The comparison chart looks very effective.
I haven't tried it yet, but after looking at the comparison chart, I can see that it significantly improves the quality of professional photography portraits (and doesn't affect other features). If there is a z-image base version without the specific trigger words (which might require longer training), this should be adopted by many people who like to perform LORA merging in models with "realism" labels.
The effectiveness of your first training has exceeded the quality of many models that have gained attention in the community.
thank you! The LoRA has be trained on the base model (not de-turbo) but the samples have been generated on Z-Image Turbo. The trigger word doesn't have any effect actually, about half of the samples do not include it and it seems to make no difference