Ari Arkaza — Character LoRA for Anima
My independently trained Anima LoRA of Ari Arkaza, associated with @KyosSkittles.
The character design combines long brown hair with vivid purple sections, blue eyes, colorful candy-shaped hair ornaments, and gold lightning-bolt earrings. A purple hoodie with yellow lining over a black high-neck top is a recurring outfit in the references.
WHY I TRAINED THIS
Yet another VTuber trashing publicly her fans for internet points. If you don't like AI that is ok, but to humiliate a fan publicly for it crosses a line.
Then promoting Nightshade, known to do nothing as a snake oil to other artists.
So I did the funny and poisoned her dataset to show it does nothing.
A nightshaded LoRA.
CHARACTER CREDIT
Character credit: Ari Arkaza / @KyosSkittles
This is an independent training project. It is not an official release or an endorsement by the character's creator.
THE DESIGN
The main features to preserve are:
• Long brown hair with purple front sections and purple accents through the lengths
• Bright blue eyes
• Colorful candy-shaped hair ornaments
• Gold lightning-bolt earrings
• A small upward-curving hair tuft
• Purple and yellow accents in the familiar outfit
For the familiar outfit, try:
purple hoodie, yellow hood lining, black turtleneck
Clothing can change. Describe the outfit you want rather than including the hoodie in every prompt.
AVAILABLE VERSIONS
• 3000 Steps
The earlier checkpoint from this training run.
• 3600 Steps
A later checkpoint saved after another 600 training steps.
Both use the same trigger. Compare them with the same prompt, seed, checkpoint, and LoRA strength to make the differences easier to judge.
TRIGGER WORD
arinimaExample:
arinima, solo, purple hoodie, yellow hood lining, black turtleneck, looking at viewer, simple backgroundStart with the trigger and your scene description. Reinforce individual appearance details only when needed.
Optional supporting tags:
long hair, brown hair, purple hair, multicolored hair, blue eyes, candy hair ornaments, lightning bolt earringsThese are supporting descriptions, not additional trigger words.
STRENGTH
Try 1.0 as an initial setting, then adjust for your checkpoint and prompt.
Lower it if the LoRA overwhelms the style or restricts the result. Raise it if the character's features aren't coming through clearly.
This is an initial setting to try, not a tested recommendation for every setup.
EXAMPLE PROMPTS
Simple character check:
arinima, solo, upper body, purple hoodie, yellow hood lining, black turtleneck, relaxed smile, looking at viewer, white backgroundDifferent outfit:
arinima, solo, full body, navy tracksuit, white sneakers, standing beside a bicycle, outdoorsQuiet scene:
arinima, solo, cream sweater, blue jeans, sitting by a window, holding a book, soft afternoon lightCOMPATIBILITY
Character LoRA for Anima. Results depend on the checkpoint, prompt, strength, and other LoRAs in your setup.
Training details for each checkpoint are listed under its version.
Description
The 3600-step Ari Arkaza checkpoint for Anima.
This is a later checkpoint from the same training run as the 3000-step release, saved after another 600 steps.
QUICK START
Trigger: arinima
Initial strength to try: 1.0
Example:
arinima, solo, cream sweater, blue jeans, sitting by a window, holding a book, soft afternoon light
Adjust the strength according to the result. If the LoRA dominates the style or makes outfit changes difficult, lower the weight or compare with the 3000-step checkpoint.
VERSION DETAILS
• Base model family: Anima
• Training steps: 3600
• Network rank: 6
• Format: SafeTensors
• Training software: AI Toolkit Experimental 0.13.23
• File: arkaza_anima_000003600.safetensors
More steps don't guarantee a better result. Compare both checkpoints using the same prompt, seed, base model, and strength.
The image-protection experiment behind this training is described in the main model description. This checkpoint is not, by itself, a controlled test of Glaze or Nightshade.
CREDIT
Character: Ari Arkaza / @KyosSkittles
Independently trained LoRA. Not an official or endorsed release.



