Visionary: Open-Source macOS App for Building Image Training Datasets Locally
RADIO02118 · reddit · 2026-07-23
Visionary is a native macOS app designed to streamline image dataset preparation and curation. It integrates deduplication, grouping, tagging, and exporting, running fully offline.
Key Features
- Grouping & Dedupe: Groups by near-duplicate, color, person, or resolution, marking non-highest-res duplicates.
- Tag Management: Rename, merge variants, and flag rare tags in bulk.
- Captioning: Integrates WD taggers, JoyCaption, Moondream, and Qwen2.5-VL with configurable prompt grammar.
- Export: Supports formats for kohya, ai-toolkit, and diffusers.
Performance
Handles 100,000 images at 120 fps with 1.5 GB memory. Face detection and feature extraction leverage the Apple Neural Engine, taking about 11 minutes for perceptual features.
More from Apps
- Tesla says FSD is driving demand as French carmakers lobby to block approval — mitchdeg · 2026-07-23
- Open-source flat-animation Skill turns one topic into an explainer video — tinyfool · 2026-07-23
- Google CodeLab shows how to try GenUI on Flutter — davemccollough · 2026-07-23
- Coco adds a hide-avatar mode so the AI only chimes in when needed — EchoShao8899 · 2026-07-23
- Google adds Gemini to Slides so users can generate fully editable decks in minutes — Scobleizer · 2026-07-23
- TwelveLabs shows a video memory layer that stores moments, entities and themes — AI Engineer · 2026-07-23