v1.1 - slightly less concept and style bleed.
A metal yoke that restrains the arms of the person wearing it. Just prompt it with wearing a metal yoke.
If the metal is fusing into clothes or isn't well formed, you can prompt metal collar or metal arm cuffs .
The only variation is single bar yoke which gets the yoke to have only 1 bar more frequently.
Is generally consistent enough from front and back views. struggles with side views.
If anyone has any advice on increasing the concept's cohesion so it shows the exact yoke shown in the dataset, please let me know. I used almost entirely 3d renders of the same object from multiple viewpoints and they all had the same physical structure so you think it would generate the same yoke every time.
Description
FAQ
Comments (2)
It works but it seems to also affect the style of the image (it veers towards the style in your demo images)
May I ask which tool did you use to train the lora?
yeah I strongly tagged the art style of most of the images, so the one image i didn't tag as strongly bleed through a lot. I'll fix that in a 1.1 version later, but you should be able to describe a style to overwrite it.
I just use kohya_ss, or rather Anima-Standalone-Trainer which uses sd-scripts just like kohya.