ComfyUI AI: create artistic images without using artist names; SDXL Lightning Lora + IP adapter

hello and welcome to my new video This Time Again narrated by my AI voice of choice Charlotte I've been following the debate around image generating AIS and copyright or image rights since I started using stable diffusion last year as an author and photographer who benefits from copyright myself I can certainly understand people's objections for this reason I only use my own pictures which I either took myself with my camera or created myself in some other way and in the prompts for stable diffusion I never use artist names to add their style to my images in general I prefer a more realistic Style Both in my writing and in the images I create with the AI nevertheless the problem itself has not let me go that's why I'm going to show you my approach to using art in stable diffusion without copying in the artists or rather their images not only the sdxl models help but also the IP adapters and more recently the sdxl lightning Laura model the video is divided into three parts the first part is about creating the images that we will need in the second part to generate IP adapter embeds you can see how this works in detail in my last video in the third part we will use the embeds to create create artistic images ourselves you can already see a few of my results here part one as we all know all art movements have their own characteristics for example and the video will be about this art movement these terms are attributed to surrealism the unreal and the supernatural dreams and Visions wishes desires and suffering and my favorite states of intoxication and general knowledge which is a public good and may be used by everyone also includes the fact that surrealism is divided into two different schools absolute surrealism or abstract surrealism and paranoic critical surrealism the surrealists developed their own painting techniques and also their own approaches to the painting process itself there are five techniques in total frottage grattage collage de and drip painting and the two approaches are called automatism and hyperrealism but of course I don't want to give an art seminar here even if that might be of interest to some people the important thing here is that stable diffusion and the sdxl base model can process these it has been shown that the basic model benefits from the fact that it has had such broad training so it is very based as the saying goes we can take these and finally use the sdxl text in code clip G and clip L accordingly as far as I know G and L stand for group and list these list terms can therefore be used in clip L as you can see here the lightning Laura only runs if you set the CFG value to eight in the sampler at least for this version there is also a one-step twostep and a four-step version but eight simply delivers better quality and you have to set theuer to sgm uniform in my test the dim sampler proved to be the best however the speed upgrade for the sdxl models via sdxl lightning Laura proves to to be less helpful at this point in the workflow as you can see here in a direct comparison but in the third part of the video at the latest it becomes clear how drastically this can speed up the entire generation process it takes a while until we have all the necessary images together we need a few sample images for each technique to create the embeds in the second part it is interesting to note that basically only the sdxl base model adequately represents the respective painting techniques the other models are less broadly educated but sometimes deliver pretty cool interpretations there are of course other painting techniques that you can combine with the surrealistic ones it's not so rare for interesting results to emerge and it's also exciting to try out the control after generate functions in the sampler I don't normally use much more than randomize and fixed but sometimes it's worth running increment and decrement a few times part two creating embeds after selecting the images created in the first part you can now process them into embeds using the IP adapter I have done the work of creating these for the plus model and the normal model in order to compare them afterwards you can watch the whole process C in the uncut version of this video if you like in any case there were a few surprises the lightning Laura simply works brilliantly here instead of the required 50 55 steps for the CFG value in the exampler eight is now sufficient in any case you should also be careful here to set the first value in the encode IP adapter image node to true if you want to use an IP adapter Plus model and of course set it to falce if you are not using it with so many embeds I still made the mistake a few times part three three creating artistic images this is the really fun part with the lightning Laura you can Cobble together the embeds insanely quickly and in high resolution and create cool images as you can see the embeds from the normal IP adapter model manage to easily transfer their stylistic content to any image that is sent to the sampler via control net during my tests I found that the IP adapter plus embeds are excellent at painting the blank pages in the image mask but more complicated to use with the control net in any case with both models you can determine the foreground and background of the image via the start at an end at values in the apply IP adapter from encoded node as you can see here all in all there was a lot to smile about with all the things the AI wanted to read out of my surrealistic prompt entries in any case it is possible to have your pictures generated in a selected art style without copying an artist and their pictures this is even better for my own projects as they already have a certain way of their own anyway and I would find it a shame if someone looked at my work and could only recognize a copy in any case I hope you enjoy trying it out thanks for watching the video if it was interesting and or helpful for you I would appreciate a like and a subscribe last but not least have a nice day

Channel: Show, don't tell!

Views: 900

Keywords:

Id: UltnfnkTXa0

Length: 11min 21sec (681 seconds)

Published: Mon Feb 26 2024