search

jhc13 / taggui

Tag manager and captioner for image datasets
GNU General Public License v3.0
774 stars 37 forks source link

Feature Request: Token length threshold #195

Closed CognitiveDiffusion closed 4 months ago

CognitiveDiffusion commented 5 months ago

Hey there jhc13,

I am very happy to see your Taggui mentioned more and more in the SD community. It's becoming one of the most well-known tools for captionint from what I can gather.

Since Tags+Natural-language captions are getting more and more popular (and proof to be effective), please add an option to

  1. Search tags (lower right UI menu)
  2. Search & replace text (Ctrl+R)
  3. Delete tags

by token threshold.

Ideally there would be sliders (or an input field) to specify the token length and if the threshold is more or less tokens.

That way we could for example easily delete all CogVLM captions (usually 50 tokens or more) if it turns out the prompt or settings were bad. Another example would be that we could change words in shorter tags (such as "green hair", "medium hair" or even manual tags like "holding a sword in his left hand"), while leaving longer CogVLM (& Co.) tags alone.

When it comes to searching for tags filtering out the individual long captions would help immensely as well.

jhc13 commented 5 months ago

Have you tried sorting the tags by length (Sort by -> Length)? I think this will solve some of your problems.