---
license: apache-2.0
pipeline_tag: depth-estimation
library_name: diffusers
---

# NormalCrafter: Learning Temporally Consistent Normals from Video Diffusion Priors

[NormalCrafter](https://normalcrafter.github.io/) generates temporally consistent normal sequences with fine-grained details from open-world videos of arbitrary lengths.  This model is based on the paper [NormalCrafter: Learning Temporally Consistent Normals from Video Diffusion Priors](https://huggingface.co/papers/2504.11427).


## 🚀 Quick Start

### 🤖 Gradio Demo
- Online demo: [NormalCrafter](https://huggingface.co/spaces/Yanrui95/NormalCrafter) 
- Local demo:
    ```bash
    gradio app.py
    ``` 

### 🛠️ Installation
1. Clone this repo:
```bash
git clone git@github.com:Binyr/NormalCrafter.git
```
2. Install dependencies (please refer to [requirements.txt](requirements.txt)):
```bash
pip install -r requirements.txt
```



### 🤗 Model Zoo
[NormalCrafter](https://huggingface.co/Yanrui95/NormalCrafter) is available in the Hugging Face Model Hub.

### 🏃‍♂️ Inference
#### 1. High-resolution inference, requires a GPU with ~20GB memory for 1024x576 resolution:
```bash
python run.py  --video-path examples/example_01.mp4
```

#### 2. Low-resolution inference requires a GPU with ~6GB memory for 512x256 resolution:
```bash
python run.py  --video-path examples/example_01.mp4 --max-res 512
```