Illustrious XL & NoobAI Native Characters: Complete Guide

<script type="application/ld+json> { "@context": "https://schema.org", "@type": "HowTo", "name": "How to check if a character is native to Illustrious or NoobAI", "step": [ { "@type": "HowToStep", "name": "Load a base checkpoint", "text": "Use Illustrious XL v2.0 STABLE or NoobAI XL ε-pred/v-pred in your UI." }, { "@type": "HowToStep", "name": "Prompt the character name with franchise", "text": "Format: character_name (franchise_name). Example: hoshimi miyabi (zenless zone zero)." }, { "@type": "HowToStep", "name": "Add species tag for Pokémon/Digimon", "text": "Append pokemon (creature) or digimon (creature) to avoid generic creature bleed." }, { "@type": "HowToStep", "name": "Evaluate the result", "text": "Green = perfect. Yellow = major gaps, LoRA needed. Red = not recognized, LoRA mandatory." }, { "@type": "HowToStep", "name": "Cross-reference the community list", "text": "Check the Civitai article 10242 (720 characters, 46 franchises) for pre-tested results." } ] } </script>

TL;DR

• Test first, LoRA second: Prompt character_name (franchise) on a base Illustrious or NoobAI checkpoint. If the result is Green tier, you don't need a LoRA.
• Use the community tier list: The Civitai article 10242 documents 720 characters across 46 franchises with Green/Yellow/Red ratings — but it was tested on fine-tunes, not the raw base model.
• Raehoshi ≠ native: Raehoshi Illustrious XL only adds 8 author-trained characters. They require manual hair, eye, and outfit tags. Illustrious/NoobAI native characters work from the base model's Danbooru/e621 training data.

What "native character" actually means

Both Illustrious XL and NoobAI XL are SDXL derivatives trained on massive tag-caption datasets. Illustrious uses Danbooru2023; NoobAI adds the e621 dataset on top. When a character name appears in those datasets with enough varied, well-tagged images, the name itself becomes a valid prompt token. You type hoshimi miyabi (zenless zone zero) and the model reconstructs the character — no LoRA required.

The catch: recognition is not binary. A character might render perfectly (Green), lose the signature outfit (Yellow), or collapse into a generic anime person (Red). The community tier list captures this gradient.

The two model families and what they know

Source: Illustrious XL site, NoobAI XL site, Civitai article 10242

NoobAI's e621 addition means better coverage for Western IPs, furry characters, and artist style tags. Both models support Danbooru-style character tags natively.

How to test a character yourself in 30 seconds

1. Load a base checkpoint — Illustrious XL v2.0 STABLE or NoobAI XL (ε-pred or v-pred).
2. Prompt the standard format: character_name (franchise_name)

• Example: furina (genshin impact)

1. Add species tags for Pokémon/Digimon: pikachu (pokemon), pokemon (creature)
2. Generate 4–8 images at 1024×1024 or 1536×1536, 20–30 steps, CFG 5–7.
3. Rate the output using the tier system below.

Tier definitions (from community testing)

Source: Civitai article 10242

The community character list (720 characters, 46 franchises)

The only structured, public resource is the Civitai article "Illustrious XL (v0.1) Recognized Characters List" by h_madoka. It tests 720 characters across franchises including:

• Fate/Grand Order (Artoria Caster, Nero Claudius, Mash Kyrielight, Ishtar, etc.)
• Blue Archive (Takanashi Hoshino, Sorai Saki, Misono Mika, Tendou Aris)
• Genshin Impact (Furina, Citlali, Varesa)
• Azur Lane (Eldridge, Pamiat Merkuria, Leonardo da Vinci)
• VTuber groups (.LIVE, Aogiri High School, Amaryllis Gumi, Blast Project, Guild CQ)
• Fire Emblem Engage (Alear, Celine, Veyle, Yunaka)
• Danganronpa (Kirigiri Kyoko, Nanami Chiaki, Maizono Sayaka)
• And 39 more series

Each entry links to a dedicated Civitai model page showing sample generations.

Critical caveat: The list was generated on WAI-NSFW and Hoseki fine-tunes, not the raw Illustrious base model. Fine-tunes can amplify or suppress character knowledge. Treat it as a strong prior, not ground truth.

Read the full community list →

Raehoshi: a different thing entirely

Raehoshi Illustrious XL (model ID 846917) is a merge that adds 8 author-trained characters by Raelina. The character list article explicitly states:

"This list contains only characters trained specifically by me. Characters natively supported by the Illustrious or NoobAI base models are not included here."

The list provides names only. To get accurate results you must manually add:

• Hair color and hairstyle
• Eye color
• Specific accessories or outfit details

Wildcard files and series-categorized versions are available in the article attachments and on rentry.org/xbbfsdua.

Source: Raehoshi article 23522, Raehoshi model page

Prompting formats that work

Illustrious XL: hybrid prompting

Write a sentence, then append Danbooru tags.

A girl sitting on a bench, hoshimi miyabi (zenless zone zero), solo, white hair, blue eyes, school uniform, pleasant smile

Quality tags (masterpiece, best quality, very aesthetic, absurdres) help but aren't mandatory.

Source: Illustrious XL prompting guide

NoobAI XL: quality-first tagging

Lead with aesthetic percentile tags, then character + artist.

score_9, score_8_up, score_7_up, furina (genshin impact), solo, hydro vision, white dress, blue hair, heterochromia

Source: NoobAI XL guide

Common errors and fixes

Sources: Civitai article 10242, Civitai article 9148, Civitai article 31875

When to train a LoRA instead of prompting

Train a LoRA when:

1. The character is Yellow or Red tier on the community list (or untested and prompting fails).
2. You need a specific outfit/version the base model doesn't capture (e.g., alternate ascension, seasonal costume).
3. You're merging a style + character and want clean separation.
4. You need consistent multi-angle generation for comics or reference sheets.

Community-proven LoRA settings (Kohya-ss)

Source: Civitai article 9148, Civitai article 31875

What we still don't know (and how to work around it)

FAQ

Which anime characters does Illustrious XL know without a LoRA?

Roughly 720 characters across 46 franchises tested on WAI-NSFW and Hoseki fine-tunes (v0.1). Major series: Fate, Blue Archive, Genshin Impact, Azur Lane, VTuber groups, Fire Emblem Engage, Danganronpa. No official list exists for v1.0/v2.0/v3.5.

How do I check if a character is native to NoobAI XL?

Load a NoobAI XL checkpoint (ε-pred or v-pred), prompt character_name (franchise), generate 4–8 images. NoobAI's e621 training often improves Western/furry characters compared to Illustrious.

When do I still need a character LoRA with Illustrious XL?

When the character rates Yellow (major outfit/color errors) or Red (generic person) on the community tier list. Also when the base model consistently misses a specific outfit variant.

What is the difference between Raehoshi trained characters and Illustrious native characters?

Raehoshi adds only 8 author-trained characters (v11). The list gives names only — you must add hair, eye, and outfit tags manually. Illustrious/NoobAI native characters come from Danbooru/e621 training data; Green-tier ones work from name + franchise alone.

Do I need to add 'pokemon (creature)' tag for Pokémon characters?

Yes. Community testing shows Pokémon and Digimon require pokemon (creature) or digimon (creature) to disambiguate from similarly tagged non-franchise images.

Does the Pony Diffusion XL character list apply to Illustrious?

The Illustrious v0.1 article says it's "very likely" because Illustrious has a more recent dataset, but this is explicitly labeled an inference, not a tested equivalence.

What prompting format works best for Illustrious XL native characters?

Hybrid: natural English sentence describing the scene, followed by comma-separated Danbooru tags. Example: A girl sitting on a bench, hoshimi miyabi (zenless zone zero), solo, white hair, blue eyes, school uniform.

Is there an official searchable database for Illustrious/NoobAI native characters?

No. Neither OnomaAI Research nor Laxhar Dream Lab publishes one. The only structured resource is the community Civitai article 10242 (720 characters, tested on fine-tunes). Verification requires manual prompting.

Sources

1. Illustrious XL official site — model architecture, dataset, prompting format, version history
2. NoobAI XL official site — model lineage, e621 addition, prompting guide, prediction lines
3. Raehoshi Illustrious XL model page — version history (v8→v11), author info
4. Raehoshi character list article — 8 trained characters, prompting requirements, wildcards, rentry link
5. ilxl character list article — 720 characters, 46 franchises, Civitai model links per character
6. Illustrious XL v0.1 recognized characters — tier system, testing methodology, Pony comparison note
7. Illustrious LoRA training discussion — LR ranges, dim/alpha, optimizer comparisons
8. Yeetyah's training guide — dataset prep, style LoRA settings, noise offset fix

Further reading

• Illustrious & NoobAI: LoRA training settings that work
• Danbooru tagging for SDXL: a practical cheatsheet
• Merging Illustrious fine-tunes without breaking character knowledge