Decided to try making some style loras for Z-image, in particular of traditional artists I enjoy. No trigger word required but i'd suggest starting your prompt with:
A painting of Enjoy.
Description
FAQ
Comments (11)
in your experience so far, how good is z image at understanding style compared to other models?
I think it's a bit too early to tell for sure, but so far it's done great. I've trained 10 or so loras and it been super receptive to new styles after about 1500-2000 steps. All the fails i've had so far have largely been my lack of knowledge or trying things that didn't make sense.
@imb101 that's promising! btw, are you using musubi or something else?
@olivereads38255 I’m using AI toolkit for Zimage and diffusion pipe for WAN and qwen image edit. Though I have to say AI toolkit is really great in terms of usability
i've been as part of my tests trying to recreate many cool images and art styles. it's often difficult at first but once you find right prompting and words to use it comes close. i'll start with AI analyzing image and giving me prompt styles and so on for that image, but often it doesn't work right away and requires much manual tweaking. I am sure Lora's of specific styles will make it much easier to achieve the style you want however.
@imb101 i see, thank you for the tip imb :)
@zengrath idk if anyone else has my problem, but i haven't been able to get a decent style lora to train properly since sd 1.5 lol
@olivereads38255 I found @ostris vid super helpful for zimage styles :
@imb101 omg thank you i needed this!