A $10 Million Bet on AI That Streams Worlds Forever, Not Just Clips
Visko's Orbis model generates live, interactive 4K video indefinitely, and a Stanford-trained founder says it could train the robots heading to your home.
Key points
- Visko Platform Inc. raised $10 million in pre-seed funding, led by Llama Ventures, announced today.
- Its Orbis model streams 4K video at 24 frames per second continuously, without stopping to render a finished clip.
- Users can change their text instructions mid-stream and the video updates in real time instead of restarting.
- Orbis topped eight competing systems in a human preference study on long-form video quality, according to Visko's own technical report.
- The company sees its biggest near-term market in training data for robots and autonomous vehicles, not consumer entertainment.
Most AI video tools work like a photo lab. You hand over a prompt, wait while the machine develops the image, and get back a finished clip you cannot touch. Visko Platform Inc., a 16-person startup based in Sunnyvale, California, wants to replace that model entirely.
Today the company opened free public access to Orbis, its foundation model (a large, general-purpose AI trained on vast amounts of data that other products can be built on top of). At the same time, it announced it has raised $10 million in pre-seed funding led by Llama Ventures.
What does Orbis actually do differently?
Orbis streams video the way a river flows: it never stops. The first frame appears almost instantly, and each new frame is generated using everything that came before it, keeping the scene consistent as it grows.
Think of it less like a video camera and more like a video game engine that AI is running in real time. Type a new instruction mid-stream and the world shifts around it. The video does not restart. It adapts.
"Most video models optimize every frame of a clip at once and show nothing until the whole thing resolves," founder and CEO Qing (Will) Yin told The Robot Report. "Orbis finishes the first frame, streams it, then generates the next conditioned on what came before."
Visko says Orbis can hold 4K quality at 24 frames per second for up to an hour without the colours washing out or the scene losing its internal logic. Its technical report placed Orbis first among eight systems in a human preference study measuring long-form stability.
| What Orbis claims | The number |
|---|---|
| Video resolution | 4K |
| Frame rate | 24 fps |
| Max continuous duration tested | 1 hour |
| Pre-seed funding raised | $10 million |
| Team size | 16 people |
| Funding lead | Llama Ventures |
Who is actually paying for this?
Not gamers. Not filmmakers. At least not yet.
The company's clearest near-term revenue case is selling synthetic training footage to robotics and autonomous vehicle companies. Gathering real-world video of dangerous edge cases, say a robot dropping a fragile object or a car hitting ice, is expensive and sometimes impossible to stage safely. Orbis can generate that footage on demand.
"Change the prompt mid-run, and the world updates around it instead of restarting," Yin said. "That makes it usable for simulation and for training footage that is unsafe or expensive to stage."
Advisory board chairman Michael I. Jordan of UC Berkeley also flagged gaming, live commerce, and education as longer-term opportunities.
What still needs fixing?
Visko published its own shortcomings, which is worth noting. Yin says scene consistency across very long runs still needs work, and the model cannot yet follow precise physical instructions like "raise the arm exactly 45 degrees." The team is also exploring whether to feed in touch-sensor data alongside video.
Here is the honest caveat: this is a pre-seed company with one published model, self-funded benchmarks, and a road map that depends on winning robot-training contracts that do not yet exist at scale. Yin himself said general-purpose home robots could still be a decade away.
If you want to see what the technology looks like today, Visko opened public access at visko.ai.
Your takeaway: If you work in robotics, simulation, or autonomous systems, Orbis is worth a look as a source of synthetic training data. For everyone else, file this one under "watch closely" rather than "act now."
