Hey, thank you for everyone's likes, I just started a buymeacoffee account, since I had some serious hardware damages... It seems training does take a toll on the hardware. Please support~ thank you again o(////▽////)q https://www.buymeacoffee.com/reweik
Note:
For v15:
It feels like that the model can't tell the difference between 3 G units, so just ust "g01" for the best result. if I can find more individual (and sorted) data for the specific character, I'll retain it.
The AptomFrost has the same problem, but it is over fitting rather than not showing up ;D.
Now the Guyver looks more authentic thanks to my computer spending whole night reading the classic manga novel.
Now it must fall in love with the story and characters, and wants to continue the legend.
======================================================================================================================
When using to generate Guyver, you can use full strength.
When generate AptomFrost, use at lower strength, that part of model is overfit...
I've changed trigger word to g00, g01, g02f, g03, aptomf, so it has no conflict with models' key words. Use helmet and bodysuit to enhance your result if the model don't know what to do.
Using at full strength to get Guyver, lower strength to get Guyver's costume.
Using the lora alone without trigger word to get retro colored img.
===========================================================
===========================================================
Childhood obsession, never ending story.
I did try multiple times and got no idea why it is so hard to train...
Maybe just because I don't have a good dataset.
Mainly was trained on the model pictures, some manga covers. Maybe I should try to use the manga pages too?