Type: Software
Status: Beta
Tech Stack: Python, FastAPI, Uvicorn, Transformers (Florence-2, JoyCaption), PyTorch, React, Vite, TypeScript
Problem Statement
Training AI image models (LoRA, DreamBooth, fine-tuning) requires image datasets with high-quality text descriptions. Manual captioning is extremely time-consuming, and automatic tools rarely offer a visual gallery view, multi-model support, or the ability to switch between caption formats (full text, tags, short description). Batch processing with subsequent manual correction in an integrated workflow is missing.
Description
A lean tool for creating image descriptions for AI training datasets. Images are displayed in a React gallery, captions saved as same-name .txt files. Supports auto-captioning via Florence-2 and JoyCaption (NSFW-capable) with selectable format (full text, short text, tags). Batch save all captions with one click.
Use Case
Tag images with descriptions so AI models can learn from them.
Link: https://github.com/rawk7000/captioner (private repo)