EnerVerse-AC
AgiBot · Released May 2025
An action-conditioned robot world model from AgiBot that generates the camera view a robot arm would see while it carries out a given sequence of actions. The released weights were trained only on the open AgiBot World dataset and are licensed for non-commercial use.
- Weights
- Open weightsCC BY-NC-SA 4.0
- Commercial use
- Non-commercial
- Real-time
- No real-time claim
- Where to try
- Run it yourself
01 Controls
What you control
- One start frame from the robot's camera
- An action sequence for the robot
- The camera's extrinsic and intrinsic parameters
02 License
Can you use it commercially?
No. The model card declares CC BY-NC-SA 4.0: no commercial use, and anything you build from the weights must use the same license.
Declared license: CC BY-NC-SA 4.0 · Model card
A reading of the published license text, not legal advice.
03 Speed
Does it run in real time?
No real-time claim. The cited sources make no real-time claim.
Figures are the authors' own, with their settings and hardware. They are not comparable across models.
04 Setup
How to run it
- 01Create a Python 3.10 conda environment and install requirements.txt and PyTorch3D.
- 02Download the single-view checkpoint from Hugging Face and the CLIP weights, and set both paths in the configs.
- 03Prepare a start frame, an action file and camera parameter files; the repository includes a script that converts AgiBot World episodes into this format.
Summarised from the project's own instructions. Follow the source for exact commands and versions.
05 Access
Where to try it
No hosted version was verified. Run the published weights yourself.
Listed alphabetically. Windflow maintains this directory and is one of the hosts; a listing does not promise capacity at any given moment.
06 Questions
EnerVerse-AC FAQ
Is EnerVerse-AC open source?
EnerVerse-AC publishes downloadable weights under CC BY-NC-SA 4.0. Read the license terms before you build on it.
Can I use EnerVerse-AC commercially?
No. The model card declares CC BY-NC-SA 4.0: no commercial use, and anything you build from the weights must use the same license.
Does EnerVerse-AC run in real time?
The cited sources do not claim real-time operation.
Where can I try EnerVerse-AC?
No hosted version was verified; run the published weights yourself.
What does EnerVerse-AC take as input?
You give it one start frame from the robot's camera, an action sequence for the robot and the camera's extrinsic and intrinsic parameters.
How do I run EnerVerse-AC?
Following the official instructions: Create a Python 3.10 conda environment and install requirements.txt and PyTorch3D. Download the single-view checkpoint from Hugging Face and the CLIP weights, and set both paths in the configs. Prepare a start frame, an action file and camera parameter files; the repository includes a script that converts AgiBot World episodes into this format.
What are alternatives to EnerVerse-AC?
Similar models in this directory include 1X World Model (open weights, permissive license), Ctrl-World (open weights, permissive license) and LingBot-World 2.0 (open weights, non-commercial).
Similar world models
All 56 in the directoryRobotics
1X World Model
1X Technologies
Robot action and video modeling
Permissive licenseOpen weightsRobotics
Ctrl-World
Stanford University and Tsinghua University
Keyboard commands converted into robot action chunks
Permissive licenseOpen weightsExplorable world
LingBot-World 2.0
Robbyant (Ant Group)
Interactive world navigation
Non-commercialOpen weightsReal-time claimTry on fal, LingGuang, Reactor, Windflow
Explorable world
Neural World Models
Meta (FAIR)
Navigation action conditioning
Non-commercialOpen weights