ComfyUI AI: create artistic backgrounds with Fibonacci, Euler, Heisenberg and Brown's formulae

hello and welcome to my new video again spoken by Charlotte the friendly AI voice from the neighborhood I experimented a bit more with the IP adapters and the idea of using them and the sdxl lightning models to create artistic images this time however the IP adapter embeds created in the workflow are not intended to add artistic style to the latent image noise but primarily to create pleasing and controllable image back backg grounds but I'm getting ahead of myself let's start with the first part of the video this is about creating abstract backgrounds and whether it's about sense relationships number relationships force and object relationships or image relationships there are formul for all of them that are so well known that even stable diffusion can do something with them if you type the Fibonacci series O's number Heisenberg's uncertainty principle and even brownie in motion into the prompt you'll at least get some great pictures as you can see I've opted for images of patterns and spirals which are then processed into embeddings if you want to use the new Juggernaut lightning model you should set the steps in the K sampler between four and six and the CFG value between one or two and above all the sampler must be set to dpmp sde anything else will lead to unsatisfactory results the Schuler should be set to normal or remain so it was interesting during testing that the models were even able to do something with the direct input of the formula when testing without formulas the results were not as good as the ones you can see here now that we have created enough spirals patterns and wildly mixed up molecules we can move on to the second part here it has been shown that two output images are not quite enough so this time there are four load image nodes with the prepare image for clip Vision nodes required for this workflow which ensure the square shape of the images for the sdxl models and enable additional interpolation this helps the apply IP adapter node to process the image information again when saving the embeddings do not forget to set the first value in the encode IP adapter image node to true if you select an sdxl plus IP adapter model once you are satisfied with the generated images it is time to move the embed files from the output folder to the input folder let let's quickly set up the third part of the workflow again we need the apply IP adapter from encoded node and the load IP adapter embeds node as well as the load IP adapter model node and now it gets exciting in the apply IP adapter from encoded node we can now influence the background of the image via the weight value and above all the starcore at value the IP adapter Plus models still need some fine tuning but my initial results look promising the difficulty lies in not transferring the style of the embeddings to the main motif of the image whose background we want to adjust as always the control Nets help us with this however Midas depth map and canny together with the IP adapter plus model have brought my computer to its knees the new Juggernaut lightning model has been able to counteract this somewhat as it only requires four steps in the case sampler and when faced with the decision of which control net to use my choice was always the depth map all in all it was a lot of fun to work with the IP adapters again especially because of the speed of the new lightning models and Fibonacci oer Heisenberg and brown have once again proven that you can count on them thanks for watching the video if it was helpful and or interesting for you I app appreciate a like and a subscribe last but not least have a nice day

Channel: Show, don't tell!

Views: 2,182

Keywords:

Id: b_tAL06G7mo

Length: 12min 39sec (759 seconds)

Published: Thu Mar 07 2024