01
AI video & UGC
Exploring more affordable, controllable visual storytelling.
A research company
The Visual
Renaissance.
We are building video foundation models to make visual generation an everyday AI capability.
Explore our research
01 / Research
A different scaling direction
More possibility.
Less friction.
Video can be a medium for thinking, creating, and interacting. To become part of everyday life, it needs to be affordable, consistent, and controllable.
Our research explores a different balance: scale logic up. Scale vision down. A simple ambition for a new generation of visual intelligence.
Our ambition
100× lower generation cost
Target economics
$1 per hour of generated video
Research goals, paired with Seedance 2.0 visual quality and stronger consistency, physics, and shot control. These are targets, not current pricing or measured production results.
A different balance
Up.
Scale logic up.
Down.
Scale vision down.
02 / Explorations
Recorded research prototypes
One medium.
New possibilities.
From stories and games to software and physical interaction, we are exploring where visual generation can become useful.
These recorded prototypes illustrate research and application directions. They are not released products or performance benchmarks.
02
Generative gaming
Exploring worlds that respond to player actions.
03
Generative software
Exploring visual interfaces that adapt to a task.
04
Robotics
Exploring visual predictions and physical interaction.
03 / Team
Experience across the visual frontier
Frontier research.
Built into reality.
Our team brings experience building video foundation models, training world models, pioneering 3D generation, and turning real-time video research into a company acquired by Roblox.
Research and engineering come together across data, model architecture, pre-training, post-training, infrastructure, and agentic workflows.
Selected prior experience
01
Adobe
Video foundation models
02
DeepMind
Genie & Veo training
03
Seedance
Early model research
04
Roblox
Real-time video acquisition
Research recognition
Best-paper recognition.
- CVPR
- ICCV
- SIGGRAPH Asia
- CoRL
- ICRA
- IROS
01 / Foundation models
Adobe. DeepMind. Seedance.
Experience founding Adobe’s video model team, co-leading DeepMind Genie and Veo training, and first-authoring the Seedance prototype.
02 / Interactive worlds
Research that became a company.
Built a real-time video startup acquired by Roblox, connecting world-model research with interactive systems.
03 / 3D generation
A new way to reconstruct.
Created large reconstruction models that helped establish modern feedforward 3D generation, including LRM and GS-LRM.
04 / Training systems
~10K GPU infrastructure
Prior experience scaling foundation-model training and infrastructure at approximately 10,000-GPU scale.
05 / Full stack
From data to deployment.
Capabilities across data, multimodal pre-training, post-training, efficient systems, and agentic research workflows.
04 / Journal
Questions guiding our work
Notes from the lab.
Build with us
A new medium
starts with new questions.
We would love to hear from researchers and builders exploring what everyday visual intelligence could become.
Get in touch