| | --- |
| | license: cc-by-2.0 |
| | pipeline_tag: text-to-image |
| | --- |
| | |
| | # ShadowDraw: From Any Object to Shadow–Drawing Compositional Art |
| |
|
| | This repository contains the models and code for **ShadowDraw**, a framework that transforms ordinary 3D objects into shadow-drawing compositional art. Given a 3D object, the system predicts scene parameters, including object pose and lighting, together with a partial line drawing, such that the cast shadow completes the drawing into a recognizable image. |
| |
|
| | - **Paper**: [ShadowDraw: From Any Object to Shadow-Drawing Compositional Art](https://huggingface.co/papers/2512.05110) |
| | - **Project Page**: https://red-fairy.github.io/ShadowDraw/ |
| | - **Code**: https://github.com/Red-Fairy/ShadowDraw |
| |
|
| | <div align="center"> |
| | <img src="https://github.com/Red-Fairy/ShadowDraw/raw/main/assets/teaser.gif" alt="ShadowDraw Teaser" width="600"/> |
| | </div> |
| |
|
| | ## 🔧 Installation |
| |
|
| | First, create a conda environment and install the dependencies: |
| |
|
| | ```bash |
| | git clone https://github.com/red-fairy/ShadowDraw.git |
| | cd ShadowDraw |
| | conda env create -f environment.yml |
| | ``` |
| |
|
| | Then, install Blender 4.3.2 following the instructions [here](https://download.blender.org/release/Blender4.3/), unzip it, and specify the path to the Blender executable in the `blender_path` argument of the `main.py` script. |
| |
|
| | ## 🎨 Generate Your Own Shadow-Drawing Art |
| |
|
| | ### Shadow-Drawing Art from a Single Object |
| |
|
| | First, download the LoRA weights for the line drawing generation model from HuggingFace: |
| | ```bash |
| | huggingface-cli download RedFairy/Flux-ShadowDraw-LoRA --local-dir ./checkpoints |
| | ``` |
| |
|
| | Then, prepare your 3D object file in `.obj`, `.glb`, or `.ply` format, and run the following command: |
| |
|
| | ```bash |
| | python scripts/launch.py --object_filepaths PATH_TO_OBJECT_FILE \ |
| | --save_name SAVE_NAME \ |
| | --output_root OUTPUT_ROOT \ |
| | --optimize_object_params |
| | ``` |
| |
|
| | To accelerate the generation process, you can remove `--optimize_object_params` and instead add `--sample_distribution`. This will fix the azimuths and fit a distribution of object internal rotations with respect to fractal dimension and sample internal rotations accordingly. |
| |
|
| | To add gravity to the object, add `--use_gravity`. If specified, the object will be first animated by gravity to find a stable pose and then rendered. |
| |
|
| | ### Shadow-Drawing Art from Multiple Objects |
| |
|
| | For multi-object compositions, you can use the following command: |
| |
|
| | ```bash |
| | python scripts/launch.py --object_filepaths PATH_TO_OBJECT_FILES \ |
| | --save_name SAVE_NAME \ |
| | --output_root OUTPUT_ROOT \ |
| | --sample_distribution |
| | ``` |
| |
|
| | Here, `PATH_TO_OBJECT_FILES` should be a space-separated list of `.obj`, `.glb`, or `.ply` files. |
| |
|
| | For user-specified subject, add `--user_character CHARACTER_NAME` and `--system_prompt_path system_prompts/prompt_proposal_user.txt`. |
| |
|
| | ## 📊 Citation |
| | If you find this work useful, please consider citing: |
| |
|
| | ```bibtex |
| | @article{luo2025shadowdraw, |
| | title={ShadowDraw: From Any Object to Shadow–Drawing Compositional Art}, |
| | author={Luo, Rundong and Snavely, Noah and Ma, Wei-Chiu}, |
| | journal={arXiv preprint arXiv:2512.05110}, |
| | year={2025} |
| | } |
| | ``` |