White Tea Moonlight Z-Image-Turbo Pure Version fp16

Introduction to the "White Tea Moonlight" Z-image Fine-Tuned Pure Style Base Model

1. Core Model Positioning
The "White Tea Moonlight" Z-image Fine-Tuned Version is a specialized model deeply optimized based on the original pure style base model using Z-image feature transfer technology. Focusing on "enhanced portrait consistency + improved scene fusion precision," it retains the original three core styles: "light-faced ethereal vibe, first-love fragility, and everyday relaxed feel," while solving common pain points such as "facemismatch when changing scenes, blurry detail restoration, and style drift." It is tailor-made for generating portrait-grade pure portraits in batches, short-video cover matrices, and fixed virtual persona appearances in bulk, while being compatible with mainstream local deployment platforms such as ComfyUI and Stable Diffusion.

2. Core Technical Advantages of Z-image Fine-Tuning
1. Portrait Feature Locking Capability
By importing Z-image feature vectors from 3-5 reference images, it can precisely anchor facial feature baselines (such as almond-eye curvature, rounded nose tip, and lip thickness). When generating images in different scenes and outfits, the consistency of the character's core facial aesthetics is improved by over 85%, completely bidding farewell to the issue of "same model, different faces."
It supports rapid fine-tuning with low sample sizes without the need for large-scale training data. Ordinary high-performance computers (like the 5090 graphics card) can complete personalized feature injection within 30 minutes, offering high user-friendliness for beginners.

2. Scene and Character Fusion Optimization
The fine-tuned model strengthens the linkage logic between "character, scene, and lighting." Scene elements entered in prompt words (such as seaside sunset or cafe warm light) will automatically match the character's lighting and shadow effects (such as backlighting golden rims on hair, and facial soft light diffusion), avoiding the awkward sense of "disconnection between character and background." Compatible with ControlNet skeleton binding and IP-Adapter style transfer, it can overlay gesture and posture control to achieve three-dimensional generation of "fixed appearance + free movement + precise scene."
3. Style Purity and Detail Enhancement
Filtering out redundant data prone to style drift during training, the purity of the pure style is increased by 90%, preventing the mixing of non-target elements such as heavy makeup and bright colors, ensuring generated images strictly adhere to the style definition of "pseudo-bare makeup, low saturation, and soft texture." It optimizes the generation of skin texture, hair fluffiness, and fabric texture, clearly restoring sweater textures, the airy feel of long hair, and the light-transmitting feel of white shirts, reaching a portrait-grade output standard.

3. Core Character Generation Specifications (Post-Fine-Tuning Enhanced Version)
1. Facial Features and Aesthetic Baseline (Non-Drift Items)
- Feature Classification | Core Parameters | Fine-Tuning Enhancement Points
- Face Shape | Oval face / slight baby fat, melon seed face, soft jawline | Lock facial proportions to avoid extreme thinning or round face distortion.
- Eyebrows & Eyes | Almond eyes / doe eyes, light brown / amber pupils, natural wild eyebrows | Enhance the innocent look of drooping outer corners of the eyes; precisely match eyebrow thickness and hair color.
- Nose & Lips | Upturned cute nose (rounded tip), thin soft lips (nude pink / bean paste color) | Optimize the realism of nose tip highlights and lip gloss mirror texture.
- Skin | Cool fair soft-light skin, pseudo-bare makeup base, minor freckles allowed | Enhance skin translucency and avoid the plastic look caused by excessive skin smoothing.

2. Makeup and Styling Adaptation (Controllable Variable Items)
- Makeup: Strictly limited to the pseudo-bare makeup category; supports detail toggles such as "nose-tip blush, light lip gloss, and natural eyelashes," while heavy eyeshadows, thick eyelines, and sequins are prohibited.
- Hairstyle: Three basic styles—long curly hair with air bangs, fluffy high ponytail, and collarbone-length hair. Hair direction and fluffiness can be precisely controlled via prompts.
- Outfit: Low-saturation color palettes (milky white, light blue, oatmeal, mint green) + soft fabrics (cotton-linen, knitwear, chiffon). Supports style switching between campus, casual, and forest styles. Fabric folds and lighting effects are automatically optimized.

4. Adapted Scenes and Use Cases
1. Priority Adapted Scenes (Best Effects After Fine-Tuning)
- Everyday categories: Window-side desk, cafe corner, convenience store entrance, school playground.
- Outdoor categories: Seaside sunset, park picnic, rooftop evening glow, cherry blossom lawn.
- Literary categories: Bookstore bookshelf, library study area, florist entrance, film photography studio.

2. Core Application Scenarios
- Content creation: Xiaohongshu pure-style graphics, Douyin short-video cover matrix, Bilibili virtual streamer persona images.
- Design demands: Novel youth campus themed illustrations, e-commerce clothing model images (fresh style), personal customized photo albums.
- Local deployment: Supports 8K ultra-clear output; combined with ComfyUI workflows, it enables batch image generation and style iteration.

5. Local Deployment Recommendations
- Fine-tuning data preparation: Prepare 3-5 clear reference portrait images (1 front, 1 side, 1 semi-side are best), ensuring even lighting and unobstructed facial features.
- Prompt writing formula: Model name + character features + outfit + scene + lighting + detail requirements, for example: White Tea Moonlight Z-image version, light-faced pure girl, long curly hair with air bangs, milky white knit sweater + light blue jeans, cafe warm light corner, soft hair light, translucent skin, 8K ultra-clear.
- Negative prompt: Add words such as heavy makeup, bright colors, blur, deformed, hard light, plastic look, and heavy foundation to further improve generation quality.