Hugging Face's logo

svjack
/
Prince_Xiang_Z_Image_Turbo_Lora

Model card Files
xet
Community
YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/model-cards#model-card-metadata)

LoRA Model Card: svjack/Prince_Xiang_Z_Image_Turbo_Lora

Character Wang Xiang, Art Synthesis with Aesthetic Portraits

Fine-tuned Adapter:

  • Xiang_HairCut_Gls_ZIT_Lora_000004750.safetensors
  • Xiang_Photoshoot_ZIT_Lora_000005750.safetensors
  • Prince_Xiang_whbf_lora_000004500.safetensors
  • Prince_Xiang_slim_face_lora_000002000.safetensors
  • Xiang-Dream-ZIT-Lora_000001750.safetensors
  • Xiang_bottom_ZIT_Lora_000002000.safetensors
  • Prince_Xiang_noclothes_ZIT_Lora_000002000.safetensors
  • Xiang_Dream_nocloth_ZI_Lora_000002000.safetensors

Base Model: Z-Image-Turbo
Core Strengths:

  • Character Consistency: Maintains detailed facial features (fair skin, subtle facial contours) across diverse settings and lighting conditions.
  • Aesthetic Detailing: Captures delicate textures of skin, hair, and fabrics with nuanced lighting effects.
  • Dynamic Poses: Handles complex human poses with natural anatomical proportions and expressive gestures.
  • Adult Artistic Style: Sensual artistic style through sexy pose, fabric movement and lighting.

Inference Dataset: demonstrates inference samples of this lora model.

Optimized Text-to-Image Prompts

Example 1: Intellectual Elegance

Prompt: 

王翔戴着深蓝色细框窄边细眼镜，镜片后深邃的眼眸泛着微光，
浅灰色高领帽衫垂落的衣角随着他垂眸时的呼吸微微起伏，
露出锁骨处若隐若现的皮肤光泽。他指尖无意识地摩挲着帽衫褶皱，
腕表带滑过手背时带起一阵若有似无的香气，
垂落的发丝扫过锁骨凹陷处，喉结在吞咽动作中泛起微红，
睫毛在眼下投出细碎阴影，仿佛随时会从镜片后溢出灼热的凝视。

Key Characteristics:

  • Accessory Detailing: Glasses and watch are rendered with material accuracy
  • Lighting Control: Subtle shadows and highlights create depth
  • Texture Quality: Fabric folds and skin surfaces show high-frequency details

王翔头型帽衫

Example 2: Face Swap Grid Video

  • Xiang Reference Image

王翔冬装

Example 3: Peng and Xiang Kiss Grid Video

  • Peng Reference Image

image - 2026-01-31T025635.414

Example 4: Tease Grid Image

Inference Style Dataset: demonstrates inference samples of this lora model in tease style.

ComfyUI_temp_xuvsb_00002_

ComfyUI_temp_vxdne_00002_

ComfyUI_temp_usimx_00005_

Example 5: Tease Grid Video

Inference Style Dataset: demonstrates inference samples of this lora model in tease style.

Example 6: Tease Grid Video

Inference Style Dataset: demonstrates inference samples of this lora model in tease style.

Example 7: Furry Style Video

Example 8: Slim Style Grid

Inference Style Dataset: demonstrates inference samples of this lora model in glasses style.

ComfyUI_temp_henqo_00002_

Example 9: Glasses Style Grid

Inference Style Dataset: demonstrates inference samples of this lora model in glasses style.

ComfyUI_temp_cupzb_00005_

Example 10: Different Style Grid

Inference Style Dataset: demonstrates inference samples of this lora model in different style.

ComfyUI_temp_xetyg_00002_

Technical Parameters

Setting Recommendation Notes
Resolution 1280×2048 Optimal for portrait details
CFG Scale 7.0 Balances creativity & anatomical accuracy
Sampler DPM++ 2M Karras Preserves skin textures and fabric details
Steps 8-20 Ensures facial feature and clothing fidelity
LoRA Weight 0.9-1.1 (init with 1.0) Adjust for desired character strength

Performance Profile

  • VRAM Consumption: ~18-22GB (depending on base model)
  • Inference Speed: Optimized for Z-Image-Turbo's efficient architecture
  • Training Data: Focused on character consistency and aesthetic portraiture

Usage Guidelines

Trigger Words

  • Use "王翔" as the primary trigger for character consistency
  • Include descriptive terms for specific aesthetics (e.g., "soft lighting", "water droplets", "snow particles")

Weight Adjustment

  • 0.7-0.8: Subtle character influence, blends with base model style
  • 0.9-1.0: Strong character presence, maintains distinctive features
  • >1.0: Experimental, may enhance specific characteristics

Compatible Models

Designed for optimal performance with Z-Image-Turbo base model, but compatible with other realistic vision models through weight adjustment.

License

CC-BY-NC-SA 4.0 (Non-commercial, share-alike)
Community Hub: https://huggingface.co/svjack/Prince_Xiang_Z_Image_Turbo_Lora/discussions

Training Methodology

This LoRA adapter was trained using low-rank adaptation techniques focusing on:

  • Character Consistency: Maintaining stable identity and features across varied prompts. character has face:

捕获

  • Texture Quality: Enhanced skin, hair, and fabric rendering
  • Pose Variety: Supports multiple angles and expressions while preserving identity

The model demonstrates Z-Image-Turbo's capabilities in handling complex human portraits with aesthetic sensitivity, particularly excelling in lighting and texture reproduction.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including svjack/Prince_Xiang_Z_Image_Turbo_Lora