GOJO SATORU (FEMALE VER.) — SDXL LoRA [Illustrious]
============================================================
They say the strongest sorcerer in Jujutsu Kaisen couldn't possibly be a woman. Well, I made one anyway
Dear all. This is my first attempt at a genderbend character LoRA. Gojo Satoru, but make her a girl — with the same smug attitude and those iconic round shades. Five outfits, one consistent face. Trained from the ground up with hand-picked images.
It's not perfect yet. Some poses still come out weird, and occasionally a random pair of glasses sneaks into the nude shots. But honestly? She's got 90% of what I wanted. I'll keep improving.
🏷️ TRIGGER WORD
============================================================
gojofemale
Always pair with short hair and white_hair in every prompt — these two anchor the character design. One word trigger, two anchor tags.
👗 FIVE OUTFITS — HOW TO USE
============================================================
Pick the style you want. The uniform is the default — no glasses or bandage tags needed.
🕶️ ROUND SHADES
gojofemale, short hair, jjk uniform, black turtleneck, black jacket, black long pants, black boots, high_collar, white_hair, sunglasses, round_eyewear
🩹 BANDAGE OVER EYES
gojofemale, short hair, jjk uniform, black turtleneck, black jacket, black long pants, black boots, high_collar, white_hair, bandage over eyes, covered_eyes
👔 JJK HIGH-COLLAR UNIFORM
gojofemale, short hair, jjk uniform, black turtleneck, black jacket, black long pants, black boots, high_collar, white_hair
🖤 BLACK LINGERIE
gojofemale, short hair, black lingerie, white_hair
🔞 NUDE
gojofemale, short hair, nude, naked, no clothes, white_hair
Important tips:
Use sunglasses and round_eyewear for the iconic black shades. Use bandage over eyes and covered_eyes for the white blindfold. Always include short hair and white_hair in every prompt — short hair keeps the bob cut, white_hair keeps the color. Skip both and she might grow long brown hair. The uniform appears by default when you skip both glasses and bandage tags.
⚙️ RECOMMENDED SETTINGS
============================================================
LoRA weight: 0.7 to 1.0
Push higher if the face drifts, lower if the outfit overwhelms.
Sampler: DPM++ 2M Karras or Euler A
Steps: 25 to 35
CFG Scale: 5 to 7
❌ NEGATIVE PROMPT
============================================================
lowres, bad anatomy, bad hands, text, error, missing fingers, cropped, worst quality, low quality, jpeg artifacts, signature, watermark
For nude: add glasses, sunglasses, clothes, clothing, shirt, towel, fabric. The base model really likes giving her glasses when she shouldn't have any.
⚠️ KNOWN PROBLEMS
============================================================
Sometimes glasses bleed into the nude and lingerie styles. Negative prompt helps a lot.
Certain poses with arms crossed or hands near the face can produce anatomical mess. This is a general SDXL LoRA thing, not just this model.
If she looks too masculine, drop the weight to 0.7 and add genderswap or feminine to your prompt.
The uniform can occasionally look like a generic black jacket instead of the JJK one. Add high_collar and gojou_satoru to reinforce it.
🖼️ HOW TO GET PLAIN WHITE BACKGROUND
============================================================
This LoRA was trained on images with varying backgrounds. The model sometimes "remembers" them and bleeds backgrounds into your generations. Here is how to fight it.
Recommended baseline for clean studio shots:
gojofemale, short hair, white_hair, <outfit tags>, plain white background, simple background, solo, <pose tags>
Settings that help:
LoRA Weight: 0.5 – 0.65 (lower = less background memory)
CFG Scale: 5
Sampler: DPM++ 2M Karras
Steps: 25
Aggressive negative prompt for background:
background, scenery, room, gradient, blurry background, detailed background,
indoors, outdoors, building, wall, floor, ceiling, window, complex background,
messy background, furniture
If the background STILL bleeds at weight 0.65, drop to 0.5. At that weight the face may start to drift — compensate by adding 1girl and keep the outfit tags minimal. A LoRA trained on non-isolated datasets will always carry some background memory; the trick is to starve it with low weight and drown it with negative tags.
For the cleanest possible output, use ComfyUI with a background removal node after generation.
📊 TRAINING DETAILS
============================================================
Base Model: ntrMIXIllustriousXL_v777ForLora (Illustrious SDXL)
Network: LoRA — Dim 16 / Alpha 8
Epochs: 6
Batch Size: 4
Optimizer: AdamW8bit with Cosine LR
Resolution: 1024x1024 (bucket training)
Total Steps: 6,300
Final Loss: 0.0796
GPU: RTX 4070 Ti SUPER (16GB)
Images: 330 (curated from 1,130 generated)
Captions: WD14 Tagger v1.4 — Danbooru-style tags
I started with 1,130 images across 5 concept folders. Trimmed way down after the first training run was taking forever. Too many images and the model started memorizing backgrounds instead of learning the character. WD14 handled all tagging automatically. No manual captioning work needed.
✅ COMPATIBLE WITH
============================================================
Illustrious SDXL, NoobAI, and most other SDXL anime checkpoints. Tested on ntrMIXIllustriousXL and WAI-illustrious. Should be portable since it was trained on the ForLora variant.
============================================================
"Throughout Heaven and Earth, I alone am the honored one."
Still applies, even in a skirt.
💡 TIPS: USE WITH LORA MANAGER
============================================================
This LoRA is fully compatible with ComfyUI LoRA Manager's Trigger Word Toggle. Each style group will appear as a separate toggleable slot, so you can switch between outfits without retyping prompts.
After downloading from CivitAI through LoRA Manager, the 5 style trigger groups will auto-populate. Simply drop a Trigger Word Toggle node, connect it to your CLIP Text Encode, and toggle the style you want.
No more copy-pasting prompts. One click, one style.
🎨 COMMISSIONS & SUGGESTIONS
============================================================
Got a character you want to see genderbent? Need a specific pose or outfit variation? I take suggestions — and yes, I do commissions too.
If you want your own custom LoRA trained (character, style, pose, outfit — anything), hit me up. I'll work with you on dataset curation and training to get the result you want.
Drop a comment or DM if you're interested. Always happy to chat about new ideas.