WOW! NEW ControlNet feature DESTROYS competition!

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
Hello friends another game changing update to stable fusion and control net where you can have an input image and keep the same face style well whatever have a face put that person into different poses make the person laugh cry angry whatever let's have a look at reference only it is truly mind-blowing and again this is not a click bait let me show you a book just fell on my head I only have my shelf to blame so we are back into our trusty old stable diffusion and we are going to be playing with a control net this little down right here and make sure you have the latest version as of recording this video it's going to be [Music] 1.1.162 and it's updated all the time so as long as you got something later than that you're gonna be fine and you're gonna go into extensions here Crystal check for updates and whenever that's finished you're gonna see here to the right it's gonna say either latest or you're going to be able to update it you do that by pressing the apply and restart UI button here if that doesn't work for you you also need to update your automatic 1111 or vladder Fusion whatever you're using the easiest way is just go into your stable Fusion folder and type git pull and that will show you if you're already up to date or if you you need to update and that will be done automatically anyway let's get back to the new control net reference only which is a truly game changing control net preprocessor here so I've added this image here and let me show you quickly how this works you don't need to make any changes now apart from enabling control net and setting the preprocessor to reference only and adding an image and let's change this in depth instead to woman smiling gonna load my usual Styles as always these can be found in the video description free to download I'm using control Arrow up to wait this woman smiling and then we're gonna have 25 steps on oil array we're gonna do it four batches I'm just doing a regular old 5 12 by 5 12 to show you guys with the deliberate model and as you can see coming up on the screen now we have women that are similar to our input image we're not using any dream Booth we're not using loras or anything like that but these images of this woman or at least some of the women smiling are similar to the one here we have a purple hair even the shoulder here on three of them have this sort of similar um color gradient now the images aren't great however you can see that they're kind of messed up a little bit and that's actually something that's being worked on and if you see the official GitHub discussion here Elias villalo is a collaborator and author of this as they will be investigating blurring collapsing problems today but hopefully we'll be able to find a fix so whenever this video is released that might be fixed however if you change control mode from balance to either my prompt is more important or control net is more important you can get around this so let's use the same seed here now and let us run four new ones and you will see that these four new ones will actually work as intended as you can see we're getting the same sort of style that we had in our original image we have the purple hair we have the shoulder there coming in placed in two of them and now the image is not blurred grainy or messed up anymore and actually in all four of the images she's actually smiling now so I'd say that's a success for now it will not resemble the person 100 so if you're putting yourself in and you think oh it's going to be 100 me we're not quite there yet but you can combine this with Dreamboat Laura's whatever however I'm not sure we're gonna need to for much longer since this is getting much better each day let's not stop here with the power of multi-controllet we can actually do more with this so I have a pose here so this is an open pose so I have this is the hands going up there and this is the face here the the pink and purple and the blue is the neck there and then I'm gonna set here I'm gonna set the open pose now we don't need a preprocessor because it is this is already um open pose image now this is something that I got from post my art so I just uh took a model here loaded that exported to open pose and that will get you that image anyway this has been enabled we're not changing anything here and we are running this again so now we have a control Net Zero which is the previous one enabled we have the new controller one which is the open post also enabled and now I'm expecting to see four new images in this post all of them hopefully resembling this woman here or at least some parts of her so let's see now we have the four women here we have our original inputs so all of them have pink purplish hair in a similar tone we have the nebula gradient on the clothing apparent in well almost all of actually all of the images in various ways now our second image here didn't follow the pose but number one three and four did and did so fairly well and she is still smiling like we asked in the prompt here so I think this is a great success in the words of Borat now if we were to change her just simply saying something like woman crying and reducing everything else the same we have the same control net input zero we have the same post so we should hopefully get everything well not everything but most of the images similar but still have her crying now instead of smiling and as you can see here our second image still is the one without the arm gun going up on three and four have the arms now the clothes have changed a little bit so we don't have exactly the same image where she's crying but it's fairly similar and you can still see that it's her but it's just variations so this is an extremely powerful tool and I think this is the next step forward when it comes to control Nets and if you've been around for some time and know your way around control it you're going to be able to do amazing images with this and honestly if you're just starting out you're also going to be able to do amazing images with this because control net is so very very powerful let's do a quick test with another image let's take this old man here we're gonna disable the second control Nets we're changing this into old man Angry and we're running four images and let's see what we get out of this now our input image is a photo realistic Style and the Styles I have loaded are more digital oil paintings so we're not going to see a similar Style on the image as you can see here it's a little more photorealistic on the left ones and these are more hyper realistic paintings-esque it works how about if we take actually half the prompt for that image so let's just take this one and we're gonna lower that up in here make sure to remove the other styles I'm gonna have old rough man crying let's do crying this time and let's see if we can get him to cry and we should get a little more photorealistic style as we change the prompt here it's not dependent on the input image now it will reference our image obviously but the prompt is also quite important so as you can see now we have a more photorealistic style but the man is is crying so I hope you had fun watching is game changing update to control Nets I'm going to keep tabs on what's going on and since this was just released today I'm expecting it to get a little better with bug fixes and stuff the coming days I hope I'll see you in the next video as always have a good one see ya
Info
Channel: Sebastian Kamph
Views: 338,238
Rating: undefined out of 5
Keywords:
Id: tBwmbTwMxfQ
Channel Id: undefined
Length: 9min 7sec (547 seconds)
Published: Sat May 13 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.