🌦️ WeatherEdit: Controllable Weather Editing with 4D Gaussian Field

Chenghao Qian^1,*, Wenjing Li^1,†, Yuhu Guo², Gustav Markkula¹

A Controllable, Scalable and Efficient Framework for Realistic Weather Editing.

---

• 🎨 Flexible control over weather types (snow, fog, rain)
• 🌡️ Precise weather severity adjustment (light, moderate, heavy)
• 🖼️ Global consistency for multi-view driving scenes (temporal, spatial)

---

⭐ If you like our work or find it useful, please give us a star or cite below. Thanks!

💡 The Quickest Start

You can use our provided pretrained model for the easiest start. If you would like to try the whole pipeline, please go to the subfolder for training instructions

A. General Weather (Takes ~1hr)

Please first configure the environment with conda:

git clone https://github.com/Jumponthemoon/WeatherEdit.git
cd General_Scene
conda env create --file environment.yml
conda activate gaussian_splatting

We provide a complete pipeline to train and render Gaussian scenes with integrated weather effects.

1. Train Your Scene

Download the Dataset Pt.1 from Mip-NeRF 360 and put the garden under data folder, then run:

python train.py -s path/to/data/

After training, the model will be saved under output folder

2. Render with Weather Effects

python render.py -m path/to/model --weather snow --fps 10

🔥 Plug into your GS-based code? 👉 Check it out here

---

B. Driving Scene Editing (Takes ~45mins)

Please first clone the repo and configure the environment with conda:

git clone https://github.com/Jumponthemoon/WeatherEdit.git
cd Driving_Scene

conda create -n weatheredit python=3.9 -y
conda activate weatheredit

pip install -r requirements.txt
pip install git+https://github.com/nerfstudio-project/gsplat.git@v1.3.0
pip install git+https://github.com/facebookresearch/pytorch3d.git
pip install git+https://github.com/NVlabs/nvdiffrast

cd third_party/smplx/
pip install -e .
cd ../..

• Note： if you encounter error ImportError: cannot import name 'cached_download' from 'huggingface_hub', please follow this instruction.

1. Download sample dataset & pretrained model

cd particle_construction

Download sample dataset and pretrained model, then place them in the data and the output directory separately.

2. Render with Weather Effects

Run the script to generate rainy weather in pandaset:

export PYTHONPATH=$(pwd)
python tools/gen_particle.py --resume_from ./output/pandaset/44/checkpoint_final.pth --weather rainy

The rendered video will be saved under ./output/pandaset/44/video_eval folder

---

📌 Citation

@article{qian2026weatheredit,
  title   = {WeatherEdit: Controllable Weather Editing with 4D Gaussian Field},
  author  = {Chenghao Qian and Wenjing Li and Yuhu Guo and Gustav Markkula},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  volume  = {40},
  number  = {10},
  pages   = {8511--8519},
  year    = {2026}
}

---

📬 Contact

For questions, suggestions, or collaborations:

• 📧 tscq@leeds.ac.uk

---

🤝 Acknowledgements

This work builds upon 3D Gaussian Splatting, img2img-turbo and OmniRe. We thank for their amazing works!

Also, thanks for your interest in WeatherEdit! We hope it helps bring new life to your 3D scenes 🌧️🌨️🌫️