Type | |
Stats | 40 62 43 |
Reviews | (5) |
Published | May 18, 2025 |
Base Model | |
Training | Steps: 1,050 Epochs: 15 |
Usage Tips | Clip Skip: 2 |
Trigger Words | sazervisuel, orange bodysuit, brown leotard, belt, armor, gauntlets, black gloves boots helmet, visor, wearing helmet, covered face, covered mouth, covered eyes helmet, visor, helmet removed, holding helmet claw \(weapon\) |
Hash | AutoV2 B0BE420CA1 |
My Pony version of this LoRA has an issue of generating the helmet's "cat ears" when prompted for not wearing helmet, due to all of the images of no helmet being "head out of frame" with the face cropped off to prevent the AI from training on the actress's face. Fixed that issue accordingly by replacing it with "faceless, censored identity" instead, then decided to use the training data to make an Illustrious LoRA first to compare the two.
From my tests, seems like Pony is better at costume accuracy but Illustrious is better at understanding what you want and giving you the image in your mind.
Will train an updated version of the Pony LoRA with the issue fixed subsequently.
Main prompts:
sazervisuel, orange bodysuit, brown leotard, belt, armor, gauntlets, black gloves
Add if doing "full body" or "feet out of frame" (training data set had "boots" as prompt, but from my own testing, "knee boots" seem to get you a more accurate depiction of her boots including her silver knee guards):
boots
For helmet with face covered:
helmet, visor, wearing helmet, covered face, covered mouth, covered eyes
For helmet with face visible:
helmet, wearing helmet, [eye color prompts], [facial expression prompts]
For helmet to be held and not worn:
helmet, visor, helmet removed, holding helmet
For her claw weapons:
claw \(weapon\)