Tenka Izumo (Mato Seihei no Slave) — Illustrious SDXL Character LoRA

中文说明 | English (current)

An unofficial character LoRA for Tenka Izumo (出云天花) from Mato Seihei no Slave, trained on waiIllustriousSDXL v17. Formal E1–E7. Comparative evaluation recommends E6 at strength 0.8. Dataset: abbuibuibui/TenkaIzumo-Dataset.

Fan-made derivative, not official. Rights remain with the original owners.

Files

Epoch File Role
E1–E4 tenka_izumo-000001 … 000004 Early / alternate.
E5 tenka_izumo-000005.safetensors Matrix candidate.
E6 tenka_izumo-000006.safetensors Recommended default.
E7 tenka_izumo.safetensors Final epoch (unnumbered = E7).

No tenka_izumo-000007.safetensors.

E6 SHA-256: 94BBB01CEF240062C0EE25717809D7552014467A696B582F6EF2ABE41C2FF66A

Each file is 456493564 bytes.

Showcase

E6 @ 0.8, ComfyUI, Euler a / 28 / CFG 4.5 / CLIP skip 2 / 832×1216. No hires / ADetailer.

Face Main outfit full
Face Main
Alt outfit full Night street
Alt Night

No outfit token

Trigger Words and Settings

Captions used keep_tokens = 2; hair / eye color often omitted — add them.

Role Token Typical look
Identity tenka_izumo long black hair, purple eyes
Outfit tenka_izumo_combat_uniform Main eval (combat_uniform).
Outfit tenka_izumo_purple_turtleneck Alt eval (purple_turtleneck).

Example:

<lora:tenka_izumo-000006:0.8>, masterpiece, best quality, very aesthetic, absurdres, anime screencap, 1girl, solo, tenka_izumo, tenka_izumo_combat_uniform, long black hair, purple eyes, looking at viewer

Why E6

Round1 screened E1–E7 @ 0.8; Round3 matrix on E5–E7 × 10 scenes × 4 strengths.

Candidate Observed Decision
E5 Long black hair / purple eyes usable. Alternate.
E6 Portrait: long black hair, purple eyes, gold epaulettes, red V trim. Combat full body is trousers + knee boots + white gloves (commander silhouette, not the short-skirt cut of younger corps girls). Purple turtleneck alt is lavender turtleneck + black trousers + heels on a dirt path. Recommended @ 0.8.
E7 Close to E6. Final alternate.

Chinese notes: evaluation/横向测评记录.md.

Public Horizontal Evaluation

Round 1: E1–E7 @ 0.8

Front portrait Main full
r1a r1b

r1c

Round 3: E5 / E6 / E7 × 0.4 / 0.6 / 0.8 / 1.0

Face Three-quarter Profile
01 02 03
Main full Alt full Sitting
04 05 06
Simple Night No outfit
07 08 09

10

Training

  • Base waiIllustriousSDXL v17; LoRA rank 64 / alpha 32; AdamW bf16; buckets cap 1536; 7 epochs
  • Dataset: 69 pairs (Combat_Uniform / Purple_Turtleneck / Teal_Sleeveless_Blouse), repeats 2
  • Train / eval: NVIDIA DGX Spark + ComfyUI

Related

License

creativeml-openrail-m

Downloads last month
55
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train abbuibuibui/TenkaIzumo-Lora