Metadata-Version: 2.1
Name: captionforge
Version: 0.1.1
Summary: CaptionForge is a local ComfyUI captioning toolkit for preparing better LoRA-ready image datasets. It runs multiple caption witnesses, compares their agreements and contradictions, distills the strongest visual claims, validates them against the image with a VLM, and exports long, short, and taggy TXT captions alongside JSONL audit trails. CaptionForge does not ship model weights: Python-backed Hugging Face models such as Joy and Qwen are downloaded on demand and cached locally, while Ollama-backed stages require a local Ollama installation with the selected model tags installed or pullable.
Home-page: https://github.com/Damkohler/CaptionForge
Requires-Python: >=3.10
Requires-Dist: transformers
Requires-Dist: accelerate
Requires-Dist: huggingface-hub
Requires-Dist: pillow
Requires-Dist: numpy
Requires-Dist: safetensors
Requires-Dist: qwen-vl-utils
