Time to share my own private(not commision ones) LoRAs of SD1.5(cuz now i go to ponyXL and don't continue use this ones).
Donation link: https://boosty.to/blessedworld/donate
Zip file is triggerwords for using in wildcards
Description
FAQ
Comments (8)
The Implementation is awesome. Can you share your skills how to train?
this is for each character: wd14 tagged+TW on images, as much images as you can get from title(balanced repeats to each others character) ~45% + TW only tagged best images of characters(30 if character have) ~55%(repeats).
It not best way, but it "lazy" cuz you don't need to much manual work(if you have screenshots already).
Answering only because this one is made on https://huggingface.co/BangumiBase dataset from here,
it not my true pipeline for making best result, ReZero and Monogatari (and many others(most of uploaded) i crop manually and do some additional things to increase quality)
@randomizer89 Can you explain tw is summerizing?
@isibalrom trigger word is just char name (here for ochako it char name + one of most frequent clothess, but most times i don't do it cuz it not correct)
@randomizer89 sorry for jumping in on the questions but i'd like to hear your tactic with trigger words.
I usually remove all tags that are linked to the character and i want to be associated with it.
Like Orange Hair, Green Eyes, Scar on right eye etc.
And then replace them by having the character name as a token. You mentioned that you use wd14 tagged images and the trigger word, does that mean that you add the trigger word to the images without removing any of the character traits?
I know this is always a topic people fight over depending on preference, but i really like the results you're achieving with your lora's so i'd like to know how you do it :)
Thanks in advance for a reply!
@vslinx true way(partly) is with remove char tags like hair color/hair style/eye color/breast size and add front/side/from behind, but here i don't do this step(if you have file of all this type of tags it be helpfull if you send PM me with it(plan to made it soon, but lazy, at now i just use BooruDatasetTagManager for it))
P.S. here ~same thing do this ~55% of high quality images of char tagged by just trigger word
Could you make an SDXL version of this lora?
it hard for me.
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.