World models

Google DeepMind's real-time interactive world model, which generates explorable worlds from text and image prompts with native audio and lets users steer the scene mid-simulation in natural language, with Gemini models keeping it coherent.

Released

Oct 27, 2026

Output

Live video stream

Session

—

Tencent Hunyuan's open-weight free-form interactive world model that builds a world one video segment at a time from text, which users keep steering with multi-turn text instructions and camera trajectories as generation continues.

Released

Oct 21, 2026

Output

Live video stream

Session

—

Odyssey's foundation world model, an autoregressive diffusion transformer that predicts in real time how a text-prompted environment changes as a person or agent acts in it; also used to drive robot, vehicle, drone and game policies.

Released

Oct 8, 2026

Output

Live video stream

Session

—

PixVerse's real-time 'omni' world model that keeps one audiovisual world running while the user is inside it, updating it from text, visual references, audio and movement controls, with earlier actions carried forward.

Released

Sep 22, 2026

Output

Live video stream

Session

—

Runway's real-time interactive world model that turns video-and-audio generation into a steerable simulation: users define environment, subjects, style, physical rules and ambience, then steer with free-form text actions and continuous camera motion.

Released

Sep 3, 2026

Output

Live video stream

Session

—

World Labs' 'omni' world model, trained from scratch on text, images, video and 3D, that generates and reconstructs scenes as video or as point clouds / 3D Gaussian splats with precise camera control.

Released

Sep 1, 2026

Output

3D scene

Session

60 s

Open-weight, three-step, CFG-free interactive video world model from Alaya Lab with camera control and mid-rollout prompt switching, aimed at endless long-horizon rollouts.

Released

Aug 13, 2026

Output

Live video stream

Session

—

Open-weight 5B action-conditioned interactive video world model from Alibaba's AMAP CV Lab that streams open-ended worlds on a single consumer GPU.

Released

Jul 9, 2026

Output

Live video stream

Session

—

Open-weight real-time interactive world model from Robbyant, Ant Group's embodied-AI unit, with an unbounded interaction horizon, rich in-world actions, and Pilot/Director agents.

Released

Jul 9, 2026

Output

Live video stream

Session

—

Decart's real-time, action-conditioned world model for physical AI that streams three synchronized camera views of one continuous environment, sold by the second through Decart's API.

Released

Jun 10, 2026

Output

Live video stream

Session

—

NVIDIA's open omnimodal world foundation model family for physical AI that reasons about and generates text, images, video, ambient sound and actions, usable as a world/video model, VLM, or backbone for robot policies.

Released

May 31, 2026

Output

Simulation video

Session

—

SpAItial's 3D world generation model that turns a text prompt or image into a persistent, explorable 3D Gaussian-splat scene (with semantic segmentation) rather than a video.

Released

Apr 28, 2026

Output

3D scene

Session

—

Real-time open-ended world model from Alibaba's ATH (Alibaba Token Hub) unit with joint audio-video generation, offering a Directing mode to steer a live scene and a Wandering mode to explore a world in first person.

Released

Apr 16, 2026

Output

Live video stream

Session

3 min

Tencent HY's open multimodal 3D world model that generates, reconstructs and simulates editable 3D worlds (meshes, 3D Gaussian splats, point clouds) from text, images or video, exportable to game engines.

Released

Apr 16, 2026

Output

3D scene

Session

—

NVIDIA research framework that turns a single image plus a camera path into a large, persistent, walkable 3D Gaussian scene by generating camera-controlled video and lifting it to 3D.

Released

Apr 14, 2026

Output

3D scene

Session

—

Overworld's open-weight real-time world model that runs locally on consumer GPUs or streams in the browser, generating a playable first-person world that reacts to keyboard and mouse input.

Released

Apr 9, 2026

Output

Live video stream

Session

—

World Labs' generally available multimodal world model that turns text, images, multi-view images, panoramas or video into navigable, editable 3D worlds (Gaussian splats and meshes), with a Plus variant for larger worlds.

Released

Apr 7, 2026

Output

3D scene

Session

—

Skywork AI's open-source real-time streaming interactive world model with camera-aware long-horizon memory, fine-tuned from Wan2.2-TI2V-5B.

Released

Mar 27, 2026

Output

Live video stream

Session

—

Google DeepMind's general-purpose real-time world model, available to consumers through the Project Genie prototype (with Nano Banana Pro and Gemini) to sketch, explore and remix playable worlds from text and images.

Released

Jan 29, 2026

Output

Live video stream

Session

60 s

Tencent HY's open-source real-time interactive streaming video world model with long-term geometric consistency, responding to keyboard and mouse input; the video-stream counterpart to HY-World 2.0's 3D output.

Released

Dec 17, 2025

Output

Live video stream

Session

—