Meta/Oxford multi-agent video world model that adds fixed auxiliary cameras to keep a shared world consistent.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
OverSeer is a research multi-agent video world model from the University of Oxford and Meta. It generates each player's egocentric view jointly with a few fixed auxiliary camera views. Because the auxiliary cameras never move, they act as a persistent shared reference: places stay as they were when a player returns, and one player's actions are visible to the others.
To keep the extra streams cheap, a δ-VAE codes each auxiliary camera as one reference frame plus a compact residual per later frame. This uses 108 tokens per latent frame instead of 864. The model generates all views frame by frame from each agent's first frame, its actions and the camera poses.
The project page shows demos in Minecraft, Doom and real multi-camera studio capture (two head-mounted and six fixed cameras). It compares OverSeer with an ego-only variant and with baselines named Solaris and γ-World on tasks such as reuniting after walking apart and returning to the outcome of an event. According to a YouTube creator summary, the model itself is not released and only a dataset is. No code, weights or paper links were found on the page.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
World model simulating a shared consistent world for multiple agents using fixed extra camera views. Demos in Minecraft, Doom and real life; model not released, only dataset.
Closest neighbours by meaning, not just by category.
Model · Tencent ARC Lab
Video world model from Tencent ARC Lab that turns an image or text into camera-controlled explorable scenes with 3D-aware memory.
Model
Research multi-player video world model with hierarchical communication, trained on 4 players and run with up to 10.
Research
Research framework where a coding agent maintains world state as code, rendered to video by a fine-tuned MiniMax-H3 model.
Model · Meta
Meta's open-weights 30B dense model for local agents, tuned for tool use and long tasks, Apache 2.0 licensed.
Research
Research video world model that reasons in a learned discrete "physical language" before rendering video.
Model
14B autoregressive world model that generates interactive, camera-controlled video worlds in 3 denoising steps.
