Sign In
homemodelsimagesvideospostsarticlesbountieschallengeseventsupdatesshop

V67 Female Faces – Global Beauty & Twin Versatility for Z-Image Turbo

0

angelomaiota's Avatar

angelomaiota

Bronze Tier Supporter Badge March 2026

Mar 28, 2026

(Updated: 2 days ago)

announcement
V67 Female Faces – Global Beauty & Twin Versatility for Z-Image Turbo

V67 Female Faces – Global Beauty & Twin Versatility for Z-Image Turbo

This LoRA component is a challenge to a certain https://civitai.com/user/Peli86 (Peli86) who created a LoRA component for each nation, I am convinced that one is more than enough, and the right prompt is enough. I want to let it be known that this guy also blocked me.

After weeks of training and careful curation, I'm excited to present my new LoRA for Z-Image Turbo: V67 Female Faces.

This model is designed to generate realistic, consistent portraits of young women (18–22 years) from diverse ethnic backgrounds, with special attention to cultural details and the ability to create identical twins. Best of all, thanks to Z-Image Turbo's distilled architecture, generations take just seconds—even on consumer hardware.

Download the model here:
👉 https://civitai.com/models/2499503/v67-female-faces

...

👉 Also for men: https://civitai.com/models/2499813/v67-male-faces

📌 Overview

V67 Female Faces is a LoRA trained on Z-Image-Turbo, the 6B-parameter distilled model from Alibaba's Tongyi Lab . Z-Image Turbo is optimized for ultra-fast inference (sub-second to a few seconds) while maintaining exceptional photorealistic quality and bilingual text rendering .

This LoRA allows you to generate high-quality female faces while maintaining:

  • Precise age range (18–22 years)

  • Distinct ethnic features across multiple global populations

  • Consistency for twins (identical facial structure, complementary expressions)

  • Fast generation — works with Z-Image Turbo's 8–15 step inference

The dataset includes over 200 curated images, balanced across the following main categories: Irish, Swedish, Japanese, and albino, plus additional ethnicities. The model excels at both close‑up portraits and twin compositions.

🧬 Feature Table

EthnicityKey TraitsIrishGinger/copper hair, fair skin with freckles, green/hazel eyes, Celtic vibeSwedishPlatinum blonde hair, very fair skin, light blue/gray eyes, Nordic minimalismJapaneseStraight black hair, pale to light skin, dark eyes, modern or traditional styleAlbinoWhite/platinum hair, very pale skin, light blue/violet eyes, ethereal lookTwinsIdentical features, symmetrical poses, complementary expressions

All subjects are in the 18–22 age range and are typically framed in close‑up portraits (headshot, shoulder‑up).

🌍 Examples by Ethnicity

Below are tested prompts for each main ethnicity. You can copy them into your Z-Image Turbo interface (ComfyUI, fal.ai, etc.) and vary the seed for different results. Important: Z-Image Turbo does not support negative prompts, so all constraints must be placed in the positive prompt .

🇮🇪 Irish – Single

text

close-up portrait of a 22-year-old Irish woman, bright ginger hair, pale skin with freckles, green eyes, soft smile, misty green hills background, golden hour light, V67 Female Faces, young adult, headshot

🇸🇪 Swedish – Single

text

shoulder-up shot of a 22-year-old Swedish woman, platinum blonde hair in loose waves, fair skin, light blue eyes, serene expression, Scandinavian forest backdrop, soft overcast light, V67 Female Faces, early twenties, face focus

🇯🇵 Japanese – Single

text

portrait of a 22-year-old Japanese woman, straight black hair with subtle highlights, modern streetwear, confident gaze, city night bokeh, neon reflections, V67 Female Faces, college age, close-up

👩🏻‍🦳 Albino – Single

text

close-up of a 21-year-old albino woman, platinum white hair, pale skin with very light eyelashes, light blue eyes, ethereal look, soft diffused light, V67 Female Faces, young adult, portrait

👯‍♀️ Twins – A Special Feature

One of the most appreciated capabilities of this LoRA is generating identical twins with consistent facial features, hairstyles, and skin tones. Here are prompts for the four main twin variations.

Irish Twins

text

two Irish twin women, identical, 21 years old, bright ginger hair, pale skin with freckles, green eyes, standing side by side, misty green hills background, soft overcast light, V67 Female Faces, identical twins, shoulder-up shot

Swedish Twins

text

two Swedish twin women, identical, 20 years old, platinum blonde hair, fair skin, matching light blue eyes, wearing minimalist white sweaters, symmetrical pose, birch forest background, V67 Female Faces, identical twins, portrait

Japanese Twins (Traditional Style)

text

two Japanese twin women, identical, with long black hair, wearing matching colorful kimonos, gentle smiles, cherry blossom garden, V67 Female Faces, identical twins, kimono, headshot

Albino Twins

text

two albino twin women, identical, with platinum white hair, pale skin, light blue eyes, symmetrical pose, ethereal lighting, white background, V67 Female Faces, identical twins, face close-up

🖼️ Additional Ethnicities

Beyond the four main groups, the model also handles other ethnicities thanks to the diversity of the dataset. Here are a few extra prompts:

EthnicityExample PromptItalianItalian woman with olive skin, long dark brown wavy hair, expressive brown eyes, elegant style, Roman pines background, V67 Female FacesEgyptian*Egyptian woman with dark almond-shaped eyes, black hair partially covered by a sheer scarf, warm skin tone, golden jewelry, desert dunes, V67 Female Faces*EthiopianEthiopian woman with deep brown skin, tight curly black hair, high cheekbones, traditional white kemis dress, African highlands, V67 Female FacesIranianIranian woman with long dark wavy hair, striking hazel eyes, defined eyebrows, delicate gold necklace, Persian garden, V67 Female FacesPeruvianPeruvian woman of Quechua descent, dark braided hair, colorful traditional aguayo shawl, Andean mountain landscape, V67 Female Faces

Z-Image Turbo has specific requirements due to its distilled architecture. Unlike other models, it does not use classifier-free guidance, meaning cfg_scale should always be set to 1.0 .

ParameterRecommended ValueNotesLoRA Weight0.7 – 1.0Start at 0.8, adjust based on desired influenceSteps8 – 158 steps for speed, 12–15 for maximum quality Guidance Scale (CFG)1.0Must be 1.0 — Z-Image Turbo doesn't support CFGResolution1024×1024Native resolution; up to 4MP supported SamplerEuler + beta or DPM++For photorealistic results

Important Inference Notes:

  • Negative prompts are not supported — include all constraints in your positive prompt

  • Use num_inference_steps=8 for speed, 12-15 for higher detail

  • For portrait work, set guidance_scale=1.0 (this is fixed for Turbo models)

🔧 Training Details

Training Z-Image Turbo LoRAs requires special care to preserve the model's "Turbo" acceleration. Direct fine-tuning can disrupt the distilled trajectory, causing the model to lose its 8-step generation capability .

This LoRA was trained using:

ParameterValueTraining MethodStandard SFT + DistillPatch at inference (Scheme 4) Base ModelZ-Image-Turbo (Tongyi-MAI)Dataset200+ images at 1024×1024 resolutionTraining Steps~3000 stepsLoRA Rank (r)16Learning Rate1e-4 VRAM Required12GB (comfortably)

Why This Matters: With this training approach, the LoRA retains Z-Image Turbo's signature speed. You can generate high-quality portraits in 8–15 steps without sacrificing the model's efficiency .

🏆 Civitai Cover Prize

If you use my model and your image is chosen for the CIVITAI cover, I will give you 500 Yellow Buzz, even if the image combines multiple models.

R+ rated images will be hidden from the gallery.

💡 Pro Tips for Z-Image Turbo

Based on community testing and official documentation, here are some tips for best results with this LoRA :

Prompt Structure

Z-Image Turbo responds best to clear, structured prompts. Put the main subject in the first 8–10 words, describe lighting and background in short phrases, and avoid stacking multiple conflicting styles .

For Portraits

text

ultra-detailed studio portrait photo of a young woman, soft window light, 50mm lens look, f/1.8, freckles, shallow depth of field, sharp eyes, realistic skin texture

Speed vs Quality Balance

  • Turbo mode (8 steps): Best for concept exploration, thumbnails, rapid iteration

  • Quality mode (12–15 steps): For final assets, client presentations, higher detail

Text Rendering

Z-Image Turbo handles short text well (1–3 words) but may struggle with longer phrases. For best results, keep text short and use ALL CAPS in the prompt .

🔗 Resources

💡 Final Notes

This model celebrates beauty and cultural diversity through AI art. The twin capability was achieved by training on both real and synthetic image pairs, ensuring facial consistency and symmetry.

Z-Image Turbo's distilled architecture makes it one of the fastest and most accessible high-quality models available today . Combined with this LoRA, you can generate diverse, age-specific portraits in seconds—even on consumer hardware with 12GB VRAM .

Have fun creating your own global character stories!
Share your creations and tag them with #V67FemaleFaces.

Sláinte! / Skål! / 乾杯! / Cheers! 🍀

🔗 Direct model link:
https://civitai.com/models/2499503/v67-female-faces

0