LLM Captioning workflow

Use ollama or koboldccp, a vision model that is not censored, good the Pixtral model. In the ollama generate node, edit the prompt if necessary. In the image loading node, specify the folder from which to load the images and set the seed to 0. If there are many images, you can stop the infinite generation and save the workflow, then continue later. In the save text node, specify the path where to save the text files. You can disable overwrite and run it again, adding, for example, danboru tags to the text document after the main prompt. Be careful with the context size. 2000 is not enough.