---
title: UniAction
canonical_url: "https://www.modelscope.cn/models/AI-ModelScope/UniAction"
md_url: "https://www.modelscope.cn/models/AI-ModelScope/UniAction.md"
repository: AI-ModelScope/UniAction
last_updated: 2026-10-26
license: apache-2.0
pipeline_tag: image-to-video
tasks:
  - image-to-video
base_model:
  - Wan-AI/Wan2.2-I2V-A14B
base_model_relation: finetune
parameters: 1.0B
tensor_type:
  - F32
  - BF16
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
language:
  - en
downloads: 9
stars: 2
tags:
  - uniaction
  - video-generation
  - action-conditioning
---

# UniAction

> UniAction - AI-ModelScope 在 ModelScope 开源的模型。UniAction: Unified Action Conditioning for World Models

AI-ModelScope/UniAction 是 ModelScope 魔搭社区上的 1.0B 参数image-to-video模型，采用 apache-2.0 许可，基于 Wan-AI/Wan2.2-I2V-A14B 构建。

- **Repository**: AI-ModelScope/UniAction
- **License**: apache-2.0
- **Tasks**: image-to-video
- **Parameters**: 1.0B
- **Base model**: Wan-AI/Wan2.2-I2V-A14B
- **Tags**: uniaction, video-generation, action-conditioning
- **Downloads**: 9
- **Stars**: 2
- **Last updated**: 2026-10-26

Source: https://www.modelscope.cn/models/AI-ModelScope/UniAction

---

# UniAction: Unified Action Conditioning for World Models

[Paper](https://arxiv.org/abs/2610.32977) · [Project Page](https://uniaction.github.io/) · [Code](https://github.com/KBRASK/UniAction) · [Training set](https://huggingface.co/datasets/BowenXue/UniAction-Train) · [Evaluation set](https://huggingface.co/datasets/BowenXue/UniAction-Eval)

UniAction conditions a shared video world model on force, camera control, hand pose and robot pose.

Download `stage1.safetensors`, `stage2.safetensors` and `release.json` together, and follow the [inference example](https://github.com/KBRASK/UniAction#inference) in the code repository. The official [Wan2.2-I2V-A14B backbone](https://huggingface.co/Wan-AI/Wan2.2-I2V-A14B) is downloaded separately.

```bibtex
@article{xue2026uniaction,
  title={UniAction: Unified Action Conditioning for World Models},
  author={Xue, Bowen and Li, Zizhang and Yu, Hong-Xing and Wu, Jiajun},
  journal={arXiv preprint arXiv:2610.32977},
  year={2026}
}
```

The UniAction adapters are released under Apache License 2.0. The Wan backbone retains its original license. Training sources include EgoDex, Goal Force, DROID and SolarWM; source materials retain their respective terms.