Hello everyone! :)
I was one of the first people to generate dreambooth for 1.5 and then one of the first to do LyCORIS/LoCoN extraction but after that I was usually late to game (Flux, WAN, pretty much skipped SDXL).
When Flux 2 appeared I thought - I have my stuff set up so I can jump right in and do some lora trainings. Well, sadly I can't do that locally (yet). Fortunatelly 2 days later Z Image Turbo appeared and after my initial tests I was pretty confident - this might be a model that will stick around for good (especially with the BASE model coming soon and a possibility of finetunes on that, which is saomething we have been missing since SDXL).
Anyway, I was away for almost the whole weekend but I did manage to do the following:
There are five models trained on Z Image Turbo using AI Toolkit (check my article that comes after this one for the training template :P)
https://huggingface.co/malcolmrey/zimage/tree/main
I've played a bit with some parameteres and so far the default ones seem to be quite good, but there are of course some caveats, I will write about them in my training article in a bit.
Besides that, I've updated my model browser:
https://huggingface.co/spaces/malcolmrey/browser
It not only supports the ZImage models but I've updated links to all WAN models/samples :)
I've also uploaded simple Z Image workflow with lora added to it: https://huggingface.co/datasets/malcolmrey/workflows/tree/main/ZImage
So, what are the plans?
My custom AI Toolkit is ready to print Z Image LORAs as you can see :-)
I'm shifting my priority from WAN to Z Image. Hopefully those will work well enough on Z Image Base (if not, I will retrain :P)
Expect a lot of new Loras by the end of this week :-)
I suggest to follow me on huggingface as you will get the notifications right away.
I am also posting from time to time on my subreddit: https://reddit.com/r/malcolmrey
And I am happy from the support you give me on my coffee page :) https://buymeacoffee.com/malcolmrey
Cheers and see you soon! :)