ComfyUI_SLK_joy_caption_two

Why we included this project

For anyone building image datasets inside ComfyUI, this node handles the captioning step without forcing you to switch tools. It wraps the JoyCaptionAlpha Two model, which pairs a SigLIP vision encoder with a Llama 3.1 instruct model, so captions come out as natural descriptive sentences rather than bare keyword tags. The batch node is the part most people will care about: it processes whole folders of images, lets you prepend or append text to every caption (useful for injecting trigger words during training), and exposes top_p and temperature for sampling control. Setup takes a few manual steps because you have to download the JoyCaption model and a compatible LLM into the right ComfyUI model folders, but the README walks through each path. That makes it a solid fit for dataset builders who want captioning tightly integrated into their existing ComfyUI pipeline.