StarLumen is a material-focused LoRA trained on Z-Image, designed to enhance surface depth, fabric behavior, and controlled emissive glow in character outfits.
Instead of forcing a new art style, this LoRA improves how materials react to light, especially in environments where Z-Image outputs tend to look sterile or overly plastic.
What this LoRA does well
- Improves fabric realism and surface layering on sci-fi and futuristic suits
- Stabilizes internal glow elements (white-blue emissive structures) under varied lighting
- Adds material softness and depth, especially noticeable in anime or hybrid renders
- Preserves Z-Image´s core structure and proportions while enhancing texture perception
What this LoRA is not
- Not a face enhancer or portrait realism LoRA
- Not a world-building or cinematic environment model
- Not a one-token “magic style” LoRA (specific prompting is recommended)
Usage Concept
StarLumen works best when the outfit or suit is explicitly described in the prompt.
The LoRA responds strongly to material language such as fabric, internal glow, emissive panels, layered suit construction.
Example prompt:
"A hyper-realistic photo of a woman in the Starlumen_Suit standing in a dark, rain-slicked cyberpunk alley. The internal white-blue glow of the suit reflects vividly on the wet pavement and rain puddles. High contrast between the dark environment and the glowing suit elements. 8k, cinematic lighting, sharp details."
Example Two: "A photo of a woman standing in a dimly lit sci-fi corridor at night. She wears the Starlumen_Suit with a consistent internal white-blue glow. The glow softly illuminates the surrounding surfaces."
Description
Base Model: Z-Image
Model Type: LoRA
Training Focus: Material behavior, fabric depth, emissive light stability
Dataset: Original, curated sci-fi suit and material-focused character renders
Training Details
- Pictures: 37
- Repeats: 3
- Epochs: 16
- Total Steps / Iterations: 1776
- Training Objective: Improve material perception (fabric, synthetic surfaces, internal glow) while maintaining compatibility with Z-Image’s default lighting and anatomy.
🔧 Recommended Parameters
LoRA Strength (Weight):
- 1.0 → Commercial / semi-realistic renders Best for subtle material enhancement Preserves Z-Image’s native look
- 1.3 – 1.5 → Hybrid anime / stylized realism Balanced softness and glow visibility Recommended default range
- 1.8 – 2.0 → Strong stylization / anime override Clear anime influence Maximum material softening and glow emphasis
CLIP Skip:
- CLIP Skip: 1 (recommended) Maintains prompt fidelity and material control
- CLIP Skip 2 can be tested for stronger stylization, but may reduce material precision.
⚠️ Usage Notes
- Explicitly describe the suit or outfit to activate the LoRA’s strengths
- Works best on sci-fi, futuristic, or synthetic clothing
- Backgrounds are intentionally not modified; this LoRA focuses on subject material quality







