Instructions to use abbuibuibui/FukaKikuchi-Lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use abbuibuibui/FukaKikuchi-Lora with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("waiIllustriousSDXL-v17", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("abbuibuibui/FukaKikuchi-Lora") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
Fuka Kikuchi (Bottom-tier Character Tomozaki) — Illustrious SDXL Character LoRA
中文说明 | English (current)
An unofficial character LoRA for Fuka Kikuchi (菊池风香) from Bottom-tier Character Tomozaki, trained on waiIllustriousSDXL v17. This repository includes formal E1–E7 weights. Comparative evaluation recommends E6 as the default, at strength 0.8. Training images and captions: abbuibuibui/FukaKikuchi-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 | fuka_kikuchi-000001.safetensors |
Early. Ash-brown hair already lands. |
| E2 | fuka_kikuchi-000002.safetensors |
Alternate. |
| E3 | fuka_kikuchi-000003.safetensors |
Alternate. |
| E4 | fuka_kikuchi-000004.safetensors |
Alternate. |
| E5 | fuka_kikuchi-000005.safetensors |
Matrix candidate. Vest uniform is clean; blazer full-body at 0.8 can spawn a second copy of her. |
| E6 | fuka_kikuchi-000006.safetensors |
Recommended default. |
| E7 | fuka_kikuchi.safetensors |
Final epoch. Face close to E6; blazer full body may add a yellow sweater under the blazer and still risk a double. |
There is no fuka_kikuchi-000007.safetensors. The end-of-training save is the unnumbered file, and it is E7.
Download E6 for general use. The other files are kept for comparison.
E6 SHA-256: C7C69CB978B2B3B2231470DD66FB67B9A52EC5E4D6E635DA1928156363615BEE
Each file is 456495956 bytes.
Showcase
Direct E6 @ 0.8 outputs, no retouching. ComfyUI, Euler a, 28 steps, CFG 4.5, CLIP skip 2, 832×1216. Hires and ADetailer off.
Note on the blazer full-body showcase: Round 1 s04_main_full at 0.8 for E5–E7 often draws two Kikuchis in one frame (foreground lean-in + standing copy). The night-street and vest-uniform cells are single-subject. Keep solo / 1girl strong and consider regenerating blazer full-body shots if you hit the double.
Trigger Words and Settings
Captions used keep_tokens = 2. Training captions often omit hair / eye color. Identity words below were read from the training stills. Close-ups sometimes push the eyes toward olive / hazel even when you ask for brown — keep brown eyes and put green eyes in the negative if that drift bothers you. A small ahoge is common on late epochs.
| Role | Token | Typical look |
|---|---|---|
| Identity | fuka_kikuchi |
long messy ash-brown hair, bangs, brown / hazel eyes, quiet face |
| Outfit | fuka_kikuchi_blazer_uniform |
Navy blazer, white shirt, large red bow, purple-blue plaid skirt, dark knee socks, brown loafers. Main eval outfit. |
| Outfit | fuka_kikuchi_vest_uniform |
Navy sweater vest with crest, white shirt, red bow, blue plaid skirt, white mid-calf socks. Alt eval outfit — usually the cleaner full-body read. |
| Outfit | fuka_kikuchi_casual_cardigan / fuka_kikuchi_maid_dress / fuka_kikuchi_sailor_casual |
Other trained outfits |
Recommended: waiIllustriousSDXL v17 or an Illustrious / NoobAI merge; strength 0.8; Euler a, 28 steps, CFG 4.5, CLIP skip 2, 832×1216. Prefer the vest token when you want a reliable single full-body school look.
Example:
<lora:fuka_kikuchi-000006:0.8>, masterpiece, best quality, very aesthetic, absurdres, anime screencap, 1girl, solo, fuka_kikuchi, fuka_kikuchi_vest_uniform, long hair, brown hair, brown eyes, looking at viewer
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, black hair, pink hair, blue hair, green hair, short hair, twintails, blue eyes, green eyes, red eyes, purple eyes
Why E6
All seven epochs were screened at 0.8 on a front portrait, blazer-uniform full body, and vest-uniform full body. E5 / E6 / E7 went into the weight matrix (10 scenes × 4 strengths).
| Candidate | Observed result | Decision |
|---|---|---|
| E5 | Face is the quiet ash-brown look. Vest full body is a clean single subject. Blazer full body at 0.8 can duplicate her. | Matrix alternate. |
| E6 | Portrait stays consistent. Vest uniform at 0.6–1.0 stays a single girl with crest vest + red bow. Night street at 0.8 is one subject under shop signs. Blazer full body still risks the double — treat that as a scene issue, not a reason to prefer E7. | Recommended at 0.8. |
| E7 | Face close to E6. Blazer full body may insert a yellow sweater under the blazer and still double. | Final-epoch alternate. |
Without an outfit token, 0.8 becomes a sleeveless white dress and white strap flats — casual, still the same hair and quiet face. Without the name trigger, mid strengths leak this ash-brown look.
Chinese notes: evaluation/横向测评记录.md.
Public Horizontal Evaluation
Sheets use columns = epochs, rows = LoRA strengths. Original outputs only. Eval images were generated on a DGX with ComfyUI, then copied back. No retouching.
Round 1: E1–E7 @ 0.8
Round 3: E5 / E6 / E7 × 0.4 / 0.6 / 0.8 / 1.0
Training
- Base: waiIllustriousSDXL v17
- Network: LoRA rank 64, alpha 32
- Optimizer: AdamW, bf16
- Resolution: SDXL buckets, long edge cap 1536
- Epochs: 7
- Dataset: 81 image–caption pairs (Blazer_Uniform / Vest_Uniform / Casual_Cardigan / Maid_Dress / Sailor_Casual), repeats 2
- Train host: NVIDIA DGX Spark
- Eval host: same DGX, ComfyUI (Euler a / 28 / CFG 4.5 / CLIP skip 2 / 832×1216)
Related
- Hugging Face model: https://huggingface.co/abbuibuibui/FukaKikuchi-Lora
- Hugging Face dataset: https://huggingface.co/datasets/abbuibuibui/FukaKikuchi-Dataset
- ModelScope model: https://www.modelscope.cn/models/abbuibuibui/FukaKikuchi-Lora
- ModelScope dataset: https://www.modelscope.cn/datasets/abbuibuibui/FukaKikuchi-Dataset
Compatibility and Limits
Illustrious / NoobAI lineage checkpoints work best. Same series as Aoi Hinami / Minami Nanami — keep the name trigger. Watch for two girls on outdoor blazer full-body prompts; vest uniform and night street were cleaner singles in this eval.
License
creativeml-openrail-m
- Downloads last month
- 90

















