I love SDXL. But the "photos" it creates are soooo.... DRAMATIC! Always with dark tones, soft lighting, and shallow depth of field. Do you ever just want photos that look like you took them on a phone: sharp, bright colors, and everything in focus?
Introducing SOAP, the LoRA that makes SDXL "photos" look like they were shot on a mobile phone (and not in portrait mode!) Sharper, brighter tones, and less bokeh. This LoRA will make the background of your images come to life.
No need to use any activation word, just apply this LoRA and watch the magic. I recommend using it at 25% for mild effect, and 50% for a strong effect.
Recommended settings on Auto1111: Euler, CFG 4-5, strength 0.25 - 0.5
A REQUEST: If this LoRA blows you away, like it has me, please remember to come back and post sample images and vote! Thanks for giving it a try.
Description
FAQ
Comments (15)
This is really cool. Please make a version for SD1.5 too.
Thank you. I'll look into that.
Hey, any chance you'd consider adding a 1.5 version?
Thanks for the interest! I'm doing a quick training to see if it improves images on 1.5.
I second this. Pretty happy with my prompts and character models at this point, but wrestling SD to produce anything but studio-level pictures is quite a challenge. Sometimes worse is actually so much better.
can you share the details of trainning ? I hope to apply in SD1.5
I carefully selected about 200 images for the dataset. Each image has something (person / animal / car/ etc) in the foreground, and the background is completely clear with no blur. Each one was captioned with a description of the scene. I then trained with batch size of 4, which seems to make SD focus more on the style of the images rather than the content.
I am currently increasing the size of the dataset to train the next version, and will also train a 1.5 version at the same time.
@vegakh thank you your response. I anticipate your amzing results and i will try it myself.
Hey hey, thank you for training this lora! :)
When I'm using it, the people in my images are getting more ugly. do you have an idea how that can be prevented? I'm already using the obvious negative prompts like 'ugly, unattractive' etc.
dude we need 1.5 version of this sooooo bad
Hey, this work is really good. Would you mind releasing how you trained this lora?
Doesn't work for me... guess that means I have 1.5 like these other people. Good to know I guess.
Massive imprvement in realism! ty!
You need to make one for flux too. Great job anyways