🌸 zimage_nina_v0.1 — Realistic Asian Portrait Checkpoint

zimage_nina_v0.1 is a custom fine-tuned checkpoint built on top of

Z-Image Turbo

using a LoRA merge technique. This model is designed for high-quality, photorealistic portrait generation with a strong emphasis on East Asian facial aesthetics, natural skin tones, soft lighting, and expressive detail in eyes and hair.

This is version 0.1 — an early release to gather community feedback. Results are already clean and consistent at just 12 steps, making it a highly efficient model for portrait workflows.

✨ Model Strengths

📸 Photorealistic female portraits with natural skin & soft features

🎭 Versatile across multiple styles — casual, formal, dramatic, and night-scene lighting

💡 Works great with simple prompts — the base Z-Image Turbo instruct-following is preserved

⚡ Ultra-fast generation — only 12 steps needed

🌙 Handles low-light and neon/night-scene environments well

⚙️ Recommended Settings

Parameter Value

Parameter Value

Steps 12

CFG 1.0

Sampler Euler Ancestral

Scheduler Normal

Denoise 1.00

Base Model ZImageTurbo

🛠️ How to Use (ComfyUI)

Load the model as a diffusion model using the native ZImageTurbo node setup

Use qwen_3_4b.safetensors as your text encoder

Use ae.safetensors as your VAE

Set sampler to euler_ancestral, scheduler to normal, and keep CFG at 1.0

No negative prompt required at CFG 1.0

📝 Notes

This is an early experimental release (v0.1) — feedback welcome!

The model was created via LoRA merge technique on the ZImageTurbo base

Best results with portrait/person prompts; landscape prompts have not been extensively tested

Trained and tested in ComfyUI using the Efficiency Nodes workflow

🏷️ Suggested Tags

zimage, zimageTurbo, portrait, realistic, asian, photorealistic, lora-merge, female, comfyui