Version 2 - What's New
Improved anatomy and muscle structure
Variety of body types: From slim to curvy body shapes
Natural coloring and balanced contrast
Wide age range
Enhanced backgrounds with better perspective and depth
More natural face and body contours, reduced artificial look with improved skin texture
Various styles: Analog, amateur, professional, and more
Strengthened ethnic diversity
Wide range NSFW capabilities
Better prompt understanding and a broader range of poses
Better facial expressions
Recommended Generation Parameters:
Sampling method: LCM / Exponential
Sampling steps: 8-12
Resolution: 896x1152 / 832x1216 / 1344x768 and landscape
CFG: 1There is no need to add DMD LORA separately to the model; just load the checkpoint and run it!
I would appreciate your Buzz donations. I am developing this model solely with the Buzz points I’ve earned through the site
Warning!
This model is intended for ethical and responsible use only. Any attempt to use the model to create harmful, illegal, or unethical content will result in immediate disqualification and may lead to legal consequences. By using this model, you agree to comply with all relevant laws and ethical standards. Please use responsibly.
Description
FAQ
Comments (17)
The hands and feet are impeccable, were these done without loras?
And do you need to use the DMD2 lora with this? I'm new to dmd
I don't think you need the DMD2 Lora. DMD is just a technique afaik. Just use LCM and 8 steps or more and generate the images.
No, the DMD2 LoRA was already merged with this checkpoint.
@focalbluebell Gotcha, much appreciated
@dev_sysneon708 I wouldn't say it's all good with hands here. Three-fingeredness is not uncommon.
Seems to wig out when loras are introduced, is just a DMD2 thing?
loras should generally work with DMD2
Do you plan to make an improved version of the model? I think the model is very good and different from others—it produces unique but great results in the images. I like it, excellent work!
Fantastic model! Is there a way to get negative prompts to work with this? using a CFG scale above 1 doesn't seem to work
install neg pip
So far it is the best SDXL model I used. Keep updating! Thank you for your work!
In general, everything is fine - the details, lighting, texture of the skin, face, hair. But in about seven to eight generations out of ten, the number of fingers is like that of a bird. The fingers themselves look great. If you could solve the problem with the number of fingers, this would be the best realistic SDXL model.
After this model of yours, everyone else can give up, no one will create a better one, Thank you so much for such a GIFT!
Couldn't get it to work. Everything came out overcooked - no matter my settings.
Try a lower CFG scale per the instructions. Anything over 1 will break it. It's some kind of quirk of DMD2 stuff I don't understand.
Seems fine at higher sampling steps, but it looks great even at low counts.
dvnnt - Below 1 on the CFG? Ok - I'll try... and see how it goes. Thanks for the tip.
We need update. Pleeeese.