Hello everyone!
You can check my Z Image loras at https://huggingface.co/malcolmrey/zimage/tree/main (currently 5 of them).
There is really no need for a trigger, the simple "woman" will suffice.
I have attached my current AI Toolkit config (zimage-config.yaml)
When you click "New Job", just copy paste the config then click "Show simple" and fill the two things:
training name (name of your subject)
target dataset (folder with images)
And then just "Create Job".
Again, I suggest around 20-25 images. You can include mostly headshots, but you can change some of them to body shots if you are interested in the costume/body physique.
Those 2500 steps work fine with 20-25 images. If you want to include LESS or MORE images - I suggest to experiment with steps, because there IS a correlation between STEPS and QUANTITY of images in your dataset.
For a character/person you do not need to include captions, just images.
My models were trained using 512x512 and as you can see the loras can generate higher resolution images with ease.
Z Image is new, we do not have yet the BASE model, I will be experimenting with different parameters, etc. So far, the results from the attached template are quite good, but if we find something new and cool - I will definitely update you with that finding :-)
Cheers and have fun!