santa hat
deerdeer nosedeer glow
Sign In

Marle Barrock (Reincarnation Colosseum | Tensei Colosseum) | LoCon

315
1.3k
5.2k
50
Verified:
SafeTensor
Type
LyCORIS
Stats
1,316
5,153
Reviews
Published
Aug 16, 2023
Base Model
SD 1.5
Training
Steps: 4,080
Epochs: 40
Usage Tips
Clip Skip: 2
Trigger Words
MarleBarrock, dark-skinned female
Muscular, large breasts
muscular, bandages, bandaged arm, detached sleeves, bandaged legs, black thighhighs, pelvic curtain, tabard
smile, sharp teeth, grin
Hash
AutoV2
7F6CA39648
default creator card background decoration
Manityro's Avatar
Manityro

I have opened up a request form

Marle Barrock from Reincarnation Colosseum an Ecchi/Erotica Manga. Works best on 0.8-1.0. Training done on NAI.

This is a LoCon which can be placed in LoRA folder on Automatic1111 1.5+ so make sure to update otherwise place in Lycoris folder.

Trigger: "MarleBarrock, dark-skinned female" always needed, for accuracy add "Muscular, large breasts".

Suggestions:

  • Outfit from the Manga "muscular, bandages, bandaged arm, detached sleeves, bandaged legs, black thighhighs, pelvic curtain, tabard".

    • Even "from behind" works ok given the 2-3 back pics ;)

  • Iconic face: "smile, sharp teeth, grin"

  • Hair is very stable but may rarely lose the yellow color on the single bang, just regen at 1.0 weight and it will return.

  • Her outfit loves to bleed into other clothes but many clothes that do not resemble tabards/suits work pretty well.

    • Things like "sideboob, side cutout, tabard" can help counter some of the outfit bleed.

My goal was to improve upon existing LoRAs for this character by getting the look right while having less/no style baked in so it can be used on more models successfully. I manually edited the dataset to cleanup the images and I used Regularization images to achieve a more stable look with less style burn in.

I bumped into this Manga by chance once I for sure wanted to make her once I found out there were colored pictures out there. I got the hair + single bang working very well and consistent even with a 50% black & white dataset. I hope you all enjoy.

Feedback and reviews are always appreciated.