abbuibuibui/HayasakaAi-Lora

Ai Hayasaka (Kaguya-sama) — Illustrious SDXL Character LoRA

中文说明 | English (current)

An unofficial character LoRA for Ai Hayasaka from Kaguya-sama: Love is War, 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/HayasakaAi-Dataset.

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

Files

Epoch	File	Role
E1	ai_hayasaka-000001.safetensors	Alternate. Screened out in round 1: school full-body misses the sweater around the waist.
E2	ai_hayasaka-000002.safetensors	Alternate. Screened out in round 1: early identity, unfinished school details.
E3	ai_hayasaka-000003.safetensors	Alternate. Usable identity; school sweater starts to appear; softer than E5/E7.
E4	ai_hayasaka-000004.safetensors	Alternate. Screened out in round 1: mid-pack portrait detail.
E5	ai_hayasaka-000005.safetensors	Recommended default. Best balance of portrait, maid, school, casual, sitting, and night scenes at 0.6–0.8.
E6	ai_hayasaka-000006.safetensors	Alternate. Not advanced to the weight matrix. Nearly the same as E7.
E7	ai_hayasaka.safetensors	Alternate. Final-epoch filename has no -000007. Close to E5; 1.0 blooms or swaps socks more easily.

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 lighter, less baked look matters.

E5 SHA-256: 7CFA62E199E61B49E7D05F7392296CBBA810339E8DF4CC9E310A84FB52158997

Showcase

These 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.

Maid portrait	Maid full body

School uniform	Casual

Trigger Words and Settings

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

Role	Token	Typical look
Identity	ai_hayasaka	Blonde hair, blue eyes, side ponytail, mole under the eye
Outfit	hayasaka_maid	Black-and-white maid dress, frilled apron
Outfit	hayasaka_school_uniform	White shirt, black skirt, sweater tied around the waist, kneehighs
Outfit	hayasaka_casual	Casual blouse / skirt looks from the set

Recommended: waiIllustriousSDXL v17 or an Illustrious / NoobAI-family model; LoRA strength 0.8 (0.6 already identifies the character; school sweater and maid apron are more stable at 0.8; 1.0 often adds little beyond sharper faces and occasional sock swaps); Euler a, 28 steps, CFG 4.5, CLIP skip 2, 832×1216.

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

Example:

<lora:ai_hayasaka-000005:0.8>,
masterpiece, best quality, very aesthetic, absurdres, anime screencap,
1girl, solo, ai_hayasaka, hayasaka_school_uniform,
blonde hair, blue eyes, side ponytail,
white shirt, black skirt, sweater around waist,
upper body, looking at viewer, outdoors

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 school-uniform full-body shot (14 images). E1/E2 missed the waist sweater; E4 was middling. Only E3 / E5 / E7 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 blonde hair, blue eyes, side ponytail; maid / school / casual control; profile and sitting; night-street identity; default-outfit leak without clothing words; and trigger isolation without ai_hayasaka.

Candidate	Observed result	Decision
E3	Usable identity; softer outfit detail than E5/E7.	Alternate.
E5	Most consistent identity across portrait, maid, school, casual, sitting, and night tests at 0.6–0.8.	Recommended default at 0.8.
E7	Nearly the same as E5, with more bloom / sock swaps at 1.0 and no stable gain.	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: E1–E7 @ 0.8

Front portrait	School full body

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

Front portrait	Three-quarter view

Maid full body	School full body

Training

• Data: 35 manually selected Hayasaka stills with matching .txt captions. Captions start with ai_hayasaka, <outfit>, keep_tokens = 2, remaining tags shuffled. Outfit counts: maid 13 / school 13 / casual 9. Published as abbuibuibui/HayasakaAi-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, 980 optimizer steps (140 per epoch). One weight file saved per epoch.

Related repositories

• Dataset: abbuibuibui/HayasakaAi-Dataset — the 35 image / caption pairs used for this LoRA.

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. Without an outfit token, the school look often returns at mid-to-high strength. Write other clothes explicitly.

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.