Avatar Set · Realistic · AI Lab

Dark Elf Nobles — Avatar Set

Male of the pair — the same house palette, silver hardware and crescent motif.

Stack

Role: Set art direction, matched-prompt pair generation, consistency QA & retouch
Base model: Krea 2 (open-weights)
LoRA: Self-trained painterly realism × ornate fantasy LoRAs, stacked ↓
Tools: ComfyUI, two self-trained style LoRAs stacked + Krea 2, directed generation, retouch
Year: 2026

Process

Generated as one set

Both characters come from the same pipeline — two self-trained style LoRAs stacked on an open-weights base, one shared prompt skeleton binding the pair: the same noble-house palette of obsidian black, silver-white and a restrained amethyst accent, the same brushed-silver hardware language, a shared crescent motif. Selection from seed batches, then consistency QA and targeted retouch.

From the AI Lab

Model Card Cinematic Painterly Realism — LoRA Low-saturation painterly realism for splash art and character concepts. 85-image dataset.

Overview

A style LoRA for cinematic, low-saturation painterly realism — the workhorse behind the splash-pipeline posts (Empress of the Lotus Pond, Saint of the Storm, Golden Emperor).

Card


Base model	Krea 2 (open-weights, turbo)
Dataset	85 curated images, pure-content prose captions, no trigger word
Network	LoRA dim 32 / alpha 32 · 16 epochs · final loss 0.067
Validation	Epoch sweep with fixed-seed probes; epoch 16 selected as standard
Operating notes	Dark cinematic backdrops are part of the style contract — flat studio backgrounds suppress it

Notes

Prompting rules discovered in validation (front-loaded subject, positive-only phrasing at CFG 1, backdrop placement) are documented per post in the Process sections.

Open in LAB →