Avatar Set · Flat · AI Lab

Chibi Party — Avatar Set

Stack

Role: Set art direction, paired generation in one pipeline run, consistency QA & retouch
Base model: Krea 2 (open-weights)
LoRA: Self-trained chibi game style LoRA ↓
Tools: ComfyUI, self-trained style LoRA + Krea 2, directed generation
Year: 2026

Process

Generated as one set

All characters come from a single pipeline run — one style LoRA, one prompt skeleton, shared palette and design language — so they read as a matched set out of the box. Selection from a seed batch, then consistency QA and light retouch.

From the AI Lab

Model Card Chibi Game Style — LoRA Clean chibi game-art style. 60-image dataset, dual-epoch operation (stable / cute).

Overview

A style LoRA for clean chibi game art — used for the Chibi Party set and the Gala Diva splash.

Card


Base model	Krea 2 (open-weights, turbo)
Dataset	60 curated images, prose captions, no trigger word
Network	LoRA dim 32 / alpha 32 · final loss 0.0914
Validation	Epoch sweep; two operating points kept — a stable epoch for fidelity, a later epoch for cuter, simpler shapes
Operating notes	Prompts must stay short; the style expects chibi proportions and collapses on adult ratios

Epoch sweep — out-of-distribution probe

Fixed seed and prompt across checkpoints. The stable epoch holds proportions; the later epoch pushes rounder, simpler shapes — both are kept as deliberate operating points:

Epoch sweep

Open in LAB →