Jujutsu Kaise Uniform — Female JJK Uniform Outfit LoRA (Illustrious)
============================================================

Jujutsu Kaisen uniform,

This is an outfit LoRA for female characters wearing a JJK-style uniform: a cropped navy jacket with gold buttons, pleated skirt, black belt, and pantyhose. It's not a character LoRA — it's purely the outfit. Works with any character LoRA or on its own.

Trained on Illustrious SDXL. Use this version if you're on Illustrious, NoobAI, or other Danbooru-based SDXL checkpoints. If you're on Pony, grab the Pony version instead — same trigger word, different base model.


🏷️ TRIGGER WORD & COMPANION TAGS
============================================================

Trigger word:
jjkf_uniform

The trigger word alone is NOT enough. You MUST include these companion tags:

jjkf_uniform, cropped jacket, high collar, pleated skirt, black jacket, black pantyhose, black skirt, belt, high-waist skirt, small_head, pouch on belt


⚙️ RECOMMENDED SETTINGS
============================================================

LoRA weight (model): 0.7 is the sweet spot
LoRA weight (CLIP): 1.0
Sampler: DPM++ 2M Karras or Euler A
Steps: 25 – 35
CFG Scale: 5.5

Tested checkpoints: WAI-illustrious-SDXL-v17 , Bismuth Illustrios v80

IMPORTANT — LoRA strength is contextual:
- Sweet spot: 0.7 — best balance of outfit accuracy and clean framing. Works for both solo and combined with character LoRAs.
- Above 0.7: crop behavior from the training dataset becomes increasingly visible. At 0.75+ the head cropping is noticeable. At 0.8+ it's very aggressive — heads get cut off consistently.

📝 POSITIVE PROMPT TEMPLATE
============================================================

masterpiece, best quality, amazing quality,
1girl, solo, jjkf_uniform, cropped jacket, high collar,
pleated skirt, black jacket, black pantyhose, black skirt,
belt, high-waist skirt, small_head,
[character tags: hair, eyes, expression],
[pose tags], simple background, full body


❌ NEGATIVE PROMPT
============================================================

lowres, bad anatomy, bad hands, text, error, cropped,
worst quality, low quality, blurry, ugly, male, 1boy,
head out of frame, faceless, neck cut off,
pants, trousers, trench coat, long skirt

Additional notes: head out of frame, faceless, neck cut off — the training dataset was cropped neck-down, so the LoRA can learn to crop heads at high strength. These tags prevent that.


⚠️ KNOWN PROBLEMS
============================================================

Trigger word is not standalone., this LoRA's trigger word must be paired with the full companion tag string. Without them, the outfit won't render cleanly — especially the cropped jacket and high collar details.

Head cropping above 0.7. The training dataset is 60 images cropped neck-down (outfit focus). Above 0.7 weight, crop behavior emerges — subtle at 0.75, aggressive at 0.8+. Fix: stick to 0.7, and include head out of frame, faceless, neck cut off in the negative prompt. Lower weight REDUCES cropping — do not push weight higher to "fix" anything, it makes cropping worse.

Outfit detail loss at low strength. Below 0.4 weight, the LoRA signal fades and the cropped jacket silhouette blurs into a generic top. Fix: don't go below 0.4 for solo use.

Color drift. The uniform can sometimes shift to slightly different tones under certain lighting conditions. Add any unwanted colors to the negative prompt. Training captions were filtered for consistent black/navy color tags.

CFG sensitivity. Sweet spot is exactly 5.5 — outfit renders cleanly with defined shadows and accurate details. Below 5, outfit detail softens. Above 6, contrast increases but cropping risk also rises. Stick to 5.5 for consistent results.

Character tags (hair, eyes) sometimes not followed. This is expected — the LoRA was trained on neck-down crops and knows nothing about faces or hair. Character appearance comes entirely from the base model + your prompt. Strengthen your character tags and add unwanted features to the negative prompt. Don't blame the LoRA for a face it never learned.


📊 TRAINING DETAILS
============================================================

Base Model: WAI-illustrious-SDXL-v17 (Illustrious SDXL)
Network: LoRA — Dim 16 / Alpha 8
Epochs: 8
Batch Size: 4
Optimizer: AdamW with Cosine LR scheduler
Learning Rate: 1e-4 (text encoder: 5e-5)
Resolution: 1024×1024 (bucket training)
Total Steps: 1,232
Final Loss: 0.092
Dataset: 60 images (neck-down crops, female JJK uniform focus)
Captioning: WD14 Tagger — Danbooru-style tags, color-filtered

Why Dim 16? Illustrious is much more receptive than Pony — new concepts stick easily. Dim 16 is enough to capture the cropped jacket, pleated skirt, and belt details cleanly. No need for Dim 32 like the Pony version.

Why only 8 epochs? Illustrious learns fast. The loss curve hit 0.092 at epoch 8 and started flattening. More epochs would overfit — the outfit would memorize backgrounds instead of learning the concept. Compare: the Pony version needed 15 epochs just to fight its military uniform prior.

Why text encoder LR 5e-5? Critical for trigger word stability. Without it, the LoRA might respond to random tag combinations instead of the actual trigger. Keep tokens set to 1 so the trigger word has room to breathe.

Why only 60 images? Outfit LoRAs don't need character or face variety. Variation comes from angles, poses, and lighting. 60 images is plenty for a single outfit — the LoRA needs to learn one thing and learn it well.


✅ COMPATIBLE WITH
============================================================

Illustrious SDXL, NoobAI, and most Danbooru-based SDXL anime checkpoints. Tested on WAI-illustrious-SDXL-v17. Not compatible with Pony SDXL — that needs the Pony version (check my other models).


💡 TIPS
============================================================

Combine with character LoRAs — this is the main use case. Same 0.7 weight works for both solo and combined use. Run your character LoRA at 0.7–0.8 and this outfit LoRA also at 0.7 — they work together without conflict.

Generate character datasets. Use this LoRA to generate images of female characters in the JJK uniform, then train them as character LoRAs. Useful for building a character series.

CFG 5.5 is the sweet spot. Below 5, outfit detail softens. Above 6, contrast increases and crop risk rises.

No score tags needed. Unlike Pony, Illustrious doesn't use the score_9 / score_8_up prefix system. Start your prompt with masterpiece, best quality instead.

If character tags are inconsistent (e.g., asking for black hair but getting blonde), that's the base model, not the LoRA. This outfit LoRA doesn't touch face or hair at all. Strengthen your character tags in the prompt.

For the cleanest results, generate at 1024×1024 or higher resolution. The cropped jacket and pleated skirt details render best at native resolution.


🎨 COMMISSIONS & SUGGESTIONS
============================================================

Got a character you want to see in the JJK uniform? Need a specific outfit variation or a custom clothing LoRA? I take suggestions — and yes, I do commissions too.

If you want your own custom LoRA trained (character, style, outfit, pose — 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.