Valindia — Krea 2 Character LoRA

My first character LoRA!

I’m really happy with how Valindia turned out. She holds her identity surprisingly well across different outfits, locations, expressions, cosplay, fantasy scenes, and both SFW and NSFW generations.

If you use her, please post your generations! I genuinely want to see what people make with her.

Quick Start

Trigger: Valindia

Recommended Strength: 1.0

I trained and primarily tested her with the expectation that you can simply load the LoRA at 1.0 and start prompting.

For more flexibility, especially when changing strongly learned features, you can also experiment around 0.8–1.0.

Example:

A photo of Valindia, standing in a coffee shop...

A cinematic shot of Valindia, casting magic in a ruined cathedral...

A selfie of Valindia, sitting on a bed...

About the LoRA

Valindia was trained on Krea 2 Raw using AI Toolkit.

Training:

- 30 curated training images

- 1500 steps

- Krea 2 Raw base

- AI Toolkit

- Character identity focused dataset

The goal was strong facial consistency without locking her into a specific outfit, location, pose, or type of scene.

She has been tested successfully with portraits, full-body images, fantasy scenes, cosplay, different expressions, suggestive content, and explicit NSFW generation.

Showcase Images

The images on this page were generated using Lustify with the LoRA at 1.0 strength.

Lustify has a fairly strong NSFW bias, so results may look different when using stock Krea 2 Turbo, Krealism, or other Krea-based checkpoints.

The LoRA itself was trained on Krea 2 Raw rather than Lustify, so while I've only really tests on Lustify (its my go to right now) it should work on other checkpoints.

Prompting Tips

Valindia generally works well by simply starting your prompt with her name:

A photo of Valindia...

A cinematic shot of Valindia...

A selfie of Valindia...

You do not need to describe her normal appearance every time. Let the LoRA handle the identity and just focus on describing the scene, clothing, expression, pose, lighting, etc.

For multiple people, describe the other character clearly. Hair color, appearance, clothing, and other identifying features help keep the second person distinct.

Attached Workflow

All showcases (and most of my stuff) has the workflow attached - a modified version of the Lustify workflow by Coyotte - so you can drop it into comfyui, download the models, and go to town if you like.

Known Issues

Hair:

Her default hairstyle and color are fairly strongly learned. Different hair colors and styles can be prompted and do work, but results are currently hit or miss. Lowering LoRA strength slightly can sometimes help.

Eyebrow Piercing:

In some distant or full-body generations, Krea occasionally duplicates her eyebrow piercing. This appears to be mostly a small-detail consistency limitation of Krea rather than a major LoRA issue.

Multiple Women:

If you simply prompt "Valindia with another woman," the model may decide that you obviously wanted two Valindias (Twins Basil, TWINS!).

Describe the second woman clearly and the problem is significantly reduced.

Future Updates

For now, I’m very happy with this version.

If people enjoy using her and I start seeing a good number of community generations, I’ll consider doing another training round focused on improving hair flexibility and separation between Valindia and additional characters.

So seriously — post what you make. SFW, NSFW, weird cosplay, fantasy nonsense, whatever. I want to see where people take her.