# EdenLive

EdenLive is a generative renderer: you feed it a rough source video, such as game footage or a live camera, plus a scene prompt and a reference image, and it re-renders each chunk as a new world while keeping the source motion. Movement comes from the source video, not from direct keyboard commands.

Organization: VAST. Released: 12 October 2026. Kind: Live video.

## What you control

- A source video: game footage or a live camera
- A scene prompt
- A reference image for appearance

## License

Weights: Available (https://huggingface.co/VAST-AI/EdenLive). Declared license: Mixed component licenses.
Only under its own license (Mixed component licenses). The video pipeline's weights are Apache-2.0, with an MIT CLIP encoder and a vision encoder under Meta's DINOv3 License. The optional first-frame workbench uses FLUX.2 [klein] under a non-commercial license.

## Real-time

Real-time, per authors. 24 fps or higher at 480p or 720p, minute-scale generation.

## How to run it

1. Clone VAST-AI-Research/EdenLive, create a Python 3.10 conda environment, and install PyTorch and the pinned packages from the README.
2. Download VAST-AI/EdenLive from Hugging Face.
3. For offline clips, run python -m edenlive.scripts.infer.
4. For the live service, start the TURN relay (bash turn_server.sh), then the model service (bash start.sh 14b, or 5b).

## Where to try

- [Reactor](https://www.reactor.inc/models/eden-live)
- [Windflow](https://windflow.online/models/edenlive)

## Sources

- [Project source](https://github.com/VAST-AI-Research/EdenLive)
- [Weights and model card](https://huggingface.co/VAST-AI/EdenLive)

## Questions

### Is EdenLive open source?

EdenLive publishes downloadable weights under Mixed component licenses. Read the license terms before you build on it.

### Can I use EdenLive commercially?

Only under its own license (Mixed component licenses). The video pipeline's weights are Apache-2.0, with an MIT CLIP encoder and a vision encoder under Meta's DINOv3 License. The optional first-frame workbench uses FLUX.2 [klein] under a non-commercial license.

### Does EdenLive run in real time?

Yes, according to its authors. 24 fps or higher at 480p or 720p, minute-scale generation. This directory has not measured it independently.

### Where can I try EdenLive?

You can try it on Reactor and Windflow.

### What does EdenLive take as input?

You give it a source video: game footage or a live camera, a scene prompt and a reference image for appearance.

### How do I run EdenLive?

Following the official instructions: Clone VAST-AI-Research/EdenLive, create a Python 3.10 conda environment, and install PyTorch and the pinned packages from the README. Download VAST-AI/EdenLive from Hugging Face. For offline clips, run python -m edenlive.scripts.infer. For the live service, start the TURN relay (bash turn_server.sh), then the model service (bash start.sh 14b, or 5b).

### What are alternatives to EdenLive?

Similar models in this directory include Helios (open weights, permissive license), Krea Realtime 14B (open weights, permissive license) and LongLive 2.0 (open weights, custom terms).

Checked 2026-10-23. Corrections: https://github.com/windflow-labs/world-model-directory