Vision-Language Models with In-Context Learning for In-Camera Portrait Composition Guidance

Duotun Wang, Rui Cao, Kunyu Feng, Wanling Li, Qian Zhong, Zhijing Shao, Yue Jiang, Zhiming Hu, Mingming Fan, Zeyu Wang*

The 39th Annual ACM Symposium on User Interface Software and Technology (UIST Adjunct ’26), 2026.

---

---

Abstract

Capturing aesthetically pleasing portraits requires specialized compositional knowledge that ordinary users often lack, making it difficult for them to achieve optimal camera framing. While vision-language models (VLMs) excel at general image understanding, they frequently produce overly positive feedback and lack the granular structural awareness needed to generate actionable compositional guidance. To address this, we present a visual in-context learning (ICL) framework that equips VLMs with augmented aesthetic evaluation capabilities. By retrieving structurally similar, professionally reframed image-composition analyses as visual prompts, our framework grounds VLM reasoning in established aesthetic standards, such as the rule of thirds and leading lines, and enables mapping compositional flaws to specific camera instructions.

---

Links

---

BibTeX