Look, Compare and Draw: Differential Query Transformer for Automatic Oil Painting

[Project Page] [Paper]

This work has been accepted by IEEE Transactions on Visualization and Computer Graphics, 2026.

Abstract

This work introduces a new approach to automatic oil painting that emphasizes the creation of dynamic and expressive brushstrokes. A pivotal challenge lies in mitigating the duplicate and common-place strokes, which often lead to less aesthetic outcomes. Inspired by the human painting process, i.e., observing, comparing, and drawing, we incorporate differential image analysis into a neural oil painting model, allowing the model to effectively concentrate on the incremental impact of successive brushstrokes. To operationalize this concept, we propose the Differential Query Transformer (DQ-Transformer), a new architecture that leverages differentially derived image representations enriched with positional encoding to guide the stroke prediction process. This integration enables the model to maintain heightened sensitivity to local details, resulting in more refined and nuanced stroke generation. Furthermore, we incorporate adversarial training into our framework, enhancing the accuracy of stroke prediction and thereby improving the overall realism and fidelity of the synthesized paintings. Extensive qualitative evaluations, complemented by a controlled user study, validate that our DQ-Transformer surpasses existing methods in both visual realism and artistic authenticity, typically achieving these results with fewer strokes. The stroke-by-stroke painting animations are available on our project website.

Prerequisites

• Linux or macOS
• Python 3.9
• PyTorch 1.7+ and other dependencies (torchvision, visdom, dominate, and other common python libs)

Training

  cd train
  bash my_train.sh

• models would be saved at checkpoints/painter folder.

Inference

  cd inference
  python inference.py

Citation

If you find this code helpful for your research, please cite:

@article{liu2026look,
author = "Liu, Lingyu and Wang, Yaxiong and Zhu, Li and Liao, Lizi and Zheng, Zhedong",
title = "Look, Compare and Draw: Differential Query Transformer for Automatic Oil Painting",
journal = "TVCG",
code = "https://differential-query-painter.github.io/DQ-painter/",
year = "2026" }

Acknowledgments

This repository is benefit from Paint Transformer. Thanks for the open-sourcing work! We would also like to thank to the great projects in Compositional Neural Painter.