Since the release of Z Image Turbo and Z Image Base, I’ve conducted extensive LoRA training and testing based on these two models.
Some LoRAs built on ZIB and ZIT are already available on the site for free download, and more will continue to be released in the future.
At this point, LoRAs based on these models have reached a relatively stable level of output quality. They are now generally reliable in terms of character feature reproduction and consistency.
Some clients have already tried custom orders, and the feedback so far has been very positive.
Z Image Base and Z Image Turbo LoRA commissions are now officially open.
A brief comparison of key features between models and FLUX1 is shown in the table below:
| Feature | Z Image Base | Z Image Turbo | FLUX |
|---|---|---|---|
| Image Quality | High | Medium | Medium |
| Realism | High | High | Medium |
| Diversity | High | Low | High |
| Prompt Adherence | Very High | High | Medium |
| Output Stability | Medium | High | Medium |
| Target Similarity | High | Medium | High |
| Generation Speed | Medium | Fast | Medium |
| LoRA Stability | Medium | Medium | High |
| LoRA Success Rate | Medium | Medium | High |
Notes:
- LoRAs trained on ZIB can generally be used with ZIT. However, the strength typically needs to be increased to around 1.2–1.5, and some quality loss may occur compared to using them with ZIB.
- If you commission LoRAs for the same target across multiple base models, each additional base model version is available for an extra $5.
For more details about model orders, please visit:
https://sololo.xyz/commission
