abbuibuibui/ChikaFujiwara-Lora

Chika Fujiwara (Kaguya-sama) — Illustrious SDXL Character LoRA

中文说明 | English (current)

An unofficial character LoRA for Chika Fujiwara from Kaguya-sama: Love is War, trained on waiIllustriousSDXL v17 (commonly saved as waiIllustriousSDXL_v170). This repository includes every unique epoch file from four resume rounds. Comparative evaluation recommends R4E6 (chika_fujiwara_rE6S951-000006.safetensors) as the default, at strength 0.8. The training images and captions are in abbuibuibui/ChikaFujiwara-Dataset.

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

Files

Unnumbered finals are not SHA-256 duplicates of the numbered epoch next to them, so every file below is kept.

Label	File	Role
R1E1	chika_fujiwara-000001.safetensors	Alternate. Round 1: weak identity, pink hoodie, blue jumper instead of sailor.
R1E2	chika_fujiwara-000002.safetensors	Alternate. Round 1: odd black hair block and star bow.
R1Final	chika_fujiwara.safetensors	Alternate. Usable identity; flatter than later rounds. Unique SHA vs E1/E2.
R2E3	chika_fujiwara_rE3S476-000003.safetensors	Alternate. Hoodie full-body starts to lock; not advanced to the matrix.
R2Final	chika_fujiwara_rE3S476.safetensors	Alternate. Unique SHA vs R2E3.
R3E4	chika_fujiwara_rE4S581-000004.safetensors	Alternate. Usable; school pose a bit stiff.
R3E5	chika_fujiwara_rE4S581-000005.safetensors	Alternate. Livelier portrait; entered the weight matrix.
R3Final	chika_fujiwara_rE4S581.safetensors	Alternate. Unique SHA vs R3E5. Soft compared with R4E6.
R4E6	chika_fujiwara_rE6S951-000006.safetensors	Recommended default. Best balance of portrait, hoodie graphic, sailor uniform, sitting, and night street at 0.8.
R4Final	chika_fujiwara_rE6S951.safetensors	Alternate. Unique SHA vs R4E6. Nearly the same; 1.0 blooms extra hair/socks more easily.

Download R4E6 for general use. The other files stay so other developers can compare resume rounds.

R4E6 SHA-256: 8aba943186521bee0ebb98a41f13ff3ac7bd476bb6c4213c30a95a7422529ad7

Showcase

These images are direct R4E6 @ 0.8 outputs with no retouching. Settings: Euler a, 28 steps, CFG 4.5, CLIP skip 2, 832×1216; Hires and ADetailer were off.

Portrait	Hoodie full body

School uniform	Night street

Trigger Words and Settings

This LoRA has an identity trigger and outfit tokens. Captions used keep_tokens = 2.

Role	Token	Typical look
Identity	chika_fujiwara	Pink hair, blue eyes, side ponytail, black hair bow
Outfit	chika_hoodie	Light hoodie (often a chest graphic), skirt
Outfit	chika_school_uniform	Shuchiin sailor: white shirt, black skirt, red ribbon, kneehighs
Outfit	chika_casual	Casual looks from the set
Outfit	chika_dress	One dress still in the set

Recommended: waiIllustriousSDXL v17 or an Illustrious / NoobAI-family model; LoRA strength 0.8 (0.4 under-bakes hoodie color and school jumper; 0.6 already identifies her; 1.0 often adds little beyond sharper faces, extra socks, or extra dark hair); Euler a, 28 steps, CFG 4.5, CLIP skip 2, 832×1216.

If ADetailer is enabled on the face, put chika_fujiwara plus pink hair, blue eyes, and side ponytail in the face prompt, or the face can drift.

Example:

<lora:chika_fujiwara_rE6S951-000006:0.8>,
masterpiece, best quality, very aesthetic, absurdres, anime screencap,
1girl, solo, chika_fujiwara, chika_hoodie,
pink hair, blue eyes, side ponytail, black hair bow,
full body, looking at viewer, outdoors

Negative prompt:

lowres, bad, 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,
blonde hair, brown hair, black hair, green eyes, purple eyes, red hair

Why R4E6

R4E6 was not chosen because it was a later filename. All ten unique checkpoints were first screened at strength 0.8 with the same recipe and fixed seeds on a front portrait, a hoodie full-body shot, and a school-uniform full-body shot. Early R1 files missed sailor/hoodie lock. Only R3E5 / R4E6 / R4Final advanced.

The second round generated one image per candidate in 10 fixed scenes at 4 strengths (0.4 / 0.6 / 0.8 / 1.0): 120 direct outputs. The comparison covered pink hair / blue eyes / side ponytail; hoodie vs sailor control; profile and sitting; night-street identity; default clothes without an outfit token; and trigger isolation without chika_fujiwara.

Candidate	Observed result	Decision
R3E5	Livelier portrait; hoodie graphic and night chest print weaker than R4E6.	Alternate.
R4E6	Most consistent identity across portrait, hoodie graphic, sailor, sitting, and night tests at 0.8.	Recommended default at 0.8.
R4Final	Nearly the same as R4E6, with more bloom at 1.0.	Archived alternate.

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: all epochs @ 0.8

Front portrait	Hoodie full body	School full body

Round 3: R3E5 / R4E6 / R4Final × 0.4 / 0.6 / 0.8 / 1.0

Front portrait	Three-quarter view

Hoodie full body	School full body

Training

• Data: 42 manually selected Chika stills with matching .txt captions. Captions start with chika_fujiwara, <outfit>, keep_tokens = 2, remaining tags shuffled. Outfit counts: hoodie 17 / school 12 / casual 12 / dress 1. Published as abbuibuibui/ChikaFujiwara-Dataset.
• Network: LoRA rank 32 / alpha 16, UNet plus both text encoders.
• Optimizer: AdamW8bit; UNet LR 1e-4, text-encoder LR 5e-5; cosine schedule.
• Precision and resolution: bf16, 1024 buckets, batch size 1, latents cached to disk, sdpa, no_half_vae. Flip and color augmentation off.
• Steps: repeat 4 per image, 7 epochs per round. The last resume (chika_fujiwara_rE6S951) ran 1176 optimizer steps (42 × 4 × 7). Earlier rounds used the same dataset and recipe with different output_name prefixes. Learning-rate values above come from the shared kink-panel trainer command, not from a printed LR line in train.log.

Related repositories

• Hugging Face dataset: abbuibuibui/ChikaFujiwara-Dataset
• ModelScope model: abbuibuibui/ChikaFujiwara-Lora
• ModelScope dataset: abbuibuibui/ChikaFujiwara-Dataset

Compatibility and Limits

Base model	Result
waiIllustriousSDXL (any version)	Best. Training-family checkpoint.
Illustrious-XL / NoobAI-XL and merges	Usable; try 0.7–0.9.
Pony Diffusion V6 XL	Loads, but identity can drift.
Generic SDXL 1.0	Poor.
SD 1.5 / SD 2.x	Incompatible.

This is a character LoRA, not a general style model. Evaluation used waiIllustriousSDXL v170, Euler a, and 832×1216.

Known limits from the public matrix:

• Strength 0.4 often keeps a pink hoodie and a green jumper instead of the later teal hoodie / black sailor.
• Profile shots at 0.6–1.0 frequently collapse the hair to a dark color (Shuchiin cast leak). Use a front or three-quarter view, or add pink hair more strongly.
• Without an outfit token the default is not sailor; you usually get a light short dress. Write other clothes explicitly.
• The chika_dress token has only one training still.

License

• Character and source-work rights remain with © Aka Akasaka / Shueisha / 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.