Pose-ICL: 3D-Aware In-Context Learning for Pose-Controllable Subject Customization

2026-06-09

AI Summary

Existing image generation methods struggle to simultaneously achieve accurate pose control and consistent appearance across diverse poses in subject customization. To address this challenge, this work proposes Pose-ICL, a tuning-free framework that leverages 3D-aware in-context learning to efficiently adapt to novel subjects using only a few image-pose pairs. The key innovation lies in the introduction of Surface-Anchored Positional Encoding (SAPE), which explicitly anchors image tokens to coordinates on the surface of a 3D bounding box, thereby endowing the DiT architecture with native 3D awareness. Experiments demonstrate that Pose-ICL significantly outperforms current state-of-the-art methods on both 3D assets and real-world subjects, achieving notable advances in pose fidelity and identity consistency.

Abstract

Subject Customization is a foundational task in modern image generation. By providing a few reference images and a text prompt, users can generate images of a specific object in any desired scene. However, existing methods still struggle to achieve effective pose control for customized subjects. In practice, they often exhibit inaccurate poses or inconsistent cross-pose appearances. These limitations suggest that understanding objects in a volumetric manner remains a significant challenge for 2D-native backbones. To address this challenge, we propose Pose-ICL, a tuning-free framework that leverages 3D-aware In-Context Learning (ICL) to directly adapt to new subjects through multiple paired image-pose references. Its core mechanism,Surface-Anchored Position Embedding (SAPE), equips the model with explicit 3D awareness by anchoring image tokens to the surface coordinates of a volumetric bounding box. Dedicated refinements ensure its seamless compatibility with existing DiT models. Extensive evaluations on both 3D assets and real-world subjects demonstrate that Pose-ICL significantly outperforms current methods in both pose accuracy and identity consistency.

Problem

• pose control
• subject customization
• 3D-awareness
• identity consistency
• image generation

Innovation

• 3D-aware
• In-Context Learning
• Pose Control
• Subject Customization
• Surface-Anchored Position Embedding