Introducing the Sebastian Krüger Art Style for SDXL 1.0
Experience the iconic artistry of Sebastian Krüger, a renowned German artist celebrated for his captivating caricatures and hyperrealistic portraits. Krüger's distinctive approach involves skillfully exaggerating details in his depictions of celebrities, capturing their essence while infusing a touch of surrealism. By manipulating proportions and expressions with finesse, he crafts larger-than-life interpretations that both celebrate and satirize his subjects, often drawn from the vibrant realm of pop culture.
The artistic journey, however, isn't without its challenges.Diffusers obviously can't understand a persons personality and to create a caricature becomes a task of trial and error combined with prompting and seed surfing.
This version 1.0, represents my initial exploration into the realm of training a LORA. While it might not be flawless, I hope it offers an enjoyable outlet for personal amusement.
In my usage, I've found a weight of approximately 1 to be effective, although you may need to fine-tune it based on prompt complexity. A clip skip value of 1 was applied to the images shared on this page, and mostly CFG 7 normallly gave me decent results.
The main trigger word is kru3ger_style. As I continue to explore the capabilities of this Lora, I invite you to review the prompts I've used for the images I've posted. Additionally, your contributions and prompts shared here will provide valuable insights into effective interactions.
Personally in my testing i get good results for portraits with 3:4 ratio (1024*1365). I will post 3 example images with the same prompt & seed in this gallery to demonstrate the difference it makes in style when used with that aspect ratio,(1) with and (2) without trigger word, (3) one vanilla one
It's important to note that my creation of this Lora was fueled purely by admiration, devoid of any commercial intent. For those captivated by this unique art style, I encourage you to explore Sebastian Krüger's official channels and consider acquiring pieces from this remarkable artist.
If you're brimming with ideas for future projects, I welcome your input! Dive into the enchanting realm of this remarkable technology, experiment with its potential, and remember to share your outstanding creations for all to enjoy.
Your feedback is invaluable, so if you find this tool beneficial, don't hesitate to leave a review and showcase your captivating works of art!
Description
1.0
FAQ
Comments (7)
Niiice! Would you consider a SD 1.5 version?
I haven't planned that since i find sdxl superior. At least when control nets etc will arrive.... But maybe if its not too much work i could take a look into it when there are a couple people asking for it and i find the time. Why do you want a 1.5 version?
Not sure their reasoning, but my GPU can't handle SDXL (6GB, sadly), so I'm stuck on previous versions.
I have a 1.5 version ..... but i havent uploaded it anywhere
@gonzo I'm fully invested in 1.5 still, with a full setup of personal Loras, plus I haven't seen a reason to switch, so I'm really happy whenever I see cool stuff for 1.5. BTW would still love to see the Sokol Lora too!
@macro it would be great if you consider uploading it since some people would like to play around with it. Besides that i would like to test it to see how it behaves in relation to mine so i can improve my understanding of training.
@Eluna i was already suspecting that would be the case for some people with weaker machines. Since i have a fairly strong machine (4090+80GB Ram)i havent researched much but as far as i know you could still use SDXL on 6GB Vram if you are willing to use Comfy. On reddit i saw people doing that so maybe you check that out.
@_1_ i have a huge library of personal Loras on 1.5 myself and the amount of control that is achievable through extensions is still unbeaten, so i totally get you. On the other hand the SDXL base seems to be way superior than the 1.5 base in understanding, quality and flexibility. Imho not only a little but realy a lot, which makes it logical that over time SDXL will surpass 1.5 and the community will switch and adapt. Maybe you give it a littly try besides your normal usage so when the point of final transition comes you are not so left behind. The Sokol Lora is realy not in a shape where i would upload it here on a CivitAi. It had almost no captioning for the dataset (in relation this Lora took me 6 hours of manual describing the images while Sokol was done automatically) and is extremely overtrained (on this i ran hours of batch processing and comparing epochs to fiddle out style vs flexibility). If you still want it, i could send it to you in private but you will be dissapointed in the end.
TRES TRES BIEN ! bravo