Duchesse Style Z-Image
v2 - I don't normally release updates so quickly but I liked this version and wanted to post images with it. I tried changing some of the training settings and doubling the number of training images and it seems to work pretty well. I get fewer extra limbs and a stronger style with this one.
I trained this on Z-Image base but all the sample images are done with Z-Image Turbo
Duchesse Style Flux
I don't know if anyone but me wanted this but I had the dataset and a lot of blue buzz so here it is!
Duchesse Style v1.5
Winning auctions is expensive and I’m buzz poor, so to keep some of the Duchesse checkpoint's style available for on-site generation I’ve made this LoRA.
This version of the LoRA is more faithful to the Duchesse checkpoint's aesthetic, while also being more versatile across a wide range of checkpoints.
The samples include images from seven different checkpoints so you can see how Duchesse Style v1.5 interacts with each style and what kinds of results you can expect.
Duchesse Style v1
A LoRA I made to capture the look and feel of my Duchesse checkpoint—a blend of painterly elegance and soft character rendering. The style was originally inspired by DucHaiten-Pony v6, and this LoRA captures many of its distinctive visual traits for Illustrious checkpoints.
I tested it with a few different checkpoints—PerfectDeliberate was my personal favorite, mostly because that combo came closest to the aesthetic of the Duchesse checkpoint. I usually kept the weight between 0.8 and 1.0, and that gave me great results. I've also included examples from iLustMix, Illustrij, One obsession. I'll be trying it out with other checkpoints as well and I'll post the images here.
Feel free to try it out with those or any other Illustrious based checkpoints—I’d love to see what you come up with!
Link to the Duchesse checkpoint: https://civarchive.com/models/1798332/duchesse
Description
First attempt at a Flux LoRA
FAQ
Comments (8)
If you still have the dataset ready, i would love to see it ran against Z-Image-turbo 😍
Yeah, I still have it all. I tried to make a Z-Image lora but I was getting a lot of body horror. I'll give it another try to tonight to see I can get something better
@Chumple https://github.com/ostris/ai-toolkit worked for me, i used the Z-image-de-turbo base model for ZIT = it's an de-distilled version with CFG that works wonder to make ZIT LoRA (i didn't try the LoRA made with it on Z-image-base for now, my setup is refused this version ^^;)
@n_Arno Thanks for the link! I'll give that a try, on my first attempt I just used the civit lora creatior
@n_Arno I had some trouble with ai-toolkit so I tried it with OneTrainer. I think I can make the lora a bit better but I published my first attempt.
@Chumple On my way to try it as soon as i have finished my last picture of the day of Civchan :D
Finally took the time to test it (well more than me taking the time, it was my GPU struggling for the better part of the afternoon): https://civitai.com/posts/26631126
@n_Arno Thanks for trying it out and making so many great images!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.