Overview
A newer, more accurate image describer for generating detailed captions — mostly to caption datasets for LoRA training and other AI uses. It batch-describes a folder and writes the results as one sidecar text file per image (the layout LoRA trainers expect), one combined file with a description per line, or both at once; any string can be prepended or appended to every caption.
How It Works
Two models sit over one engine. Florence-2 (the default) actually generates a sentence describing the image — real prose captions. CLIP is a scorer, not a captioner: it ranks the image against a list of candidate labels you supply and keeps the top few, so it’s the right tool for consistent tags from a controlled vocabulary but can only output words already on your list. A Tkinter GUI and a command-line tool share the same engine.
Current Status
Working, with both the GUI and CLI front-ends over the Florence-2 / CLIP engine.