Pixxie3 (Pony XL)
This is an independent Pony XL training of the same carefully curated dataset used for my current Illustrious development, trained in parallel rather than converted from another LoRA.
The goal was to see how Pony’s architecture interpreted the character while preserving the same identity and costume design.
In testing, Pony gave Pixxie noticeably more expressive energy and exaggerated personality while maintaining strong costume fidelity, side profiles, and back views. If you’re looking for a version that leans into Pony’s strengths without abandoning character consistency, this is the one I’d recommend.
Recommended weight: 1.0 (tested 0.8–1.2).
Interestingly, using the same dataset, Pony consistently produced a livelier, more exaggerated interpretation of Pixxie than the Illustrious version. That extra energy fits her personality surprisingly well.
Description
I'm pleased to share the updated version of my Pixxie LoRA. I've found that in now consistently produces:
Improved identity preservation
Better side-profile consistency
Stronger expression variety
Reliable back-view understanding
Improved prompt generalization
It seems to have a good fidelity to detail without evidence of overtraining.