🐺 ARTURO WOLFF — Character LoRA for Illustrious

A character LoRA built to reproduce Arturo Wolff — me.

This is my Arturo Wolff character LoRA, trained for the Illustrious / SDXL ecosystem.

Arturo is my personal character, online identity, and the central figure behind most of the artwork I make. Over time, his design has accumulated a lot of little visual decisions that are easy to describe individually but surprisingly difficult to reproduce as a coherent whole: the proportions, facial structure, scruffy silhouette, color separation, eyes, ears, tail, posture, expressions, and that permanently exhausted-looking rockstar energy.

The purpose of this LoRA is to compress all of that into a reusable character representation.

The goal is:

Arturo should remain Arturo even when everything else changes.

◆ WHO IS ARTURO?

Arturo Wolff is a slim anthropomorphic wolf with a deliberately recognizable visual identity:

• Gray fur

• Dark / black countershading

• Black ears

• Red eyes

• Scruffy fur

• Large pointed wolf ears

• Digitigrade anatomy

• Large, fluffy two-tone wolf tail

• Lean proportions rather than an exaggerated muscular build

• Naturally tired-looking eyes and facial structure

• Usually carrying some degree of dry, sleepy, unimpressed, or rockstar-like expression

That last part is particularly important.

The tired appearance is part of the character design, rather than simply an expression I occasionally give him.

Ideally, the LoRA should reproduce those characteristics as part of Arturo's identity without requiring a dozen additional descriptors every time.

However, depending on the checkpoint and how strongly it responds to character LoRAs, some features may need reinforcement.

Useful tags include:

tired eyes, bags under eyes, notched ear, scruffy fur, slim, anthro, furry, digitigrade, black ears, two-tone tail

These are particularly useful when a model starts smoothing out Arturo's face, changing his fur distribution, or turning him into a more generic wolf character.

◆ TRIGGER / IDENTITY

Primary identity token

awffnai

Trigger-token note: The uploaded SafeTensors metadata does not record the actual training trigger. I've used awffnai for the Illustrious examples; replace it with the exact token from this model's training captions if different.

For most prompts, I recommend treating the character token as the anchor and then describing the scene normally.

You do not necessarily need to exhaustively reconstruct Arturo from tags every single time.

That defeats part of the purpose of a character LoRA.

A basic prompt can be as simple as:

1boy, awffnai, furry, anthro, black jacket, standing in a convenience store at night, looking at viewer

Then add pose, clothing, expression, framing, environment, style, artist tags, etc. as needed.

Illustrious checkpoints tend to be comfortable with Danbooru-style tagging, so you can use that structure to control the image without turning every prompt into a complete character sheet.

If a particular generation begins drifting, explicit anatomical or color descriptors can be added to reinforce the relevant feature.

◆ EXAMPLE PROMPTS

Simple Character Test

1boy, awffnai, furry, anthro, standing, full body, looking at viewer, relaxed posture, black shirt, jeans, simple interior

A useful test because there is very little for the prompt to hide behind.

Facial Structure Test

1boy, awffnai, furry, anthro, close-up, portrait, looking at viewer, neutral expression, tired eyes, red eyes, messy hair, simple background

This is useful for checking how well the model retains Arturo's facial structure, eye shape, ears, and fur separation.

Anatomy Test

1boy, awffnai, furry, anthro, full body, standing, side view, digitigrade, black ears, large fluffy tail, casual clothing, simple background

Useful for identifying whether a particular checkpoint struggles with his digitigrade anatomy or tail.

Scene Test

1boy, awffnai, furry, anthro, sitting sideways in a diner booth, one arm resting over the backrest, holding a coffee cup, looking out the window, night, rainy window, cinematic lighting

This is closer to how I actually want people to use the model.

Put him in a situation and see what happens.

◆ PROMPTING PHILOSOPHY

I generally recommend describing what you want Arturo to be doing, rather than spending half of the prompt repeatedly explaining who Arturo is.

For example:

Instead of:

gray wolf, black ears, red eyes, gray fur, black fur, fluffy tail, two-tone tail, scruffy hair...

try:

awffnai, sitting sideways on a diner booth, one arm resting over the backrest, holding a coffee cup, looking out the window

Then reinforce individual physical traits only when necessary.

A good identity LoRA should perform some of the descriptive work for you.

That said, Illustrious finetunes can have fairly strong preferences regarding anatomy, expressions, character proportions, and stylistic conventions.

If Arturo starts looking too muscular, too clean, too cute, or like somebody borrowed his face and forgot to return it, adding a few physical descriptors can help steer things back.

◆ LoRA STRENGTH

Recommended starting strength: 1.0

Start at 1.0 and adjust according to the checkpoint.

You can experiment with lower strengths when combining the LoRA with particularly strong styles or other character LoRAs.

Higher strengths may be useful when a checkpoint starts overriding Arturo's design.

These are prompting recommendations, not a claim that the uploaded file has been generation-tested across that strength range.

◆ BASE MODEL

Architecture: Stable Diffusion XL (SDXL)

Model Family: Illustrious / SDXL

Model Format: SDXL LoRA

Implementation: Diffusers

Network Module: networks.lora

Training checkpoint recorded in metadata:

urn:air:sdxl:checkpoint:civitai:795765@889818

This LoRA uses SDXL architecture and is intended for use with compatible Illustrious-family checkpoints and finetunes.

Its behavior may vary depending on how heavily the target checkpoint modifies the original training model.

◆ TRAINING DATA

The dataset metadata records 50 images with one repeat, while the training settings record 20 configured epochs and 3,000 maximum steps. These should not be interpreted as proof that the uploaded checkpoint completed all 3,000 steps.

◆ NETWORK CONFIGURATION

LoRA Rank / Network Dimension: 32

Network Alpha: 32

The model contains LoRA weights for the SDXL U-Net and both text encoders.

◆ OPTIMIZATION

Optimizer: Adafactor

Learning Rate: 0.0005

Model / U-Net LR: 0.0005

Text Encoder LR: 0.00005

LR Scheduler: Cosine

Max Gradient Norm: 1

Min SNR Gamma: 5

◆ CAPTION REGULARIZATION

Caption Dropout: 0.05 / 5%

A small amount of caption dropout was used during training.

Conceptually, caption dropout discourages the network from relying perfectly on every caption element being present on every iteration.

For character training, this can help reduce brittle relationships between the character and incidental tags in the dataset.

The objective is not just:

specific token combination = specific training picture

but a more generalized internal representation of the character.

Ideally, Arturo should remain recognizable even when his outfit, environment, expression, pose, or artistic style changes.

Caption dropout alone does not guarantee that behavior, of course. The variety of the dataset and the characteristics of the checkpoint also matter.

◆ PRECISION & MEMORY CONFIGURATION

Training Mixed Precision: BF16

Gradient Checkpointing: Enabled

Latent Caching: Disabled

Full FP16: Disabled

Noise Offset: 0.1

◆ TRAINING ENVIRONMENT

The embedded training metadata reports:

Training software: ai-toolkit

ai-toolkit version: 0.12.14

Training controller: Civitai Spine Controller

GPU architecture: Ada

Reported VRAM: 24 GB

Model metadata date: September 20, 2026

The file contains SafeTensors metadata covering the model architecture, training configuration, network settings, software version, and model identification.

A note on training progress

The filename identifies this release as epoch_9, but the embedded metadata contains conflicting progress information.

The training_info field reports step 1350 and epoch 8, while other metadata fields report epoch 1 and 3,000 training steps.

These values appear to combine different stages or representations of the training run, so I wouldn't treat any one of them as a definitive completed-epoch count without the original training logs.

◆ WHAT I WANT FROM THIS MODEL

1. For me to create content of my fursona.

2. For others to create fanart of my fursona.

3. Validation.

That's pretty much it.

I want to be able to put Arturo into situations without spending an unreasonable amount of time fighting the model over whether his ears should be black, whether he's supposed to be built like a refrigerator, or whether his face should look like a completely different wolf.

And I want other people to have that same freedom.

◆ ABOUT THE CHARACTER

Arturo exists at the intersection of my artwork, music, online identity, and the larger collection of characters and projects I've built around him.

He's the guitarist and vocalist of AWFF, but his existence isn't limited to the band.

Arturo is also generally portrayed as a womanizer, depressed, whimsical, and a bunch of other stuff I can't share here.

You are absolutely allowed to take him out of that context tho.

Put the wolf somewhere stupid.

Give him a normal office job.

Make him work the night shift at a gas station.

Send him to space.

Put him in the most unnecessarily elaborate outfit imaginable.

There's no obligation to respect the lore every time you generate something with him.

◆ FINAL NOTES

The model is meant to be used, pushed around, mixed with styles, given weird outfits, placed into environments completely outside of the original dataset, and generally stress-tested.

Whether you're making fanart, experimenting with different Illustrious checkpoints, or just seeing how many increasingly questionable situations you can put a tired wolf into, go for it.

If you make something interesting with him, I genuinely want to see it.

🐺

Have fun.