ShutterMuse: Capture-Time Photography Guidance with MLLMs

2026-06-24

AI Summary

This work addresses the challenge that existing photographic assistance methods struggle to provide real-time, coordinated guidance on both composition and human pose during image capture. To this end, we introduce the first dual-guidance task tailored for in-the-moment shooting, accompanied by CaptureGuide-Dataset—a large-scale dataset of 130,000 samples—and CaptureGuide-Bench, a dedicated evaluation benchmark. We further develop ShutterMuse, a unified multimodal large language model that jointly optimizes composition refinement and scene-aware pose recommendation through supervised fine-tuning and reinforcement learning. Experimental results demonstrate that ShutterMuse achieves state-of-the-art performance in photographer-level composition guidance, delivers competitive pose recommendations, and significantly reduces inference costs, thereby validating the feasibility of multimodal large language models as intelligent in-camera assistants.

Abstract

Real-world photography requires capture-time guidance for both camera framing and subject pose. Yet existing aesthetic cropping benchmarks mainly evaluate post-hoc crop prediction and overlook subject-side recommendations, leaving the capture-time guidance capabilities of multimodal large language models (MLLMs) underexplored. To address this gap, we introduce CaptureGuide-Bench, a benchmark with two complementary tasks: photographer-side composition decision and refinement, and subject-side scene-conditioned pose recommendation. Our evaluation reveals limitations: general-purpose MLLMs can make composition decisions but lack precise refinement localization, while specialized aesthetic cropping models localize crops effectively but are limited to refinement; neither provides actionable pose guidance. To support model development, we further construct CaptureGuide-Dataset, comprising 130K samples with textual rationales and structured visual annotations, and develop ShutterMuse, a unified MLLM trained with supervised and reinforcement fine-tuning. Experiments on CaptureGuide-Bench show that ShutterMuse achieves the best overall photographer-side performance among evaluated baselines and competitive subject-side pose recommendation with substantially lower inference cost, demonstrating the potential of MLLMs as interactive assistants for photography during image capture.

Problem

• capture-time guidance
• photography
• multimodal large language models
• pose recommendation
• composition refinement

Innovation

• capture-time guidance
• multimodal large language models
• photography benchmark
• pose recommendation
• composition refinement