I will provide a platform for z image lora training

This app was built while fine-tuning Z-Image on a custom character. The character LoRA was trained on a small dataset of just 20 reference images, curated for consistent pose/style/lighting variety rather than sheer volume, then trained as a low-rank adapter on top of the base Z-Image model.

Rather than eyeballing single generations, this tool generates matched pairs the same prompt and seed run once through the base model and once with the character LoRA enabled side by side, across several seeds in a row. That side-by-side view makes it easy to spot whether the trained character's face, proportions, and style stay consistent or start drifting as seeds change.