abbuibuibui/emilia-Lora

Emilia (Re:Zero) — Illustrious SDXL Character LoRA

中文说明 | English (current)

An unofficial character LoRA for Emilia from Re:Zero − Starting Life in Another World, trained on waiIllustriousSDXL v17 (commonly saved as waiIllustriousSDXL_v170). This repository includes complete E1–E7 weights. Comparative evaluation recommends E5 as the default, at strength 0.8. The training images and captions are in abbuibuibui/emilia-Dataset.

This is a fan-made derivative, not an official product. Rights to the character and source work remain with their respective owners.

Files

Download E5 for general use. The other six files are kept so other developers can compare epochs or pick a different trade-off. E3 is the most useful alternate when a more animated pose matters.

E5 SHA-256: BB2A04B46597D81D53DB73AA6E8D1ECBDE1DAAA7BD661559846E8E2B9FD5F8A7

Showcase

Both images are direct E5 @ 0.8 outputs with no retouching. Settings: Euler a, 28 steps, CFG 4.5, CLIP skip 2, 832×1216; Hires and ADetailer were off.

Trigger Words and Settings

This LoRA has an identity trigger only. There is no outfit token. The default look is the white-and-purple ceremonial dress; other outfits must be written in the prompt.

Recommended: waiIllustriousSDXL v17 or an Illustrious / NoobAI-family model; LoRA strength 0.8 (0.6 already identifies the character; ceremonial thighhighs are more stable at 0.8; 1.0 adds little); Euler a, 28 steps, CFG 4.5, CLIP skip 2, 832×1216.

If ADetailer is enabled on the face, put emilia_rezero plus silver hair, purple eyes, the white flower, purple ribbon, and pointy ears in the face prompt, or the face can drift.

Example:

<lora:Emilia_Rezero_Q-000005:0.8>,
masterpiece, best quality, very aesthetic, absurdres, anime screencap,
1girl, solo, emilia_rezero, long silver hair, purple eyes,
white flower hair ornament, purple hair ribbon, pointy ears,
white ceremonial dress, purple trim, high collar,
upper body, looking at viewer, outdoors, night, moonlight

Negative prompt:

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit,
fewer digits, cropped, worst quality, low quality, signature, watermark,
username, blurry, extra person, two girls, multiple girls

Why E5

E5 was not chosen because it was a later epoch. All seven checkpoints were first screened at strength 0.8 with the same recipe and fixed seeds on a front portrait and a ceremonial full-body shot (14 images). E1 looked stiff; E4 lost the thighhighs on the ceremonial full-body test. Only E3 / E5 / E7 advanced.

The second round generated one image per candidate in 9 fixed scenes at 4 strengths (0.4 / 0.6 / 0.8 / 1.0): 108 direct outputs. The comparison covered silver hair, purple eyes, the flower, ribbon, x-pin, pointy ears, and the white-and-purple dress; profile and full-body proportion; whether a cloak prompt was followed; sitting and night-scene identity; default-outfit leak without clothing words; and trigger isolation without emilia_rezero.

The full Chinese evaluation write-up is in `evaluation/横向测评记录.md`.

Public Horizontal Evaluation

Each sheet below is a real comparison from one scenario: columns are epochs; rows are LoRA strengths. Images are original outputs scaled into labeled grids; nothing was retouched or replaced.

Round 1: E1–E7 @ 0.8

Left to right: E1 through E7. This round only narrowed the candidates.

Round 2: E3 / E5 / E7 × 0.4 / 0.6 / 0.8 / 1.0

Left to right: E3, E5, E7. Top to bottom: 0.4, 0.6, 0.8, 1.0.

Identity and viewing angle

[image]

At 0.4 all three candidates look generic. From 0.6, E5 / E7 lock the high collar and green gem; E3 still often drops the ceremonial collar at 0.8. Profile shots keep the pointy ear and x-pin on all three.

Full body and outfit control

[image]

Ceremonial full-body shots at 0.4 / 0.6 often lose the thighhighs; from 0.8, E5 / E7 keep the full dress. All three follow a cloak prompt, so outfits can still be changed without an outfit token. E3 at 1.0 more often loses the boots.

Scene generalization

[image]

Identity holds in the night-forest tests. From 0.8, E5 / E7 keep more ceremonial detail without collapsing into a single training still.

Default-outfit leak

[image]

With only emilia_rezero and no clothing words, the default white-and-purple high-collar dress appears. That is expected for a single-outfit identity set. Other clothes must be written in the prompt.

Trigger isolation

[image]

With the LoRA loaded but without emilia_rezero, the controls are ordinary light-haired characters: no flower, x-pin, or ceremonial collar. The character binding stays on the trigger and does not obviously contaminate unprompted people.

Training

• —Data: 80 manually selected Emilia stills with matching .txt captions, published as abbuibuibui/emilia-Dataset. Captions start with emilia_rezero, keep_tokens = 1, remaining tags shuffled. No outfit token.
• —Network: LoRA rank 32 / alpha 16, UNet plus both text encoders.
• —Optimizer: AdamW8bit; UNet LR 1e-4, text-encoder LR 5e-5; cosine schedule, 56-step warmup.
• —Precision and resolution: bf16, 1024 buckets, batch size 1, latents cached to disk, sdpa, no_half_vae. Flip and color augmentation off.
• —Steps: repeat 2 per image, 7 epochs, 1120 optimizer steps (160 per epoch). One weight file saved per epoch.

Compatibility and Limits

This is a character LoRA, not a general style model. Evaluation used waiIllustriousSDXL v170, Euler a, and 832×1216. Other bases, samplers, prompts, or sizes can change the result.

Related repositories

• —Dataset: abbuibuibui/emilia-Dataset — the 80 image / caption pairs used for this LoRA.

License

• —Character and source-work rights remain with © Tappei Nagatsuki / White Fox / the relevant production committees.
• —The LoRA weights are released under CreativeML OpenRAIL-M.
• —Intended for non-commercial fan use. Generated images must also follow the license of the base model in use.
• —Do not use this model for unlawful, deceptive, harassing, or exploitative purposes.