santa hat
deerdeer nosedeer glow
Sign In

Laimu (Limealicious) PonyXL (Vtuber) | LoCon

90
494
14.2k
30
Verified:
SafeTensor
Type
LyCORIS
Stats
494
14,198
Reviews
Published
Apr 26, 2024
Base Model
Pony
Training
Steps: 1,440
Epochs: 12
Usage Tips
Clip Skip: 2
Strength: 1
Trigger Words
LaimuVT, green eyes, animal ears, tail, multicolored hair
choker, necklace
Medium hair, single braid
Short hair
Long hair
gradient jacket, off shoulder, striped thighhighs, asymmetrical legwear, mismatched legwear, animal ear legwear
Hash
AutoV2
CB894D7D0B
default creator card background decoration
Manityro's Avatar
Manityro

The Indenpendant Vtuber Laimu or Limealicious. Works best around 1.0 weight. Training done on Pony Diffusion V6 XL. Request form

Trigger: "LaimuVT, green eyes, animal ears, tail, multicolored hair" always needed. Add "choker, necklace" for the cute Lime necklace and then choose a hairstyle from below.

Hair options:

  • "Medium hair, single braid" - Classic style

  • "Short hair" - Short Alt

  • "Long hair" - Loose long Alt

Suggestions/notes:

  • Main Outfit: "gradient jacket, off shoulder, striped thighhighs, asymmetrical legwear, mismatched legwear, animal ear legwear"

    • Outfit color is a bit random so if you want a more stable jacket color change "gradient jacket" to "white jacket".

    • Sometimes the pink from the dress bleeds through but a regen of the image tends to fix it.

  • Dress Outfit: "Pink Dress"

    • Doesn't need much else to support it since it generates the arm pieces as is.

  • Flexible as much as PonyXL will allow so have fun.

    • There is a mild background bleed (windows, walls) from the stream background if you don't prompt much for it. Fix is to prompt more for the background.

Green is good and Limes are tasty, Laimu had a really cute design so I wanted to see what it would be like to train a Vtuber as the only one I made, Projekt Melody, was more 3d and had easier to work with data where with Laimu was a bit harder but fun. I hope you all enjoy.

Feedback and reviews are always appreciated.

Training data is in the Lora metadata.

Trained on 44 images.