"Hi everyone, this is my first attempt at testing Z-Image-base.

I trained this on Rank 64 for 19k steps using 820 images (90% NSFW). However, the LoRA itself works for SFW content as well, as long as you don't use direct/explicit prompts.

Since this is my first LoRA for this model, it still requires some refinement in details and understanding, so the results might not be perfect just yet.

I’m looking for feedback!
If there is a demand for these types of LoRAs, please let me know what would be better to focus on:

• Adding specific characters?

• Working with styles?

• Including special concepts?

• Or should I focus on this single LoRA to improve its quality and understanding?

Generation Settings:
All sample images were generated on Z-Image-base using:

• Steps: 40

• CFG: 4

• Sampler: euler_ancestral

• Scheduler: beta57

Note: I can't guarantee these are the absolute best settings, as I haven't tried to find the most optimal configuration myself yet."

P.S. I know I posted this in the wrong category (Z-Image-Turbo), but the Z-Image-base category wasn't available on Civitai at the time of upload. Z-Image-Turbo works too."