ComfyUI Any Style combination workflow #comfyui #controlnet #ipadapter #workflow

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
hi following the requests of some viewers today we will see how it is possible to combine quite perfectly completely different styles in one picture this workflow looks a bit complicated but if you've seen my previous videos on comfy you'll quickly understand what's going on here and this is also the place to thank you all the viewers who comment and support it's really fun to know that you are learning and using these videos and I hope you will continue to enjoy them in the future as usual we will start at the beginning and of course the relevant links will be in the description of the video here we have a node that we haven't talked about yet called The sdxl Prompt styler and it contains lots and lots of prompts for different and varied design Styles which you can choose with the click of a button if we look in the costume nodes folder then you can see the files that contain these Styles and of course you can create a preset for your prompt just pay attention to the way of writing write any prompt you want and click on Save back to comfy click on refresh and now you can see that the preset we prepared is ready for use we will use this node to select the desired Styles in our case I will choose futuristic biomechanical cyberpunk here I also write the basis of our prompt a a x closeup y y y in background high quality sharp and for those who have haven't seen the previous video the trip x and y y y are a type of codes that we replace with the help of the find and replace node this node replaces the tri x with the word dog and this node replaces the y y y in the word Suburban Street all of this goes to the positive prompt for which we Define the text as input and here I connected a show text node so that we can see what ends up going into the positive prompt as you can see the tri XX was replaced by a dog and the y y was replaced by Suburban Street all the additions comes from the prompt Styler node and this is the result we got I work with Juggernaut sdxl lightning fix seed and latent image 1216 on 832 I quite like the result but I want to have a background with a more Suburban Atmosphere by the way please note that here I am working with the preview Bridge node which allows me to continue working on this image we will soon understand why let's connect the IP adapter and to it I upload a picture with the atmosphere I want to get and press Q the IP adapter is adapted to sdxl and I work with style transfer weight as you can see the result the street is very similar to the look I wanted but I liked the previous dog we got much more to avoid the change we receive in the dog we can work in a slightly different way we will disconnect the IP adapter from the original image so that we'll get the dog that I liked and here we create a new image with the same prompt in combination with the IP adapter because I want exactly the same composition I also work with control net with depth model based on the first image here too we got a completely different dog but if you notice the composition is the same which will help us a lot to put things together the composite is quite simple we will use remove background node which also knows how to create a mask for us this is why I chose to work with the preview Bridge so we will simply connect the image and in most cases we will get a pretty good separation from the background as you can see to composite the two parts we can use latent composite mask node which is just like the image composite masked only in latent space simply connect to it the latent from the two images we created and the mask from remove background let's already enlarge the image and improve the quality of the details for this I passed the ladent through NN latent UPS scale adapted to sdxl with an increase of 1.5 and to another Cas sampler that is connected directly to the model without the IP adapter and another thing that will greatly affect the final result the Deno I will put it at about 0.45 and should change it according to the result and the combination you want to get and here is the result we were able to get a dog very similar to the first picture with an atmosphere very similar similar to the reflection picture we uploaded to the IP adapter but sometimes I have a desire to get a certain style and I don't have a suitable reference image so instead of using an IP adapter we will use the prompt Styler that we met at the beginning of the video and create the second image with it as you can see here I simply copied the same nodes and connected them to the same words dog and Suburban Street in Priest I chose art style psychedelic I put this text in into a new clip and code and I connect it to control net now we should get the same composition as in the picture the original only with a psychedelic style so that you can better control the final nature of the combined image I connect the two positive prompts to the conditioning average which allows you to create a balance between the Styles but remember that it also depends very much on the Deno you choose in K sampler and this is the result of the two Styles combined together look how beautiful the colored reflection on the dog which is obtained because of the mixing of the two Styles in the last Cas sampler some nice things you can play with let's say you want a psychedelic dog and a futuristic background you can simply add an invert mask between the remove background and the latent composite you can of course change the keywords of the prompt in short this workspace allows you in a relatively simple way to play and switch between different styles and their combinations in the image so I hope you learned and we will meet in the next lessons you are of course welcome to subscribe to the channel ask questions and like if you liked it and most importantly have fun
Info
Channel: PixelEasel
Views: 2,819
Rating: undefined out of 5
Keywords:
Id: lwendQxMMyc
Channel Id: undefined
Length: 6min 18sec (378 seconds)
Published: Thu Apr 04 2024
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.