Comfyui_CXH_joy_caption

Why we included this project

If you train image models in ComfyUI, you know the most tedious step is writing consistent captions for hundreds of reference images. This custom node pack bundles three captioning engines, Joy_caption, MiniCPMv2_6, and Florence2, into ready-made workflows so you can generate descriptive tags in bulk instead of hand-labeling each file. It handles the practical details that matter for dataset prep: batch processing folders, applying a trigger word to every image in a directory, and even classifying images into groups. The README documents real throughput numbers (roughly 4 to 5 seconds per image on a 4090) and walks through model placement, so setup is mostly a matter of downloading weights into the right folders. Teams building LoRA training sets or organizing large image collections will find it a straightforward way to turn raw folders into captioned, structured data.