Fuwa Wuffington — Personal Retrain for Anima
Fuwa belongs to GAi. This is my personally trained LoRA of his character.
Fuwa is a pokesona based on a shiny Furfrou that GAi caught during a Pokémon X/Y stream. She went on to become one of his signature characters.
WHY I TRAINED THIS VERSION
I wanted to use Fuwa alongside other character and style LoRAs.
In my setup, GAi's original LoRA tended to dominate the result and was difficult to combine with other LoRAs. I trained this version with the aim of keeping Fuwa recognizable while leaving more room for the rest of the image.
That is the goal of this release. Results still depend on your checkpoint, prompt, LoRA strengths, and the other models you're combining.
CHARACTER CREDIT
Fuwa Wuffington is GAi's original character.
Original creator and model:
https://civarchive.com/models/2395240/gais-ocs-or-fuwa-wuffington
Credit for the character and her design goes to GAi. My contribution is this particular LoRA training.
AVAILABLE VERSIONS
• V2 · 3000 Steps
The earlier of the two checkpoints from my V2 training run.
• V2 · 3600 Steps
The same run after another 600 training steps.
V2 refers to my own training revision, not a version of GAi's original release.
Both checkpoints use the same trigger. Compare them with the same prompt and settings; the later checkpoint isn't automatically the better choice.
TRIGGER WORD
fuwanimaUse this token for my retrain. GAi's original model has its own trigger words.
Start with fuwanima, then describe the outfit, pose, expression, and setting you want.
Example:
fuwanima, solo, standing by a cafe window, holding a coffee cup, looking at viewer, soft afternoon lightCHARACTER DETAILS
Fuwa's design includes short black hair, red eyes, freckles, dark blue eyeshadow, bags under her eyes, and a tall, slender build.
For her anthro appearance, describe the dog ears, tail, animal nose, and fur when needed. If you want the familiar outfit, add:
white turtleneck sweater dress, black pantyhose, white flatsOutfit and appearance tags are supporting descriptions, not additional trigger words. Start simple and reinforce whichever features need help.
USING OTHER LORAS
Combining Fuwa with other LoRAs is the reason I made this version.
Try a strength of 1.0 as an initial setting, then adjust to suit your setup. Lower it if Fuwa starts taking over the style or passing her features to another character. Raise it if her identity isn't coming through clearly enough.
For scenes with multiple characters, describe each character separately and keep their features clearly assigned. Some mixing can still happen; this retrain doesn't guarantee complete separation.
EXAMPLE PROMPTS
Simple character check:
fuwanima, solo, full body, standing, relaxed posture, white turtleneck sweater dress, black pantyhose, white flats, simple backgroundQuiet afternoon:
fuwanima, solo, sitting by a cafe window, holding a coffee cup, thoughtful expression, warm afternoon lightDifferent outfit:
fuwanima, solo, oversized black jacket, jeans, walking along a rainy street at night, neon reflectionsCOMPATIBILITY
Trained for Anima. Different Anima checkpoints and combinations of LoRAs can produce different results.
The aim is simple: recognizable Fuwa, with room for the rest of the prompt.
Description
The 3000-step checkpoint from my second Fuwa Wuffington training run.
I trained this version to make Fuwa easier to use alongside other character and style LoRAs. In my setup, the original owner's LoRA had more influence than I wanted for those combinations.
This is the earlier of the two published V2 checkpoints. Compatibility with other LoRAs is the training goal, rather than a guarantee for every combination.
QUICK START
Trigger: fuwanima
Initial strength to try: 1.0
Example:
fuwanima, solo, full body, white turtleneck sweater dress, black pantyhose, white flats, relaxed posture, simple background
Adjust the strength according to how clearly Fuwa comes through and how much room the other LoRAs need.
VERSION DETAILS
• Base model family: Anima
• Training steps: 3000
• Network rank: 6
• Format: SafeTensors
• Training software: AI Toolkit Experimental 0.13.23
• File: Fuwa_Wuffington_v2_000003000.safetensors
The 3600-step checkpoint continues the same training run for another 600 steps. Compare both using the same prompt and settings.
V2 is my training revision. Fuwa Wuffington and her character design belong to GAi.
Original creator's model:
https://civitai.red/models/2395240/gais-ocs-or-fuwa-wuffington




