Pipkin Pippa of Phase Connect
YT:https://www.youtube.com/@PipkinPippa
SDXL 1.0 - Made this lora sometime ago might as well publish it. Worked best with weight 0.6 in my experience. Hair ornaments kinda work.
Trained using FlatpieceCoreXL + FlatpieceXL VAE.
10/7/24 - Added a chinchilla version trained using images from https://x.com/RageInTheMist/status/1842770838652616912
Upper body focused, however AnimagineXL has no issues filling in the blanks :)
Weight of 0.6 recommended.
10/16/24 - Added v2 for SDXL 1.0
Weight of 0.6 recommended
base prompt: pippa, rabbit ears, bangs, virtual youtuber, pink hair, pink eyes, ahoge, looking at viewer, hair ornament,12/28/24 - Added a Flux Schnell version using the same dataset as the SDXL v2 model.
Weights of 0.75 - 1 seemed to work fine but could vary per checkpoint
1/27/25 - Added Flux Dev version
Weights of 0.5 - 0.85 recommended but YMMV
The chinchilla variant tends to appear unprompted adding "tail" to negative prompt should do the trick
Description
Optimized the dataset for accuracy