日本语版 👉
● IllustriousXL v2.0 has launched on SeaArt! This evolutionary update enhances the series' robustness and precision for superior visual control. It also expands anime style support, improves natural language parsing, and increases expressiveness, achieving the highest external TrueSkill rating to ensure excellent generation quality.
● More Detailed Update and Usage Guide for IllustriousXL v2.0
This article provides a quick guide to generating Illustrious XL characters. After reading it, you'll become more proficient in using the IllustriousXL model.
1. Why Use LoRA-Free Generation?
IllustriousXL can generate characters directly through natural language prompts without using character LoRAs. This guide contains a verified character list and optimization techniques to help streamline your character generation process.
2. Core Technique: Generating Characters Without LoRA
● Basic Rules:
○ Enter character name directly (e.g., Mirko).
○ If unsuccessful, use alias + work title (e.g., Mirko \(my hero academia)).
○ For anime, try using the character's original Japanese name.
💡 A helpful first step is to search for your desired character on Danbooru. If there are over 100 images of the character, it usually can be generated directly.
● Advanced Enhancement:
○ Add feature tags (e.g., for Kaworu Nagisa, add white hair, red pupils, and white shirt).
● Ultimate Solution:
○ If still unsuccessful, consider using or training LoRAs.
3. Character Recognition Levels and Prompting Strategies
A. Complete Recognition (No LoRA Needed)
○ Prompt Format: character name \(work title), no additional description needed.
○ Example Characters:
Rem (Re: Zero)
Harley Quinn (DC)
Iron Man (Marvel)
Batman (DC)
Pikachu (Pokémon)
B. Recognizable but May Need Additional Details
● Prompt Format: character name \(work title) + feature tags (such as species and clothes).
● Example Characters:
Shinji Ikari (EVA)
Toothless (How to Train Your Dragon)
Kratos (God of War)
💡 For example, the prompts for "Shinji Ikari (EVA)" are:
Character name \(work title) + feature tags (core character features, detail features, scene/state additions)
Shinji Ikari \(Neon Genesis Evangelion), plug suit, EVA-01, goggles, entry plug cockpit, NERV logo, determined expression, post-battle scars, angel attack background
C. Requires LoRA or Complex Prompts
● Prompt Format: character name \(work title) + detailed description + style keywords.
● Example Characters:
Mega Rayquaza (Pokémon)
Kaworu Nagisa (EVA)
💡 For example, the prompts for "Kaworu Nagisa (EVA)" are:
The version used with LoRA (the structure is the same as B-level):
Kaworu Nagisa \(Neon Genesis Evangelion), silver hair, red eyes, white plug suit, angelic aura, piano keys, Lilith-inspired cross halo, floating in LCL lake, subtle smile
The version relies on complex prompts and does not require LoRA:
Kaworu Nagisa \(Neon Genesis Evangelion), Tabris the 17th Angel, silver hair with light grey gradient, crimson irises, pale skin, white EVA pilot suit with black accents, sitting at piano, geometric angel wings, Adam's embryo in hand, NERV HQ background, neon-blue lighting, evangelion unit-02 nearby
Tabris the 17th Angel: emphasizes angel identity, avoiding confusion with ordinary characters
Adam's embryo: references core plot device
Geometric angel wings: differs from traditional wings, closer to EVA aesthetics
Neon-blue lighting: reproduces EVA's signature cool-toned scenes
D. Requires LoRA
● Characteristics: cannot be generated through prompts alone.
● Example Characters:
Rei Ayanami (EVA)
Spider-Man 2099 (Marvel)
Black Panther (Marvel)
4. Verified Recognizable Character List
Conclusion
Through systematic prompts and character lists, users can bypass LoRA limitations and generate 90% of common characters with ease. However, some rare characters still require the use of LoRA and custom training to balance efficiency and precision.
Over time, the number of recognizable characters will continue to increase. You are welcome to update your discoveries under this post to benefit more people!
