---
title: Kairos World Action Model
emoji: 🤖
colorFrom: green
colorTo: gray
sdk: gradio
sdk_version: 6.15.1
app_file: app.py
short_description: Kairos 4B world-action model for robot video generation
python_version: "3.12"
startup_duration_timeout: 1h
---

# Kairos World Action Model

This Space demonstrates [Kairos](https://github.com/kairos-agi/kairos-sensenova), a 4B-parameter native world-action model for physical AI. Given an initial scene image and a task instruction, Kairos predicts a future video showing the robot performing the described action.

## Model

- **DiT**: `kairos-agi/Kairos3.1-4B-robot-480P` — the Kairos 4B world-action model
- **Text Encoder**: `Qwen/Qwen3.5-2B` — encodes the task instruction
- **VAE**: `Wan-AI/Wan2.1-T2V-1.3B` (Wan2.1 VAE) — encodes/decodes video latents

## Usage

Upload an initial scene image (e.g. a robot arm in a workspace) and describe the task. The model generates a short video at 480×832 resolution showing the predicted future action.

## References

- [Paper](https://arxiv.org/abs/2606.16533)
- [GitHub](https://github.com/kairos-agi/kairos-sensenova)
- [Model on Hugging Face](https://huggingface.co/kairos-agi/Kairos3.1-4B-robot-480P)