Illustrious-XL-v3.0-EPS-stable

Illustrious XL v3.0-ε — Model Introduction

Overview

Illustrious XL v3.0-ε is the “stable base” checkpoint of Onoma AI’s third-generation Illustrious series. Built on the Stable Diffusion XL backbone, it adopts an epsilon-prediction training objective, specifically chosen to remain robust during future fine-tuning. v3.0-ε delivers distinct, stylish outputs and is sometimes the best model on certain aesthetic scoring at default settings.

Core Highlights

Stable Foundation for Training : Designed as a dependable base model suitable for future training tasks, with support for LoRA adaptation.

Stylish Visuals : Produces visually distinct, stylish outputs that score highly in aesthetic benchmarks.

Prompt-Level Composition : In an internal test of a two-subject prompt, v3.0-ε achieved 23 / 25 successful generations, correctly separating left/right color assignments.

Color-Control Tokens : Supports the new contrast, brightness, sharpness, dynamic colors, colorfulness, and saturation tokens introduced in the v3.x series, enabling direct palette control within the prompt.

Creators seeking a stylish, LoRA-friendly base model can start experimenting with Illustrious XL v3.0-ε today and follow upcoming blog updates for in-depth demos and comparison benchmarks.