search

Nerogar / OneTrainer

OneTrainer is a one-stop solution for all your stable diffusion training needs.
GNU Affero General Public License v3.0
1.46k stars 120 forks source link

[Feat]: IPTC: Keywords field for tags #374

Open yggdrasil75 opened 2 weeks ago

yggdrasil75 commented 2 weeks ago

Describe your use-case.

I am kinda tired of having to deal with arbitrary text sidecar files as I cant edit those from a standard image organizational tool (ie: digikam or xnview) but the iptc keywords field I can. enabling use of that field would also make it easier to just reuse the face tags as well (which appear as iptc regions often)

What would you like to see as a solution?

Use iptc keywords for tags, as well as named regions, if a named region is entirely (or mostly) hidden behind a mask then drop that tag when the mask is applied.

Have you considered alternatives? List them here.

EXIF description field can also be used or various xmp fields, but iptc would probably be the most reasonable for compatibility with free software.

O-J1 commented 2 weeks ago

Not the solution but have you considered using this for managing and editing your captions:

https://github.com/Particle1904/DatasetHelpers

yggdrasil75 commented 2 weeks ago

I use this for automated tagging: https://github.com/jhc13/taggui but I use digikam for general usage and faces in particular. its just that I want to add to taggui an option to embed the keywords into the file and add yolov10 or some other segmentation model support so it finds regions and embeds those. but for that to be useful in ai (and approved by the dev of taggui) then it would need at least 1 trainer to use it. I use this trainer so I was hoping it would. with embedding the tags in the image either as iptc/exif or xmp, then I can use the same data in digikam and onetrainer instead of having to work with 2 completely separate sets of data and either manually converting, writing a script to convert and periodically running it, or just leaving the data separate.