This lora only help emphasize the tag. Result varies depending on checkpoints and @artstyles. So go have fun experimenting with different artstyles and checkpoints.
Alert me if I got the lora file mixed up.
Weight=1.0 or above.
No need to add "standing" but you can add it, it still works but it can also give straight legs too. "Knee guard" is ok. Haven't tried adding "mechanical legs" but from previous tests it overwrite the reversed joints.
Add "The girl has a sharp crotch armor with sharp edges. The girl's crotch armor is heavily protruding out." if you want gundam-like crotch armor.
v3.0 (has trigger word) Reuploaded. Is Epoch 7, not Epoch 6
Trained using 23 images (19 from V2 and 4 from V1). Probably the last version for this lora since it already felt like perfect enough. Plus, there a few unreleased weirdly trained epochs that I can play with for fun and giggle.
networkDim: 16, networkAlpha: 8
v2.0
Trained using 19 images picked from danbooru, including non reverse joint images that help emphasize mechanical detail.
steps: 1045, epochs: 8, optimizerType: adamw8bit, Training parameter: lr: 0.0001, networkDim: 32, networkAlpha: 16, lrScheduler: cosine_with_restart, noiseOffset: 0.04
Want to turn anime girls into Barbatos Lupus Rex? Yes you can.
Steel Haze? Yes, but depending on specific artstyle or checkpoint.
V1.0
How this lora is made?
- Trained Illustrious ver using 10 from danbooru and 2 from self generated.
- Generate 8 more (too lazy) images using Illustrious lora.
- Trained Anima ver using the same 12 images and 8 generated images.
Alternative Illustrious version from other user
- Assimilated Girl Unit
- Assimilated Flesh Unit
These loras have bulky reversed joints. What I wanted to make are thin and long legs like Steel Haze.
Description
- Combined V2 dataset with a few images from V1 dataset.
- A balance mix of V2 and V1 aesthetic and consistency.
- Trained with lower network dim & alpha.
- Different tagging with actual trigger word.





