The model transforms any objects, portraits, and complex shapes into volumetric 3D structures composed entirely of interwoven letters, mechanical typography, and symbols.
Operational nuances and upcoming update:
The base Z-Image encoder excels at creating 3D volume but frequently struggles with the geometry of specific letters individual characters may break or appear distorted at the edges.
An auxiliary "prior" model (trained on clean, accurate vector font skeletons) will be released soon to stabilize letter geometry. When used in tandem (chaining), the two LoRAs will yield perfect results: one maintains consistent typography, while the other twists it into complex 3D relief. If you already have a working LoRA for clean calligraphy or crisp fonts compatible with Z-Image, try combining them right now.
Prompting and the "golden formula":
For maximum depth and volume, it is best to use abstract typography, random letter sequences, or code symbols. The model performs best when the physics of text distribution is defined through light and shadow.
A [your facility] formed by intricate mechanical typography, code symbols, and calligraphic text. The density of words compresses tightly in deep shadows and spreads apart in bright highlights, perfectly mapping the 3D shape. Crisp typography art, dramatic studio lighting, dark moody aesthetic.
Recommended generation settings
Model Weight: 1.0
CLIP Weight: 1.0
Sampler: res_multistep
Scheduler: simple
Steps: 50
CFG Scale: 4.0