Vice City Summer Pose - lora (Illustrious)

Vice City Summer Pose - lora [Illustrious]
============================================================

One pose. One hand on the head. One hand holding a drink. Golden hour. GTA 6 trailer vibe.

Hey everyone. This is a highly specific pose LoRA — the iconic pose from the GTA 6 trailer: a girl standing, one hand touching her hair/sunglasses on her head, the other hand holding a drink can. That's it. Nothing more, nothing less.


🏷️ TRIGGER WORD
============================================================

gta6_pose, 1girl, solo, standing, arm_up, one_hand_on_head, holding_beverage




🔗 PROMPT TEMPLATE
============================================================

Recommended full prompt:

gta6_pose, 1girl, solo, standing, arm_up, one_hand_on_head, holding_beverage, [your outfit], [your background], full body, masterpiece, best quality




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

LoRA weight:
0.65 - 1.0

RECOMMENDED weight :

0.7-0.75

Sampler: DPM++ 2M Karras
Steps: 20 - 30
CFG Scale: 5 - 7
Base model: Illustrious XL v2.0
Compatible: all Illustrious-based


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

worst quality, low quality, bad anatomy, bad hands, extra fingers, missing fingers, fused fingers, deformed hands, both hands up, hands on head, symmetrical pose

"both hands up" and "hands on head" are important — they reduce the generic "both hands up" pose that the base model defaults to at lower weights.


⚠️ KNOWN ISSUES
============================================================

1. Bare trigger word can be weak. Fix: bump weight to 1.2, or use the full trigger phrase.

2. Sometimes defaults to "both hands up" pose. This is the Illustrious base model's strong prior for that generic pose. Fix: raise weight, use full trigger, and make sure negative prompt includes "both hands up, hands on head, symmetrical pose".

3. Training backgrounds occasionally bleed through. The dataset was trained with varied backgrounds (pool, indoor, urban, plain). If backgrounds get messy, add "simple background, white background" and push CFG to 7.



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

Base Model: Illustrious XL v2.0 (illustriousXL20_v20)
Network: LoRA — Dim 16 / Alpha 16 (ratio 1.0)
LoRA+: loraplus_lr_ratio=16
Optimizer: AdamW, Cosine LR scheduler
Epochs: 8
Batch Size: 4
Resolution: 1024×1024 (bucket training)
Total Steps: 2848
Final Loss: 0.109
Captioning: WD14 ONNX tagger
Dataset: 178 images — varied characters, outfits, backgrounds


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

Illustrious XL v2.0
WAI-illustrious-SDXL-v17
Hassaku XL v22+
NoobAI-XL EPS
All Illustrious-base / merges



🎨 GENERATED SAMPLES
============================================================

All gallery samples use the full trigger phrase + recommended settings. Check the gallery for outfit, character, and background variations.


💬 FEEDBACK & COMMISSIONS
============================================================

If you run into issues or want another pose, drop a comment. I read all feedback.

Open for custom LoRA commissions — character, pose, style, outfit, whatever. DM me for inquiries.



💀 CREATOR'S NOTE
============================================================

This took 8 training runs, 5 days, and a GPU running at full load 24/7. Pose LoRAs are brutal — text captions have to describe a spatial position, and the base model fights you every step.

If this saved you time, drop a ❤️ or a buzz. It genuinely makes the grind worth it.