Tripo AI has released EdenLive, a real-time generative renderer that turns rough visuals from coding agents or game engines into more detailed video. In the company’s own tests, the model generated 720p video at more than 24 frames per second on a single RTX 5090, running continuously for several minutes at a stretch.
The renderer takes three inputs: a 2D or 3D scene or reference video, a target first frame, and a text prompt. It outputs at 480p or 720p. Tripo AI has open-sourced the model weights, the real-time video generation code, a skill that converts text prompts into simple playable games, and a frontend for the full prompt-to-game pipeline.
EdenLive is the second and third layer of Project Eden, Tripo AI’s world model for interactive environments first released in May. Chief Scientist Dr. Cao Yanpei described the architecture as three layers: a continuously evolving world state, a state-to-visual interface, and a generative renderer. EdenLive handles the latter two without constraining the first.
| Spec | Value |
|---|---|
| Resolution | 480p / 720p |
| Frame rate | 24+ fps |
| Hardware | single RTX 5090 |
| Duration | several minutes continuous |
Tripo AI points to a few uses: interactive game previews paired with a coding agent, scene and animation previews that keep layout and camera path fixed while varying appearance, and real-time restyling such as swapping characters or overall look. For embodied AI, the company says EdenLive can convert footage of human manipulation into a robot’s point of view, pitched as a bridge between human demonstration and robot learning.
Tripo AI, founded in 2023, builds 3D-native foundation models and world models alongside Tripo Studio and its API; its mesh-generation side was covered here when Tripo P2.0 shipped native quad mesh output.
No pricing, release date for a hosted version, or benchmark comparison against other real-time renderers was published.