Tag
StrideWorld is a benchmark for interactive video world models that fixes reference geometry and action strides before motion, revealing that most models deform scenes and fail on composed routes, while only depth-scaled command models preserve expected positions.
This paper evaluates frozen video representations and finds that latent prediction errors in video world models reflect appearance and motion rather than reliably measuring physical surprise, presenting a framework that separates representation content from error signals across IntPhys2, Physion++, and Kubric benchmarks.
The paper introduces MIST, an evaluation that separates scene setup from change execution to diagnose memory failures in video world models, and proposes M3-Agent, which recursively revises instructions to triple MIST reveal scores and boost WRBench and MBench-T results.
PPMem is a video world model that uses compact plane primitives as persistent 3D memory, jointly generating long coherent videos with explicit metric depth, normals, and exportable surfaces while outperforming prior 3D-grounded world models on DL3DV and RealEstate10K.
CoWorld Reasoner 是一个利用视频世界模型作为灵活视觉推理引擎的智能体框架,通过引导视觉想象和工具协同进化,使视觉-语言模型在七个时空推理基准上取得显著提升(如 MMSI 相对提升 11.8%、ERQA 提升 13.0%)。
WorldMemBench 提出了一个用于系统评估视频世界模型世界记忆能力的基准,涵盖260个测试用例,分别考察静态记忆(人物、物体、环境的身份与几何保持)和动态记忆(事件演进与状态持续性)。对九个代表性模型的评估显示,强静态记忆并不意味着可靠的状态演化,且重复重访会暴露记忆退化问题。
HLA-WM提出一种训练免费的混合线性注意力框架,通过几何引导的检索与递归线性状态重组合来缓解GatedDeltaNet在长时视频世界模型中的远距离遗忘问题,在60秒基准上以12倍内存节省同时提升所有一致性与相机控制指标。
The paper introduces Honeycomb, a video world model that uses HexMemory, a low-rank fixed-size scene memory represented as six spatial and spatiotemporal planes, enabling constant storage during long-horizon video generation with strong generation quality and revisit consistency on WorldScore and RealEstate10K benchmarks.
QuantWM 提出一种免训练的 2-bit KV cache 量化框架,通过量化敏感性感知聚类(QSAC)与主子空间注意力补偿(PSAC)来稳定视频世界模型的注意力 logits,缓解时间闪烁与视觉退化,并在多个基准上实现最高 6.20 倍 KV cache 内存压缩。
SolarWM introduces an open framework and unified training recipe for building interactive video world models with scalable training across diverse data sources, enabling long-horizon real-time rollouts.
ForgeWM is a progressive framework that distills bidirectional video generators into efficient few-step interactive world models, supporting low-latency interaction and replay-time refinement with demonstrated improvements on Minecraft and FPS gameplay.
A best-paper research from NVIDIA and collaborators introduces WorldTrace, a training-free framework that keeps compressed memory addressable in autoregressive video world models by assigning fixed slot-rank positions, enabling coherent long rollouts and long-range recall beyond the training horizon.
Introduces Latent Dynamics Reasoning (LDR), a video world model that integrates kinematic dynamics in a structured latent space, enabling extrapolation of learned dynamics far beyond training distributions while using far fewer parameters and running much faster than video diffusion baselines.
This paper introduces WorldTrace, a training-free memory framework for long-horizon video world models that keeps compressed cache addressable, plus LoopBench, a benchmark for episodic recall after long detours. It improves temporal consistency by +15.5% and episodic recall by +19.5% on LoopBench.
HelloWorld is a video world model that enables socially interactive characters, allowing users to prompt on-screen characters to respond via a single button press. It uses self-distillation and training-free cross-attention masking to naturalize interactions, and introduces HelloWorldBench for evaluation.
MiniWorld is a reproducible framework for training video world models from scratch using a block-causal Video Diffusion Transformer with Flow Matching, enabling efficient streaming generation and trainable in days on a single 8-GPU server.
ShadowDancer proposes a method for any-action, frame-level control of interactive video world models by learning unified dynamics representations from a video and its shadow, enabling transferable action control without labels or motion estimators. Experiments show improved action transfer and rollout performance over baselines.
AlayaWorld is an open-source framework for building interactive generative worlds that enables real-time user interaction and supports diverse actions. It unifies the complete development pipeline from data preparation to deployment.
MemLearner proposes a learning-based adaptive context query method using query tokens to improve scene consistency and memory in video world models, particularly for long sequences with occlusions and dynamic objects.
Holo-World presents a unified controllable video world model that generates videos from a single image with explicit control over camera, object motion, and weather. It introduces a novel dataset and techniques to preserve scene structure while transferring to target weather states.