Double captioning WD Tagger + LLM model

[BenDes21]

Hi there, this is a very specific feature and I don't know how it's can be easily implement into the UI but Mr Lab find a very good way to get more accurate captions with a low level of censorship. The process is pretty simple :

• Caption an image with WD-Tagger V3 ( it's will give a list of tags according the image like : " girl, bedroom, pink dress, lying, smiling " )
• Set theses tags into a variable like img_xxx_tags or directly into the caption file .txt ( img_xxx.txt ) link to this image like the current process of image interrogator
• Caption a second time with another LLM model ( more accurate like Cog VLM ) and write the prompt to describe the image by using the tags of the variable img_xxx_tags or by using the tags of the caption file ( like " make a short description of the images by using {img_xxx_tags} " OR " make a short description of the images by using {img_xxx.txt} " )

Im sorry if Im not clear but it's basically using the tags ( list of keywords ) that the wd-tagger v3 detect and use them for make a more accurate description with another llm model. Let me know if you need more infos