Odyssey-2
Type into a video and watch its future change live.
Save to collection
Choose a home for Odyssey-2.
Start your first collection below.
Why it’s worth your time
Odyssey-2 is a general-purpose world model that begins generating video immediately and lets the viewer redirect it through natural-language input while it continues to run. Instead of calculating a fixed clip with a known ending, it generates each new frame from what has already happened and the latest instruction. The experience is deliberately spare, with a large video field and a small interaction layer that leave the unfolding scene at the centre. It is best understood as a research experience with public controls, not a conventional production product.
The important moment is the first time an instruction alters what comes next. That shift turns the viewer into a participant, even when objects smear, places lose structure, or a direction is interpreted loosely. Fidelity, memory, control, and persistent geography are not yet strong enough for dependable applications, and the company is clear that the larger uses remain unsolved. As a current tool, Odyssey-2 is visibly early. As an idea, it is profound: video stops being a finished timeline and becomes a future negotiated while it plays.
Shared by
Spotted and shared by a Biltib member“The signal is not the first frame. It is the moment a typed direction changes what the video becomes next. The scene can smear, forget its structure, or absorb an instruction loosely, but my role has already shifted from viewer to participant. Odyssey-2 is not a dependable production tool today. It is a startling preview of media whose future has not happened yet.”
Community notes
Keep discovering
Explore more →
