abbuibuibui/HayasakaAi-Dataset

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

中文说明 | English (current)

This is the actual training subset used to train abbuibuibui/HayasakaAi-Lora, an unofficial Ai Hayasaka character LoRA on waiIllustriousSDXL v17. The files here are a copy of the directory named in dataset.toml (image_dir = .../_扩充终选_待打标_20260908/早坂爱). Captions were not rewritten for this release. Latent caches, WD14 logs, and review sheets that sat in the same folder are not included.

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

Contents

Item	Count
Training images (images/*.png)	35
Matching captions (images/*.txt)	35
Missing same-name captions	0
Orphan captions	0
num_repeats	4

Filenames are unchanged (for example 1030.png / 1030.txt). Captions are comma-separated Danbooru-style tags with the identity and outfit triggers fixed first:

ai_hayasaka, hayasaka_school_uniform, 1girl, solo, looking at viewer, smile, open mouth, full body

Outfit token	Images
hayasaka_maid	13
hayasaka_school_uniform	13
hayasaka_casual	9
Total	35

• Identity trigger: ai_hayasaka (blonde hair, blue eyes, side ponytail, mole under the eye)
• Outfit tokens: hayasaka_maid / hayasaka_school_uniform / hayasaka_casual
• Training used shuffle_caption = true and keep_tokens = 2, so the first two tokens stay pinned.
• Tags were manually reviewed per image.
• images/metadata.jsonl maps all 35 filenames to those captions for the Dataset Viewer. A “server busy” state right after upload is Hub indexing, not a missing file.

Preview

These eight images are copies of real training stills, not model outputs. They cover the face, school uniform, full body, profile, sitting, maid, and a night scene.

Face	School face	School full body

Profile	Sitting	Maid

Maid expression	Night

How it was trained

• Base: waiIllustriousSDXL v17, 1024 buckets, bucket_no_upscale
• Network: LoRA rank 32 / alpha 16, UNet plus both text encoders
• Optimizer: AdamW8bit; UNet LR 1e-4, text-encoder LR 5e-5; cosine, 7 epochs, 980 steps
• 35 × repeats 4 = 140 optimizer steps per epoch
• Recommended model from that run: E5 (ai_hayasaka-000005.safetensors) at strength 0.8

Related repositories

• Model: abbuibuibui/HayasakaAi-Lora — E1–E7 weights, bilingual card, and the public epoch × strength evaluation.

License

• Character and source-work rights remain with © Aka Akasaka / Shueisha / the relevant production committees.
• Dataset files are released under CreativeML OpenRAIL-M.
• Intended for non-commercial fan LoRA training and research.
• Do not use this dataset for unlawful, deceptive, harassing, or exploitative purposes.