lora-prepare-dataset

What problem does it solve?

It prevents wasted LoRA training runs by checking whether a dataset of training images meets quality, consistency, and suitability requirements before you spend compute.

Core Features & Use Cases

• Dataset gatekeeping: Verifies minimum image counts per LoRA class and produces a structured pass, warning, or fail verdict.
• Per-image quality checks: Screens for decodeable formats, minimum resolution, cropping/subject visibility, exposure/blur, and data consistency heuristics.
• Duplicate and coverage detection: Flags exact and near-duplicates and reports caption file coverage so downstream captioning or training can proceed safely.

Quick Start

Validate a dataset by providing the absolute image folder path, the LoRA class, the intended subject text, and an absolute output folder for the generated prepare_report.json.