Haiyama Mio Character LoRA

Bring Haiyama Mio to life with this character LoRA trained from anime and manga artwork and a carefully curated dataset of approximately 20 images, developed using my newer training process for Anima models.

The dataset was carefully selected to capture Mio's distinctive appearance while providing variation in poses, expressions, angles, and settings. Particular attention was given to Mio's physical appearance and recognizable features, including her long grey hair, blue eyes, mole beneath her eye, and earrings, with the goal of preserving her identity while allowing the character to remain flexible across different generations.

Testing Uncolored Manga Panels

One of the primary goals of this LoRA was to test how well an Anima model could learn a character from a dataset consisting primarily of uncolored manga panels.

Rather than relying on a large collection of fully rendered, colored illustrations, this dataset intentionally makes heavy use of black-and-white manga artwork. This provided an opportunity to test whether the training process could learn Mio's character identity from line art, shading, and limited visual information while still producing a recognizable character in fully colored generations.

Because the dataset is relatively small, this LoRA should also be viewed as an experiment in small-data character training. The results were better than expected, particularly in terms of preserving Mio's recognizable appearance while allowing her to be placed into different outfits, poses, and environments.

The final model was trained using my updated Anima-model training process, incorporating the lessons learned from previous character LoRAs and this experiment with predominantly uncolored source material.

Features

• Character-focused LoRA for Mio Haiyama

• Carefully curated dataset of approximately 20 images

• Dataset consisting primarily of uncolored manga panels

• Experimental test of learning character identity from black-and-white source material

• Focused on preserving Mio's physical appearance and distinctive features

• Trained using my updated Anima-model training process

• Supports a variety of poses, expressions, and environments

• Designed to preserve Mio's recognizable appearance

• Works with different outfits

• Manually curated dataset

This model is part of my growing collection of character and outfit LoRAs, created using careful dataset curation, experimentation, and iterative training refinement.

Have fun with Mio! 💙

If you enjoy this model, please consider leaving a rating or sharing your generations. Seeing what everyone creates is one of the best parts of releasing these projects.

Enjoy and Good Luck!

— Daraco

NOTE: ArmetDaraco | Profile | Arc en Ciel is my official model list. The Bots are scraping these, so if you are finding them on any other site, they are reposted (hopefully with proper credit). So if you want to stay updated and support me in future releases, please follow me on Arc en Ciel!!