!!! Please do not use this LORA for commercial purposes !!!
The style is more like realistic, and can be effected by some style LORA.
You can use the prompt attached by pictures.
With the Z-image published, this's a very good base model to generates image for scene composition and context,especially for those common computers
My current workflow is to use Z-image generate the basic character compositions and environments, and then use Illustrious with LORA to partial redraws. The image generated in this way will have more tension and impact.
Have Fun!!!
Description
FAQ
Comments (9)
The file that was uploaded for this is the same as Aesthetic Fashion Collection - Autumn. Can you reupload?
Fixed, Thank you for correct me.
Not a prob.
This entire set simply does not work in most checkpoints i tried.looking at example images, seems like i am not the only one.
Which checkpoints are you using? I train them with the illustriousXL20_v20 , so the adaptability should be good. Maybe you can check the VAE or other settings?
@affa1988 pretty much tried with all popular 2d checkpoints.Wai,Prefect,madly,hassaku. Looking at example images people posted also confirms. pretty much in everything people posted it turns into a white bra with barely any pearls and the flower one is straight up just a breastless bra in most example pictures people posted except for one where it still doesnt look like a flower. Success rate of this is extremely low. i tried around 30 clothing lora yesterday including 8 of your lora on 4 different checkpoints and this series just simply didnt generate accurate images even once for me.
I dont know what settings you are using and i dont know if you can fix it going forward but please include a few 2d picture examples in the future i can copy and try to confirm it is working in the future or add in the description that these only work in 3d checkpoints. thanks for the work though.o7
@Wizzi I tried "Wai" and "hassaku". And the LORA perform quite normal. I find if you delete the prompt"cupless bra" and other may be associated of NSFW , the image generation will be more accurate. I guess all these most popular models contain too much NSFW elements. You can check my example image from the Gallery below.
@affa1988 in the example images you used no character LORA, no style LORA and almost no prompts. There are a lot of LORAs on the site that can only generate upper body shots from front with no other style or character LORA. And if that is your aim i have nothing to say. But to me this looks like there is something to be improved somewhere in the workflow so it can work with various poses,prompts and LORAs as well. And since your ideas are really interesting i would love to see best versions of these LORAs possible.
But yes you are probably correct that anything that can affect the bra shape in any way is overwhelming the LORAs understanding of what a flower or pearl is and turning it into a print bikini basically.
@Wizzi My suggestion is to use no more than three LoRAs at the same time. Otherwise, they may interfere each other, make image loss or confusion of certain details or concepts. This is especially noticeable with LoRAs focused on art styles or faces, as their effects can be particularly pronounced.
I usually use ControlNet(depth or open pose) to generate a low-resolution draft, and then gradually refine the details and specific elements through methods like HD upscaling, inpainting, and face restoration.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.