrics. These results validate the proposed framework and highlight its potential for low-cost, highly controllable animation creation.
展开完整摘要收起摘要↓
Traditional animation production relies heavily on manual drawing and iterative refinement, particularly for key-pose design, in-betweening, and character coloring. While existing animation and video generation methods have made notable progress, they typically depend on RGB boundary frames, dense frame-wise conditions, or complete sketch sequences, limiting their applicability under low-cost input conditions. We present SketchKeyAnime, a video diffusion framework for generating structurally controllable, appearance-consistent, and temporally coherent animations from sparse key-sketch inputs. Given a single reference RGB image and a few temporally indexed key sketches, SketchKeyAnime introduces a dual-branch conditioning mechanism to encode local geometric constraints alongside semantic-temporal context. It leverages Sketch Cross Attention to fuse reference image and sketch conditions with learnable gating, and incorporates an Adaptive Weighted Loss to strengthen supervision on key-sketch frames and line-art regions. Experimental results on the Aesthetic subset of Sakuga-42M show that our approach consistently outperforms representative animation interpolation and sketch-guided generation baselines. Compared to the best-performing baseline, SketchKeyAnime reduces EDMD by 31.9% and FVD by 9.5%, demonstrating superior sketch fidelity and temporal coherence, while achieving the best overall performance across most quantitative metrics. These results validate the proposed framework and highlight its potential for low-cost, highly controllable animation creation.