# Dollamor — Krea 2 LoRA (v0.1)

This is a first experimental LoRA trained to capture the aesthetic I love from the [Dollamor](https://civarchive.com/models/2027393/dollamor) model.

Trained on a private synthetic dataset of roughly 350 images using Krea 2 RAW. The dataset focuses exclusively on female characters with a strong emphasis on dramatic lighting, elaborate jewelry, and carefully composed poses. Character variations include demons, cat-girls, dragons, and cyborgs.

## What to expect

- Rich, warm lighting with a tendency toward ethereal glows and light particles

- Detailed gold jewelry and ornate accessories

- Confident, sensual poses with attention to fabric and drapery

- Works best with descriptive natural-language prompts (the dataset captions are JoyCaption-style)

## Known issues

This is a first pass — there are some quirks from dataset captioning and general rough edges. The LoRA can be overzealous with light particles. Future versions will refine the captioning and dataset curation.

## Important: uncensoring LoRA required

This LoRA was trained on a Krea 2 RAW model pre-merged with the [Krea2 TextFusion Refusal-Reduction LoRA](https://civarchive.com/models/2775340/krea2-textfusion-refusal-reduction-lora) to bypass Krea 2's built-in safety filter. You must also use this bypass LoRA when generating (both RAW and Turbo), otherwise results will be degraded.

Stack them in this order:

1. Krea2 TextFusion Refusal-Reduction — strength 1.0

2. dollamor — strength 0.8 – 1.2

## How to use

Base model: Krea 2 Turbo

Required LoRA: [Krea2 TextFusion Refusal-Reduction](https://civarchive.com/models/2775340/krea2-textfusion-refusal-reduction-lora) at strength 1.0

dollamor LoRA strength: 0.8 – 1.2

Recommended settings:

- Sampler: euler

- Scheduler: simple

- Steps: 8

- CFG: 1.0

A ComfyUI workflow is included — it's the same one used to generate the sample images on this page.

## Training details

- Base: Krea 2 RAW (BF16) merged with TextFusion Refusal-Reduction LoRA

- Tool: Ostris AI Toolkit

- Rank / Alpha: 64 / 64

- Steps: 1750 (early checkpoint from a 2000-step run, with batch size 2 = 4000 images total)

- Resolution: 512–1024px buckets

- Batch size: 2 (effective: 3500 images seen)

- Images: ~350 synthetic images with JoyCaption captions

- Trigger word: dollamor

---

More versions to come when time allows. Have fun !

generations without and with the lora