PixelSmile

Toward Fine-Grained Facial Expression Editing

Continuous, linearly controllable, and identity-preserving facial expression editing.

Jiabin Hua1,2,* Hengyuan Xu1,2,* Aojie Li2,† Wei Cheng2 Gang Yu2,‡ Xingjun Ma1,‡ Yu-Gang Jiang1
1Fudan University 2StepFun
*Equal contribution †Project lead ‡Corresponding authors
arXiv arXiv GitHub Code HuggingFace Model HuggingFace Bench HuggingFace Demo

Expression Editing

PixelSmile enables fine-grained editing over 12 facial expressions with continuous and approximately linear intensity control, while preserving identity consistency across diverse portrait inputs.

Expression Blending

Beyond single-expression editing, PixelSmile supports zero-shot pairwise blending, revealing a continuous and compositional expression manifold that can synthesize plausible compound expressions.

FFE-Bench

FFE-Bench evaluates facial expression editing from four complementary aspects, including structural confusion, editing accuracy, linear controllability, and the trade-off between expression transfer and identity preservation.

FFE Dataset

Flex Facial Expression (FFE) is a 60K-image cross-domain dataset with same-identity expression variations and continuous affective annotations, covering both real-world and anime portraits for fine-grained controllable editing.

Citation

If you find PixelSmile useful in your research or applications, please consider citing our work.

@article{hua2026pixelsmile,
  title={PixelSmile: Toward Fine-Grained Facial Expression Editing},
  author={Jiabin Hua and Hengyuan Xu and Aojie Li and Wei Cheng and Gang Yu and Xingjun Ma and Yu-Gang Jiang},
  journal={arXiv preprint arXiv:2603.25728},
  year={2026}
}