Life Illustration LoRA for Z-Image-Turbo

A custom-trained LoRA that generates Laerdal Life Illustration style images—flat vector art with faceless characters used in medical and healthcare training materials.

---

Quick Start

# Using ZImageCLI (Mac)
ZImageCLI \
  -p "life_illustration, healthcare worker standing in clinic" \
  --lora life_illustration_zit_v2.safetensors \
  --lora-scale 1.0 \
  -s 9 -o output.png

Trigger Word: life_illustration

---

Requirements

Base Model

• Z-Image-Turbo (Tongyi-MAI)
  • Distilled FLUX model optimized for 8-9 step inference
  • ~12GB download

Image Generation Client (Mac)

App	Z-Image Support	LoRA Support	Notes
AIbstraction Studio	Yes	Yes	Purpose-built for Life Design assets (coming soon)
Draw Things	Yes	Yes	Native Mac app, easy to use
MindCraft Studio	Yes	Yes	Based on mflux, Apple Silicon optimized
ComfyUI	Yes	Yes	Most flexible, requires setup
ZImageCLI	Yes	Yes	Command-line tool

AIbstraction Studio (Recommended)

A custom Mac application purpose-built for creating Life Design assets with this LoRA. Features optimized workflows for healthcare illustration generation.

Coming soon

Alternative: MindCraft Studio

MindCraft Studio is a professional AI image generation app for Mac built on mflux/MLX. Runs entirely local—no subscription, no cloud, full privacy.

1. Download MindCraft Studio from the website
2. Load Z-Image-Turbo model
3. Add this LoRA via the LoRA adapters panel
4. Set LoRA scale to 1.0
5. Use life_illustration as trigger word in your prompts

Features: LoRA support, ControlNet (Canny), image-to-image, Metal-accelerated on Apple Silicon.

Note: This LoRA is trained for Z-Image-Turbo. A FLUX Schnell version is planned for future release.

---

Usage

Prompt Format

Always include the trigger word at the start of your prompt:

life_illustration, [your description here]

Example Prompts

life_illustration, healthcare worker standing in modern clinic
life_illustration, patient sitting on examination table
life_illustration, person at desk with laptop computer
life_illustration, medical professional explaining to patient
life_illustration, office workers in meeting room

Recommended Settings

Parameter	Value
LoRA Scale	0.8 (balanced)
Inference Steps	9 (fast) or 16 (quality)
CFG Scale	0 (Z-Image-Turbo default)
Resolution	1024x1024 or 1024x1536

LoRA Scale Guide

Scale	Effect
0.6	Subtle style influence
0.8	Balanced
1.0	Design adherence (recommended)
1.2	Maximum style (may over-apply)

---

Training Details

Parameter	Value
Base Model	Z-Image-Turbo
Training Adapter	ostris/zimage_turbo_training_adapter_v2
Training Steps	3000
LoRA Rank	32
Training Images	369
Trigger Word	life_illustration

Trained using AI-Toolkit on RunPod (A40 GPU).

---

Train Your Own LoRA

Want to train your own Z-Image-Turbo LoRA? Here's how using RunPod.

1. Deploy RunPod Pod

1. Go to runpod.io/console/pods
2. Click + Deploy
3. Search for template: "AI Toolkit - ostris - ui - official"
4. Select GPU: A40 or better (24GB+ VRAM)
5. Container Disk: 100 GB
6. Click Deploy (~$0.43/hour for A40)

2. Upload Training Data

Once running, open Web Terminal and run:

mkdir -p /workspace/training_data

Upload your images (PNG) with matching caption files (.txt). Each caption should contain your trigger word.

3. Access AI-Toolkit GUI

Open: https://<pod-id>-8675.proxy.runpod.net/

Default password: password

4. Configure Training

Setting	Value
Model Architecture	Z-Image Turbo (w/Training Adapter)
Quantization - Transformer	NONE
Quantization - Text Encoder	NONE
Trigger Word	Your trigger word
LoRA Rank	32
Steps	3000
Timestep Type	Sigmoid
Save dtype	fp32

Critical: Use the GUI, not CLI configs. The is_zimage: true CLI flag is broken.

Critical: Disable quantization—it causes soft/fuzzy output.

5. Start Training

Click Start Training. Monitor samples in the output folder. Training takes ~60-90 minutes on A40.

6. Download & Stop Pod

1. Go to pod Storage tab
2. Download your .safetensors file from /workspace/output/
3. Stop the pod to save costs

Cost Estimate

• A40: ~$0.43/hour
• Training: ~60-90 min
• Total: ~$0.65-1.00

---

Files

File	Size	Description
life_illustration_zit_v2.safetensors	340 MB	Recommended - Full training run
samples/	-	Example outputs

---

Limitations

• Optimized specifically for Z-Image-Turbo (may work with other FLUX models but untested)
• Best results with simple, clear prompts
• Medical/healthcare contexts work best (matches training data)
• Complex multi-character scenes may require prompt engineering

---

License

This LoRA is released for personal and educational use.

The Life Illustration style is based on Laerdal Medical's Life Illustration style guide (private). This LoRA is an independent fan creation and is not officially affiliated with or endorsed by Laerdal Medical.

---

Links

• Figma Source: Life Illustration Library v2.2.0
• Base Model: Z-Image-Turbo on HuggingFace
• Training Framework: AI-Toolkit by ostris
• Laerdal Medical: laerdal.com

---

Acknowledgments

• Laerdal Medical for the Life Illustration style guide (private)
• ostris for AI-Toolkit and the Z-Image training adapter
• Tongyi-MAI for Z-Image-Turbo