---
language:
- en
- zh
inference: false
license: other
license_name: nanoavatar-model-licenses
license_link: LICENSE
tags:
  - nanoavatar
  - talking-head
  - lip-sync
  - audio-driven
  - audio-to-video
  - real-time
  - streaming
---

<div align="center">
<h1>📱 NanoAvatar</h1>
<h3>Changing how often avatar applications can be used and how many users they can serve</h3>
<p>📱 <strong>2023 mobile chipset · 41 FPS</strong> · ⚡ <strong>103 ms first-frame latency</strong></p>
<p>
<a href="https://github.com/wpydcr/NanoAvatar/releases"><strong>📱 Android downloads (Full / Lite)</strong></a> ·
<a href="https://github.com/wpydcr/NanoAvatar"><strong>💻 Code and usage</strong></a>
</p>
<p>
<a href="#demo">🎬 Demos</a> ·
<a href="#performance">📊 Benchmarks</a> ·
English · <a href="https://huggingface.co/wpydcr/NanoAvatar/blob/main/README.zh-CN.md">简体中文</a>
</p>
</div>

**Generate high-fidelity talking-avatar videos directly on your phone, without a cloud GPU.**

<a id="demo"></a>
## 🎬 Demos

> **Generalization demo:** The examples below were not used to train the model, nor was it specifically fine-tuned for them. 

<table>
  <tr>
    <th align="center">English demo</th>
    <th align="center">Chinese demo</th>
  </tr>
  <tr>
    <td align="center" width="50%">
      <video controls playsinline preload="metadata" width="280" style="max-width: 100%; height: auto;" src="https://huggingface.co/wpydcr/NanoAvatar/resolve/main/assets/demo-en.mp4" aria-label="English demo"></video>
    </td>
    <td align="center" width="50%">
      <video controls playsinline preload="metadata" width="280" style="max-width: 100%; height: auto;" src="https://huggingface.co/wpydcr/NanoAvatar/resolve/main/assets/demo-zh.mp4" aria-label="Chinese demo"></video>
    </td>
  </tr>
</table>

**Try the on-device avatar generation model offline.** The app includes the models and an avatar. Record your voice to animate it. [Android guide](https://github.com/wpydcr/NanoAvatar#run-on-android).

<a id="performance"></a>
## 📊 Performance

| Model | Device | Model FPS | First frame | Memory / VRAM estimate |
| --- | --- | ---: | ---: | ---: |
| 📱 NanoAvatar | **2023 · Snapdragon 8 Gen 3**<br>Android 14 | **39 FPS** | **115 ms** | **834 MiB** |
| 📱 NanoAvatar Lite | **2023 · Snapdragon 8 Gen 3**<br>Android 14 | **41 FPS** | **103 ms** | **700 MiB** |
| 📱 NanoAvatar Lite | **2021 · Snapdragon 8 Gen 1**<br>Android 15 | **18 FPS** | **183 ms** | **693 MiB** |
| ⚡ Quantized | **RTX 4090**<br>Windows CUDA | **333 FPS** | **18 ms** | **834 MiB** |
| 🖥️ Full precision | **RTX 4090**<br>Windows CUDA | **224 FPS** | **37 ms** | **1119 MiB** |

Android first-frame timing runs from audio feature extraction to the first generated image.

The model generates a **256 × 256 face region**, composited at the source video's original resolution.

🌊 **Streaming generation: start speaking as audio arrives.** With a streaming LLM and streaming TTS, NanoAvatar starts speaking in **about 0.3 seconds in our tests**, without waiting for the complete audio.

<a id="models"></a>
## 📦 Choose and download a model

| Model | Files | Contents |
| --- | --- | --- |
| ⚡ Quantized | [`quantized/`](https://huggingface.co/wpydcr/NanoAvatar/tree/main/quantized) | PyTorch/CUDA: HuBERT W8A16, mixed INT8 lip-sync model and CUDA DLL |
| 🖥️ Full precision | `full-precision/`（come soon） | PyTorch: HuBERT FP16 and lip-sync model FP32 |
| 📱 Android compiled | [`android-qnn/`](https://huggingface.co/wpydcr/NanoAvatar/tree/main/android-qnn) | CPU: original FP32 face/audio encoders. QNN: HuBERT W8A16 and HTP FP16 generator |

The quantized CUDA runtime targets **Windows CUDA SM89 (RTX 4090)**. The Android package targets **Snapdragon 8 Gen 3 / HTP v75**.

Install the Hugging Face CLI, then download the package you need:

```shell
python -m pip install huggingface_hub
```

### ⚡ Quantized

```shell
hf download wpydcr/NanoAvatar --include "quantized/*" --local-dir models
```

### 🖥️ Full precision

```shell
hf download wpydcr/NanoAvatar --include "full-precision/*" --local-dir models
```

### 📱 Android compiled

```shell
hf download wpydcr/NanoAvatar --include "android-qnn/*" --local-dir models
```

Keep the complete directory structure when downloading.

Running the models and integrating them: [NanoAvatar on GitHub](https://github.com/wpydcr/NanoAvatar).

## License

[chinese-hubert-large](https://huggingface.co/TencentGameMate/chinese-hubert-large) and its quantized and compiled versions retain the upstream MIT license.

Our lip-sync model and its quantized and compiled versions use [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). Academic research and other noncommercial uses are permitted, subject to attribution and the other license terms.

**Commercial licensing:** [wupingyu@mail.ustc.edu.cn](mailto:wupingyu@mail.ustc.edu.cn).

See [LICENSE](LICENSE) for the full terms.