jhc13 / taggui

Tag manager and captioner for image datasets
GNU General Public License v3.0
727 stars 34 forks source link

Question: Not sure about how to auto tag, instead of captions #295

Open merket opened 5 hours ago

merket commented 5 hours ago

Hi, I will be training an SDXL lora, so instead of captions in sentenced format, I would like to use tags separated with commas. I read the usage of {tags} but I don't think I fully understand how it works.

How can I make it so that taggui generates tags.

This is my current prompt:

Describe the photo in the following order: Subject's age, hair style, facial hair style, facial features, expression, outfit, accessories, location, background details and the surroundings.

I would like this to be resulted as "in mid 30's, short hair, clean shaved, surprised, hawaian shirt, wooden necklace, outdoor, palm tree, car" etc...

Is this possible to achieve?

jhc13 commented 4 hours ago

The easiest way to generate tags is to use one of the tagger models like SmilingWolf/wd-vit-large-tagger-v3.

MiaoshouAI/Florence-2-base-PromptGen with the default prompt (leave it empty) and Remove tag separators in caption unchecked will also generate tags.

You can try to use one of the other models to generate tags through prompting, but the results will not be consistent.