Download
1 variant available
380 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
20 1 2 3 4 5 6 7 8 9
(4)
Aug 29, 2026

810 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
2420 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
License:
Illustrious LicenseGUWEIZ — SDXL Style LoRA
============================================================
dark-cinematic by default because that's what it trained on — but you can push it bright and warm with daylight/warm tags (see tips). And it has a slight "house face" bias; describe the face strongly (eye shape, hair, color) or stack a character LoRA on top to keep identities distinct.
🏷️ TRIGGER
============================================================
gweiz
🎨 HOW TO USE
============================================================
Default — moody painterly:
gweiz, 1girl, solo, kimono, katana, snow, night, moonlight, cinematic lighting, atmospheric
Bright / soft — push away from dark:
gweiz, 1girl, solo, school uniform, bright, daylight, smiling, classroom, soft lighting, warm colors
Scenery — no humans:
gweiz, no humans, pagoda, night, moon, stars, snow, fog, atmospheric, painterly
Distinct faces — beat the house face:
gweiz, 1girl, solo, close-up, tsurime, sharp eyes, red eyes, short hair, adult
⚙️ RECOMMENDED SETTINGS
============================================================
LoRA weight: 0.5 to 0.7 (0.5 keeps faces free, 0.7 is strong style)
Sampler: DPM++ 2M, SGM Uniform
Steps: 25 to 35
CFG: 5 to 7
Resolution: 1024x1024, or 832x1216 for tall
⚠️ KNOWN ISSUES
============================================================
House face bias — without face descriptors it falls back to one painterly face. Fix: describe eyes/hair/expression, or stack a character LoRA on top (gweiz at 0.5).
Mooded dark — trained on night/snow/moon scenes, so it starts cinematic-dark. To brighten, add daylight, bright, warm colors, smile, and drop weight to 0.5.
📊 TRAINING DETAILS
============================================================
Network: LoRA — Dim 16 / Alpha 8
Epochs: 11
Batch Size: 4
Optimizer: AdamW with Cosine LR
Resolution: 1024 (bucket training)
Total Steps: 3,190
Final Loss: 0.149
Captions: WD14 Tagger ConvNeXt v2
💡 TIPS
============================================================
Mood tags matter more than subject. Add bright/daylight/warm to go light; night/moonlight/moody to go dark. The style follows the light, so flip the light to flip the mood.
🎨 COMMISSIONS & SUGGESTIONS
============================================================
Happy to take requests and tune prompts. Drop a comment or message me — I read everything.

