World Models

WorldSculpt adapts a single-object 3D generative prior to multi-view video observations, reconstructing cluttered scenes as collections of object meshes. The approach targets environments with hundreds of heavily occluded objects without scene-level training.

ActEffect, developed by Guangxiang Technology and a Tsinghua University team, uses a world model to evaluate action consequences during training but removes it from the robot’s deployment loop. The approach aims to retain better behavior while reducing online inference costs.

Puffin-World introduces a unified multimodal architecture for physical understanding, spatial simulation, 3D generation, and reconstruction. It treats physics, depth, and images as native world states and uses an Omni-Camera representation for flexible viewpoint motion.

SolarWM addresses a central bottleneck in video world-model research: the disconnect between heterogeneous data, incompatible video backbones, and long-horizon interactive inference. Its unified data contract and three-stage training recipe are designed to turn short training clips into sustained causal rollouts.

ZimaBlue presents a three-stage framework for turning large-scale egocentric video into more generalizable robot control. Its Slow-Fast design separates high-capacity world modeling from fast action prediction for real-time deployment.

H3-World extends the 33B MiniMax-H3 video generator with temporally grounded control of characters and cameras through language. Instead of adding a dedicated action module, it adapts the model’s existing semantic representations with lightweight LoRA training.

World Labs has introduced Atlas, a multimodal world model designed to connect images, video, camera motion and 3D scene understanding. Its broader ambition is to support controllable generation and robot Real-to-Sim workflows.

A video generator may produce one physically plausible future without capturing the range of outcomes that could follow the same observation and action. PAWBench evaluates whether repeated rollouts recover that underlying behavioral distribution.

A new paper argues that scaling world models requires more than larger collections of web video. Game engines could provide executable environments, grounded rewards, and long-horizon trajectories for post-training spatial models.

GameWAM combines a visual world model with an action policy for closed-loop gameplay and GUI control. It predicts future observations, generates executable keyboard-mouse trajectories, and studies a new source of bias in generative actions.

Code World Model separates persistent world evolution from visual rendering. A coding agent maintains executable state and consequences, while a video model turns those constraints into visual observations.

A new capability-based survey measures world models against the functions of traditional simulators. It finds meaningful progress in interaction and control, but major gaps in physics guarantees, state feedback, and reproducible long-horizon evolution.

EchoWM is an omnimodal world model designed for interactive, enterable media. As users continuously navigate a scene, it jointly generates 720p video, environmental sound, music, and speech while following first-person or third-person camera behavior.

ReWorld separates short-horizon action control from long-term scene memory during training, then combines a bounded KV cache with pose-indexed landmark retrieval at inference. The approach is designed to preserve revisited scenes without allowing context and compute costs to grow without bound.

ForgeWM introduces a progressive causal training recipe for converting a bidirectional action-conditioned video generator into interactive world models that run in one, two, or four denoising steps. The design targets low-latency gameplay while preserving alignment between controls and generated motion.

WorldRover introduces an Unreal Engine-based pipeline that renders long exploration routes together with geometry, motion, and correspondence annotations. The resulting data is designed for models that must maintain and revisit coherent representations of interactive worlds.

StateFlow proposes a state-centric framework for generative previsualization across film, games, architecture, and urban design. Instead of producing a video in one shot, it maintains an editable 3D world that can be constructed, evolved, and accessed over time.

Video world models can look plausible frame by frame yet drift over long horizons. WorldCycle turns reversible action cycles into a self-verifiable training signal, so “getting back to the starting state” becomes the objective.

MiniWorld aims to make video world model research easier to reproduce by offering an end-to-end training framework that does not rely on adapting large pretrained video generators. Its focus is causal streaming generation, modest compute, and open implementation.

The paper reframes world models as agent-centric interactive proxies, shifting the focus from predicting physical states to delivering usable feedback for planning, learning, and continual improvement.

WorldExam introduces a hierarchical benchmark for controllable video world models, shifting evaluation from visual appearance and instruction following toward whether a generated world can react coherently to its own state.