Haru is an original, entirely AI-generated adult kemonomimi character, age 18. This Illustrious-trained character LoRA focuses primarily on her facial appearance, with a visual style that leans toward semi-realism, stylized 3D/CG, and anime.

Use it to bring Haru’s recognizable face into portraits, character illustrations, and wider scenes while allowing your checkpoint and prompt to shape the clothing, setting, and overall presentation.

Quick start

• Trigger word: HaruFace

• Recommended strength: 0.5–0.75

• Suggested starting strength: 0.5

• Training base: Illustrious

• Best-suited styles: Semi-realistic, stylized 3D/CG, and anime

Start at 0.5 and increase gradually if you want a stronger character influence. In my testing, strengths above 0.75 generally produced less pleasing results. Higher strength is not necessarily better for likeness or image quality.

Recommended checkpoints

My best results have come from:

• Perfect Deliberate

• Prefect Illustrious

• WAI Illustrious

These are recommendations based on my own testing, rather than a guarantee of identical results across every version or workflow. Strongly photorealistic checkpoints have been less successful for the look this LoRA is intended to produce.

What it does best

Haru works best in close-up and upper-body portraits where her facial features have enough room to resolve clearly. Neutral expressions, subtle smiles, and restrained expressive poses are particularly good starting points.

The LoRA is intended primarily to influence facial identity. It is not designed to precisely reproduce a particular body, outfit, or background.

Hair and neko features

Haru’s short pixie-cut hair and neko ears are strongly associated with the trained character. Expect them to appear readily, and be aware that changing or removing them can be difficult.

Known limitations

• Distant faces: Close-ups are more reliable than full-body compositions. When the face occupies a small part of the image, eye alignment and fine facial details can become less consistent.

• Open-mouth expressions: These can work, but teeth, mouth shape, and likeness are somewhat less predictable than with relaxed or subtly parted lips.

• High strength: Settings above 0.75 have generally reduced quality in my testing.

• Photorealism: This resource is better suited to stylized rendering than strict photographic realism.

These are occasional limitations, not failures in every image. For wider compositions, a restrained face-detailing or inpainting pass may help, depending on your workflow.

Prompting

Include HaruFace, then describe the expression, framing, clothing, lighting, and setting you want. Begin with a straightforward prompt and the LoRA at 0.5 before adding other resources or more complex instructions.

For example:

HaruFace, solo, portrait, looking at viewer, gentle smile, short black pixie-cut hair, blue eyes, cat ears, soft lighting, simple background

Adjust the style and quality terms to suit your chosen checkpoint. For expressions, describe the visible result—such as “closed-mouth smile” or “slightly parted lips”—rather than relying only on a broad mood label.

Character and source

Haru is a fictional adult character created using AI-generated imagery from Civitai and other AI-generation tools. She is not intended to represent a real person. The source imagery was created with adult, 18+ character specifications.

Feedback

If you share results or report an issue, please include your checkpoint and version, LoRA strength, prompt, and any face-detailing or upscaling steps. That makes it much easier to compare results and understand what works well.