Tripo AI Open-Sources EdenLive, Reporting 24.7 FPS on RTX 5090
Tripo AI has released EdenLive, reporting about 24.7 FPS on an RTX 5090; the system renders video from reference scenes rather than generating 3D meshes.
VAST AI Research, the team behind Tripo AI, has released EdenLive, a real-time generative renderer that converts 2D or 3D scenes and reference footage into detailed video. It takes three inputs—a reference scene or video, a target first frame and a text prompt—and produces output at 480p or 720p.
In the developer team’s measurements, the released 5B model reached about 24.7 FPS inference throughput at 1280 × 704 pixels on a single NVIDIA GeForce RTX 5090. The release includes 5B and 14B model weights, streaming inference code, the EdenGolem v0.1 text-to-graybox-game skill, and a frontend spanning game generation through rendering. The code is published under the Apache 2.0 license.
EdenLive uses a reference stream from a game engine or another source to preserve state and motion while generating appearance; it does not create or edit the underlying 3D mesh. The 24-plus-FPS result is a developer benchmark on a specific configuration, with no independent comparison yet confirmed for image quality, latency or long-duration consistency. No direct workflow for producing printable 3D geometry has been demonstrated.
Key facts
- Developer benchmark reports about 24.7 FPS at 1280 × 704 for the 5B model on an RTX 5090
- Supports 480p and 720p video output
- 5B and 14B model weights and streaming inference code are available
- Code is licensed under Apache 2.0
Context
The release may accelerate visual prototyping for games and 3D scenes, but it remains separate from generating printable 3D geometry.
