[
  {
    "id": "world-models-2018",
    "name": "World Models",
    "org": "unknown",
    "route": "latent",
    "one_liner": "在世界模型生成的梦境环境中训练智能体，再将策略迁回实际环境。",
    "input": "记忆模型 M 接收当前潜在向量和动作。",
    "output": "用于采样下一潜在向量预测的高斯混合分布参数。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://worldmodels.github.io/"
    ],
    "imported_on": "2026-07-09"
  },
  {
    "id": "sora-world-simulators",
    "name": "Sora（2024 技术报告）",
    "org": "OpenAI",
    "route": "pixel",
    "one_liner": "OpenAI 在这篇技术报告中介绍可生成最长一分钟视频的 Sora。",
    "input": "图像和提示词。",
    "output": "一分钟视频。",
    "action_conditioned": "unknown",
    "public": "unknown",
    "open_source": "unknown",
    "limit": "该报告指出，Sora 不能准确建模一些基础物理交互，例如玻璃破碎。",
    "confused_with": "unknown",
    "sources": [
      "https://openai.com/index/video-generation-models-as-world-simulators/"
    ],
    "imported_on": "2026-07-09"
  },
  {
    "id": "copilot4d",
    "name": "Copilot4D",
    "org": "Waabi",
    "route": "latent",
    "one_liner": "先将传感器观测编码为 token，再通过离散扩散预测未来的世界模型。",
    "input": "BEV token 和自车动作。",
    "output": "由预测的 BEV token 解码得到的未来点云。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://waabi.ai/research/copilot-4d"
    ],
    "imported_on": "2026-07-27"
  },
  {
    "id": "genie-3",
    "name": "Genie 3",
    "org": "Google DeepMind",
    "route": "interactive",
    "one_liner": "根据文本提示生成可实时导航的动态世界。",
    "input": "导航输入和文本交互。",
    "output": "可实时导航的动态视觉世界。",
    "action_conditioned": true,
    "public": "limited_access",
    "open_source": "unknown",
    "limit": "该发布文称，连续交互只能维持几分钟。",
    "confused_with": "unknown",
    "sources": [
      "https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/"
    ],
    "imported_on": "2026-07-09"
  },
  {
    "id": "vjepa-2-ac",
    "name": "V-JEPA 2-AC",
    "org": "Meta AI",
    "route": "latent",
    "one_liner": "冻结预训练视频编码器后，在其表征之上另行训练的动作条件预测器。",
    "input": "过去的视频帧、动作和末端执行器状态。",
    "output": "以控制动作和本体感知观测为条件，预测未来视频观测的表征。",
    "action_conditioned": true,
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "自回归预测会累积误差，限制长时程规划的可靠性。",
    "confused_with": [
      "V-JEPA 2 基础视频表征模型"
    ],
    "sources": [
      "https://ai.meta.com/blog/v-jepa-2-world-model-benchmarks/",
      "https://arxiv.org/abs/2506.09985v1",
      "https://arxiv.org/pdf/2506.09985v1",
      "https://github.com/facebookresearch/vjepa2"
    ],
    "imported_on": "2026-07-12"
  },
  {
    "id": "marble",
    "name": "Marble",
    "org": "World Labs",
    "route": "interactive",
    "one_liner": "从多种输入生成 3D 世界，并支持交互编辑、扩展和组合。",
    "input": "文本、图像、视频或粗略 3D 布局。",
    "output": "可导出为 Gaussian splats、网格或视频的 3D 世界。",
    "action_conditioned": "unknown",
    "public": "public",
    "open_source": "unknown",
    "limit": "碰撞网格为低保真网格，用于粗略物理模拟。",
    "confused_with": "unknown",
    "sources": [
      "https://www.worldlabs.ai/blog/marble-world-model"
    ],
    "imported_on": "2026-07-28"
  },
  {
    "id": "gwm-1",
    "name": "GWM-1",
    "org": "Runway",
    "route": "interactive",
    "one_liner": "基于 Gen-4.5、逐帧生成且可通过动作控制的实时自回归模型。",
    "input": "家族所述动作控制包括相机位姿、机器人指令和音频。",
    "output": "逐帧生成的视觉序列。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": [
      "单一、统一的通用模型"
    ],
    "sources": [
      "https://runway.com/research/introducing-runway-gwm-1"
    ],
    "imported_on": "2026-07-27"
  },
  {
    "id": "waymo-world-model",
    "name": "Waymo World Model",
    "org": "Waymo",
    "route": "interactive",
    "one_liner": "基于 Genie 3 的 Waymo 世界模型。",
    "input": "语言提示、驾驶输入和场景布局。",
    "output": "包含相机与激光雷达数据的多传感器模拟输出。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "模拟时间越长，计算和维持稳定质量越困难。",
    "confused_with": "unknown",
    "sources": [
      "https://waymo.com/blog/2026/02/the-waymo-world-model-a-new-frontier-for-autonomous-driving-simulation/"
    ],
    "imported_on": "2026-07-27"
  },
  {
    "id": "cosmos-3",
    "name": "Cosmos 3",
    "org": "NVIDIA",
    "route": "physics",
    "one_liner": "联合处理和生成语言、图像、视频、音频及动作序列的世界模型家族。",
    "input": "语言、图像、视频、音频和动作序列。",
    "output": "语言、图像、视频、音频和动作序列。",
    "action_conditioned": true,
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Launches-Cosmos-3-the-Open-Frontier-Foundation-Model-for-Physical-AI/default.aspx",
      "https://arxiv.org/abs/2606.02800v4",
      "https://blogs.nvidia.com/blog/siggraph-news-2026/",
      "https://www.nvidia.com/en-us/ai/cosmos/"
    ],
    "imported_on": "2026-08-06"
  },
  {
    "id": "gaia-4",
    "name": "GAIA-4",
    "org": "Wayve",
    "route": "interactive",
    "one_liner": "让 AI Driver 参与闭环模拟的 GAIA 世界模型。",
    "input": "记录的传感器数据。",
    "output": "相机与雷达的多模态模拟数据。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "World-on-rails 是可选回放约束，并非模型整体限制；不能把该模式等同于完全交互的交通模拟。",
    "confused_with": [
      "world-on-rails 模式下完全响应自车的交通模拟"
    ],
    "sources": [
      "https://wayve.ai/thinking/gaia-4/"
    ],
    "imported_on": "2026-08-05"
  },
  {
    "id": "dreamerv3",
    "name": "DreamerV3（Nature 2025）",
    "org": "Google DeepMind / University of Toronto（论文署名）",
    "route": "latent",
    "one_liner": "学习环境模型，通过想象未来情景改进智能体行为。",
    "input": "感官观测及动作。",
    "output": "未来内部表征和奖励。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": true,
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://www.nature.com/articles/s41586-025-08744-2",
      "https://github.com/danijar/dreamerv3"
    ],
    "imported_on": "2026-10-01"
  },
  {
    "id": "dino-wm",
    "name": "DINO-WM（ICML 2025）",
    "org": "New York University / Meta FAIR（论文署名）",
    "route": "latent",
    "one_liner": "将目标特征作为预测目标，通过优化动作序列进行视觉目标规划。",
    "input": "视觉表征和动作。",
    "output": "未来视觉 patch 特征。",
    "action_conditioned": true,
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://dino-wm.github.io/",
      "https://proceedings.mlr.press/v267/zhou25t.html",
      "https://github.com/gaoyuezhou/dino_wm"
    ],
    "imported_on": "2026-10-01"
  },
  {
    "id": "daydreamer",
    "name": "DayDreamer（CoRL 2022）",
    "org": "DayDreamer 研究团队",
    "route": "latent",
    "one_liner": "在实体机器人上学习世界模型和行为策略，不依赖模拟器。",
    "input": "本体感知和视觉观测。",
    "output": "动作条件下预测的内部表征序列。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "长时间在硬件上学习会造成磨损，可能需要人工干预或维修。",
    "confused_with": "unknown",
    "sources": [
      "https://danijar.com/project/daydreamer/",
      "https://proceedings.mlr.press/v205/wu23c.html",
      "https://github.com/danijar/daydreamer",
      "https://arxiv.org/pdf/2206.14176v1"
    ],
    "imported_on": "2026-10-01"
  },
  {
    "id": "diamond-atari",
    "name": "DIAMOND（Atari / NeurIPS 2024）",
    "org": "DIAMOND 研究团队",
    "route": "pixel",
    "one_liner": "通过扩散世界模型预测游戏下一帧，并在模型中训练强化学习智能体。",
    "input": "历史游戏帧和智能体动作。",
    "output": "下一游戏画面帧。",
    "action_conditioned": true,
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://diamond-wm.github.io/",
      "https://github.com/eloialonso/diamond",
      "https://arxiv.org/abs/2405.12399"
    ],
    "imported_on": "2026-10-01"
  },
  {
    "id": "ctrl-world",
    "name": "Ctrl-World（ICLR 2026）",
    "org": "Stanford University / Tsinghua University（论文署名）",
    "route": "interactive",
    "one_liner": "可与 VLA 策略共同在想象空间中滚动生成的动作条件世界模型。",
    "input": "初始画面和动作序列。",
    "output": "联合预测的多视角视频，包括腕部视角。",
    "action_conditioned": true,
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "模型对物理交互的建模仍可能不精确。",
    "confused_with": "unknown",
    "sources": [
      "https://ctrl-world.github.io/",
      "https://github.com/Robert-gyj/Ctrl-World",
      "https://huggingface.co/yjguo/Ctrl-World",
      "https://arxiv.org/abs/2510.10125"
    ],
    "imported_on": "2026-10-01"
  },
  {
    "id": "wham-2025",
    "name": "WHAM（Nature 2025）",
    "org": "Microsoft Research",
    "route": "interactive",
    "one_liner": "联合建模游戏画面与控制器动作，可生成画面、动作或二者的序列。",
    "input": "游戏图像和控制器动作条件。",
    "output": "预测的游戏画面和控制器动作序列，依具体模式而定。",
    "action_conditioned": true,
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "论文的游戏案例聚焦单一 3D 电子游戏。",
    "confused_with": "unknown",
    "sources": [
      "https://www.microsoft.com/en-us/research/project/wham/wham/",
      "https://www.nature.com/articles/s41586-025-08600-3"
    ],
    "imported_on": "2026-10-01"
  },
  {
    "id": "vista-2024",
    "name": "Vista（NeurIPS 2024）",
    "org": "OpenDriveLab",
    "route": "interactive",
    "one_liner": "面向驾驶、强调通用性与多种控制条件的世界模型。",
    "input": "历史帧与高层意图，包括指令和目标点。",
    "output": "未来驾驶场景的视觉预测。",
    "action_conditioned": true,
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://github.com/OpenDriveLab/Vista",
      "https://arxiv.org/abs/2405.17398"
    ],
    "imported_on": "2026-10-01"
  },
  {
    "id": "atlas",
    "name": "Atlas（World Labs / 2026）",
    "org": "World Labs",
    "route": "interactive",
    "one_liner": "以共享空间上下文生成多模态内容的世界模型。",
    "input": "文本、图像、相机位姿和三维深度图；视频表示为图像序列。",
    "output": "图像帧与三维深度信息，可组成点云或 Gaussian splats。",
    "action_conditioned": "unknown",
    "public": "limited_access",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": [
      "当前 Marble 产品"
    ],
    "sources": [
      "https://www.worldlabs.ai/blog/atlas"
    ],
    "imported_on": "2026-10-01"
  },
  {
    "id": "leworldmodel",
    "name": "LeWorldModel（LeWM / 2026 研究）",
    "org": "LeWorldModel 联合研究团队",
    "route": "latent",
    "one_liner": "从像素端到端训练的 JEPA 世界模型。",
    "input": "当前图像的潜在表征和动作。",
    "output": "下一帧观测的潜在表征。",
    "action_conditioned": true,
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://le-wm.github.io/",
      "https://github.com/lucas-maes/le-wm",
      "https://huggingface.co/collections/quentinll/lewm",
      "https://arxiv.org/abs/2603.19312"
    ],
    "imported_on": "2026-10-01"
  },
  {
    "id": "h-jepa",
    "name": "H-JEPA",
    "org": "H-JEPA 研究团队",
    "route": "latent",
    "one_liner": "在各层潜在空间中，用历史状态与动作预测不同时间尺度的未来。",
    "input": "过去的潜在状态与动作。",
    "output": "各层的未来潜在状态。",
    "action_conditioned": true,
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2610.06805v1",
      "https://github.com/kevinghst/H-JEPA",
      "https://huggingface.co/jepa-world-models/h-jepa"
    ],
    "imported_on": "2026-10-07"
  },
  {
    "id": "uniwam",
    "name": "UniWAM",
    "org": "UniWAM 研究团队",
    "route": "pixel",
    "one_liner": "联合学习语义理解、未来视觉预测与动作生成。",
    "input": "当前观测、本体感知状态与任务指令。",
    "output": "语言输出、未来观测与动作。",
    "action_conditioned": "unknown",
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2610.02054v3",
      "https://github.com/UniWAM/UniWAM"
    ],
    "imported_on": "2026-10-07"
  },
  {
    "id": "deltaworld",
    "name": "DeltaWorld",
    "org": "DeltaWorld 研究团队",
    "route": "latent",
    "one_liner": "预测动作引起的潜在状态增量，再更新状态并解码未来画面。",
    "input": "初始视觉上下文与机器人动作序列。",
    "output": "解码后的未来 RGB 观测。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2610.02691v1"
    ],
    "imported_on": "2026-10-07"
  },
  {
    "id": "v2x-wam",
    "name": "V2X-WAM",
    "org": "V2X-WAM 研究团队",
    "route": "latent",
    "one_liner": "以候选行驶轨迹与动作预测未来交通占据和动态流。",
    "input": "车路协同场景表征、所选轨迹与动作序列。",
    "output": "未来交通占据与动态流。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2609.37098v2"
    ],
    "imported_on": "2026-10-07"
  },
  {
    "id": "gwm-worlds-2",
    "name": "GWM Worlds 2",
    "org": "Runway",
    "route": "interactive",
    "one_liner": "由文本动作与连续相机输入驱动的自回归视听世界模型。",
    "input": "场景初始提示与首帧、当前相机和文本动作、历史生成帧。",
    "output": "逐步生成的视频与音频。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "快速相机旋转可能使细节、纹理和几何退化；长时程记忆仍不完善。",
    "confused_with": "unknown",
    "sources": [
      "https://runway.com/research/introducing-gwm-worlds-2"
    ],
    "imported_on": "2026-10-07"
  },
  {
    "id": "alayaworld-v1-1",
    "name": "AlayaWorld v1.1",
    "org": "AlayaLab",
    "route": "interactive",
    "one_liner": "以流式 3D 点缓存和重新渲染的空间条件生成视角可控视频。",
    "input": "经重新渲染的空间记忆与运动感知视觉条件。",
    "output": "按视角条件生成的视频。",
    "action_conditioned": "unknown",
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2608.13492v1",
      "https://github.com/AlayaLab/AlayaWorld",
      "https://huggingface.co/AlayaLab/AlayaWorld-v1.1-stage3"
    ],
    "imported_on": "2026-10-07"
  },
  {
    "id": "alaya-evoke",
    "name": "Alaya-EVOKE",
    "org": "AlayaLab",
    "route": "interactive",
    "one_liner": "借助外部世界状态库，持续生成受相机和文本条件控制的视频。",
    "input": "时间变化的文本、相机轨迹与投影历史。",
    "output": "交互生成的视频。",
    "action_conditioned": "unknown",
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "固定预算的循环结构本身不能保证长时程内容一致。",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2608.13546v2",
      "https://github.com/AlayaLab/Evoke",
      "https://huggingface.co/AlayaLab/Evoke",
      "https://huggingface.co/AlayaLab/Evoke-Turbo"
    ],
    "imported_on": "2026-10-07"
  },
  {
    "id": "inspatio-world-1-5",
    "name": "InSpatio-World-1.5",
    "org": "InSpatio 研究项目",
    "route": "interactive",
    "one_liner": "按用户提供的目标相机轨迹，从图像或视频生成视图视频。",
    "input": "单张图像、四张图像或视频，以及用户提供的目标相机轨迹。",
    "output": "源画面、渲染画面、遮罩与预测视频文件。",
    "action_conditioned": "unknown",
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://inspatio.github.io/inspatio-world-1.5/",
      "https://github.com/inspatio/inspatio-world-v1.5",
      "https://huggingface.co/inspatio/world-1.5"
    ],
    "imported_on": "2026-10-07"
  },
  {
    "id": "solaris",
    "name": "Solaris",
    "org": "Runway",
    "route": "web",
    "one_liner": "把用户交互作为条件，逐帧生成视觉界面。",
    "input": "用户点击、拖拽等已发生的交互。",
    "output": "逐帧生成的视觉界面。",
    "action_conditioned": true,
    "public": "application_required",
    "open_source": "unknown",
    "limit": "稳定、可读的文字仍是视频界面生成的困难。",
    "confused_with": "unknown",
    "sources": [
      "https://runway.com/news/research/introducing-solaris"
    ],
    "imported_on": "2026-10-07"
  },
  {
    "id": "odyssey-3",
    "name": "Odyssey-3",
    "org": "Odyssey",
    "route": "interactive",
    "one_liner": "以历史观察、提示和动作持续预测可交互的视觉环境。",
    "input": "历史观察、文本提示和动作输入；预览支持第一/第三人称导航与独立相机移动。",
    "output": "随交互持续生成的未来环境视频。",
    "action_conditioned": true,
    "public": "limited_access",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://odyssey.systems/meet-odyssey-3"
    ],
    "imported_on": "2026-10-10"
  },
  {
    "id": "dreamtrue",
    "name": "DreamTrue",
    "org": "DreamTrue 研究团队",
    "route": "pixel",
    "one_liner": "以机器人动作轨迹为条件，预测多视角未来视频。",
    "input": "初始场景与机器人动作条件；轨迹经几何校准后渲染到图像空间。",
    "output": "多视角未来机器人操作视频。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "动作表示、生成器和奖励模型均在图像空间工作；遮挡或视角不足时，可能生成或奖励看似合理但物理错误的交互。",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2610.12468v1",
      "https://github.com/brave-eai/DreamTrue"
    ],
    "imported_on": "2026-10-10"
  },
  {
    "id": "planwam",
    "name": "PlanWAM",
    "org": "PlanWAM 研究团队",
    "route": "latent",
    "one_liner": "从历史观察预测有助于驾驶规划的未来表征，再生成并选择轨迹。",
    "input": "历史多视角相机观察、自车状态与导航信息。",
    "output": "面向规划的未来潜在表征，以及由其引导生成和选择的自车轨迹。",
    "action_conditioned": "unknown",
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2610.11382v1"
    ],
    "imported_on": "2026-10-10"
  },
  {
    "id": "long-wam",
    "name": "Long-WAM",
    "org": "Long-WAM 研究团队",
    "route": "pixel",
    "one_liner": "把长视觉历史与未来视频预测结合，用于生成机器人动作块。",
    "input": "观测视频历史、机器人状态与语言任务指令；具体格式随变体而定。",
    "output": "未来视频潜变量与机器人动作块；IDM 先预测视频，再生成动作。",
    "action_conditioned": "unknown",
    "public": "open_weights",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2610.10528v1",
      "https://github.com/NVlabs/LongLive/blob/main/Long-WAM/README.md",
      "https://huggingface.co/Efficient-Large-Model/Long-WAM-LIBERO-IDM"
    ],
    "imported_on": "2026-10-10"
  },
  {
    "id": "ppwm",
    "name": "PPWM",
    "org": "PPWM 研究团队",
    "route": "latent",
    "one_liner": "在有限跨度内并行预测有因果关联的未来表征轨迹，用于动作规划。",
    "input": "近期潜在状态与已执行动作组成的上下文，以及待预测的动作序列。",
    "output": "有限跨度内的多步未来潜在表征轨迹。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2610.08627v1"
    ],
    "imported_on": "2026-10-10"
  },
  {
    "id": "token-world",
    "name": "Token-World",
    "org": "Token-World 研究团队",
    "route": "latent",
    "one_liner": "在策略使用的 VLM 视觉 token 空间中，按动作预测未来状态。",
    "input": "压缩的 VLM 视觉 token、本体感知与机器人动作历史。",
    "output": "未来压缩 token 状态，可映射回策略使用的 VLM 视觉表征。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://arxiv.org/html/2610.00575v2"
    ],
    "imported_on": "2026-10-10"
  },
  {
    "id": "pixverse-r2",
    "name": "PixVerse R2",
    "org": "PixVerse",
    "route": "interactive",
    "one_liner": "以历史状态和实时多模态输入，持续生成同步音视频环境。",
    "input": "历史环境状态，以及实时文本、参考材料、音频、WASD 等动作输入。",
    "output": "连续更新的同步视频与音频片段。",
    "action_conditioned": true,
    "public": "unknown",
    "open_source": "unknown",
    "limit": "unknown",
    "confused_with": "unknown",
    "sources": [
      "https://pixverse.ai/en/blog/pixverse-r2-scaling-real-time-omni-world-models",
      "https://pixverse.ai/en/blog/pixverse-introduces-r2-real-time-world-model"
    ],
    "imported_on": "2026-10-10"
  }
]