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/Megumin-Lora")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]Megumin (KonoSuba) — Illustrious SDXL Character LoRA
ä¸æ–‡è¯´æ˜Ž | English (current)
An unofficial character LoRA for Megumin from KonoSuba: God's Blessing on This Wonderful World!, trained on waiIllustriousSDXL v17. Publishes formal E1–E7. Recommended default: E6 (megumin-000006.safetensors) at strength 0.8. Dataset: abbuibuibui/Megumin-Dataset.
Fan-made derivative; character rights remain with their owners.
Files
| Label | File | Role |
|---|---|---|
| E1 | megumin-000001.safetensors |
Alternate. Early. |
| E2 | megumin-000002.safetensors |
Alternate. Early. |
| E3 | megumin-000003.safetensors |
Alternate. |
| E4 | megumin-000004.safetensors |
Alternate. |
| E5 | megumin-000005.safetensors |
Alternate. Matrix candidate. |
| E6 | megumin-000006.safetensors |
Recommended default at 0.8. |
| E7 | megumin_rE7S859.safetensors |
Alternate final formal epoch (resume). |
megumin.safetensors (first-round leftover, unique SHA) is not published.
E6 SHA-256: 38dde62b536f832ab2e2c4c42d51daa6c31a2c5d6b0d8c2857a67884c5e18659
Showcase
Direct E6 @ 0.8 outputs. Euler a, 28 steps, CFG 4.5, CLIP skip 2, 832×1216.
| Portrait | Adventurer outfit |
|---|---|
 |
 |
| Alt outfit token | Night street |
|---|---|
 |
 |
Trigger Words and Settings
| Role | Token |
|---|---|
| Identity | megumin_konosuba (short brown hair, red eyes, eyepatch, witch hat) |
| Main outfit | megumin_konosuba_adventurer_outfit (35/70) |
| Alt outfit | megumin_konosuba_crimson_demon_school_outfit (25/70) |
| Other | casual / gym / hotspring / sleepwear (small counts) |
Recommended strength 0.8. Example:
<lora:megumin-000006:0.8>,
masterpiece, best quality, very aesthetic, absurdres, anime screencap,
1girl, solo, megumin_konosuba, megumin_konosuba_adventurer_outfit,
short brown hair, red eyes, eyepatch, witch hat,
full body, looking at viewer, outdoors
Why E6
Seven formal epochs screened @ 0.8 (face + adventurer + school tokens). E5–E7 entered the weight matrix (10 scenes × 0.4/0.6/0.8/1.0). E6 best overall balance at 0.8; E7 close. School-token scenes often still lean adventurer when hat identity is present — prompt harder for the school set or drop hat tags when needed.
Full notes: evaluation/eval_notes_zh.md.
Public Horizontal Evaluation
Round 1 @ 0.8
| Front portrait | Adventurer | School token |
|---|---|---|
 |
 |
 |
Round 3: E5 / E6 / E7 × weights
| Face | Three-quarter |
|---|---|
 |
 |
Training
- 70 stills across 6 outfit folders; captions start with
megumin_konosuba, <outfit>, …;keep_tokens=2, repeats=2. - LoRA dim 32 / alpha 16, UNet + text encoders; AdamW8bit; UNet LR
1e-4, TE LR5e-5; bf16, 1024 buckets. - 7 epochs; E7 from resume
megumin_rE7S859.
Related repositories
- HF model: abbuibuibui/Megumin-Lora
- HF dataset: abbuibuibui/Megumin-Dataset
- ModelScope model: abbuibuibui/Megumin-Lora
- ModelScope dataset: abbuibuibui/Megumin-Dataset
Compatibility and Limits
Best on waiIllustriousSDXL / Illustrious family. Adventurer outfit is the strongest token. Crimson Demon school outfit needs explicit outfit prompting; identity hat tags can pull back to the adventurer look.
License
CreativeML OpenRAIL-M for weights. Character © original rights holders. Non-commercial fan use.
- Downloads last month
- -