Summary

Pianke is a local, privacy-focused photo selection tool designed for photographers. It automates the grouping of similar images from a shoot and offers an AI-assisted A/B comparison interface for rapid selection, significantly reducing post-production time while maintaining user aesthetic control.

How It Works

Pianke provides three operational modes: 'Fast' utilizing local image comparison for low-spec devices; 'Expert' leveraging local AI models (DINOv2, face recognition, EXIF) for precise grouping and quality screening; and 'Tycoon' which integrates remote multimodal LLMs for advanced analysis and natural language explanations. Core functionality includes intelligent rejection of technical flaws (blur, exposure issues), multi-signal grouping (visual semantics, facial features, metadata), and a keyboard-driven A/B "PK" selection interface. It natively supports RAW files by analyzing embedded previews and intelligently pairs RAW+JPG. Progress is auto-saved with multi-level undo capabilities.

Quick Start & Requirements

Installation is streamlined via provided one-click launch scripts (.command/.bat) which bundle a self-contained Python environment, or manually by cloning the repository, creating a virtual environment (python3 -m venv .venv, source .venv/bin/activate), and running pip install -r requirements.txt. The 'Expert' mode requires downloading approximately 600MB of AI models. Domestic users are strongly advised to configure alternative PyPI mirrors (e.g., Aliyun, Tsinghua) for faster dependency installation. The application is accessed via http://localhost:5057.

Highlighted Details

• Advanced local AI grouping using DINOv2 visual semantics and facial recognition.
• Native RAW file support, including efficient embedded preview analysis.
• Optional integration with large multimodal models (e.g., GPT-4V, Qwen) for detailed, natural language feedback on photo quality.
• Batch watermarking feature that embeds EXIF data and camera logos onto selected images.

Maintenance & Community

Project feedback and suggestions are primarily handled via GitHub Issues. No other community channels (e.g., Discord, Slack), active contributor information, or sponsorship details are provided in the README.

Licensing & Compatibility

The project's license is not specified in the provided README. Consequently, its suitability for commercial use, integration into proprietary software, or specific compatibility constraints remain undetermined.

Limitations & Caveats

HEIC image previews may not render correctly on Windows without additional system extensions, although backend processing is unaffected. The 'Tycoon' mode, while offering advanced AI explanations, transmits image thumbnails to third-party LLM services, posing a potential privacy concern. The absence of a stated license is a significant adoption blocker.