How to Use Reference Only ControlNet for Stunning Image Outputs

the reference control net is incredibly powerful what it does is it will take the reference image and that will impact the output image based on what you provide if you provide a picture of a person then it will generally produce an image with that person in it but the resemblance may not be exact so if you are going for really realistic consistency then you're not going to get that but say for example you put a Japanese mail like I've done so here you should get a Japanese mail when you get the output image and what I really like about using the reference model is you don't have to specify or go into much detail on the person that you want to generate because we've already done that with the image so by just providing this image here and by choosing the reference type and choosing the method as the reference only and the balanced option I can just put a photo of a man at a bar and I can hit generate on this and I can see what my output is going to be now I'm expecting to have an Asian man at a bar as my output and I'm also thinking because the person has a blue jacket on it might also reference that in the output so we may have a blue jacket and we should have an Asian person let's hope we get that with the output from this control net so you'll see from these outputs that we were pretty bang on here but you can see that it's definitely not exactly the same person as our reference image but we've got our Asian individual in a bar and we have the blue shirt showing in two of the four Images so this is quite fun it's quite good to play with and it's good to sort of trial and error and see how uh things are going with this uh generation one other thing that I want to show with this control net type is if we put say a item of clothing in so you'll see here I have uploaded an image of a red dress so if I was to change my prompt to say a photo of a Spanish female wearing a dress and I'm deliberately being vague in in my description of the dress I'm not saying what style it is ETC I'm just sort of saying it's a dress and we can hit generate here so you can see that we have the image outputs of our Spanish female wearing a dress whilst it isn't exactly the same dress it's got the color red on it and it's got some details in it so again experiment experimenting with this model can give you some awesome results just to go over some of the specific uh settings on this reference model because there are a few there is the control mode where you can choose between balance prompt and control this basically means if you choose prompt then it will put more waiting towards what the input prompt is rather than the image whereas if you put control it will put more waiting towards the reference image compared to um The Prompt and balance sort of gives equal waiting to the two and then the method there is the reference only method there is the reference Adan method and the reference adane attention these are all just different ways of coming to the AI image generation output you can experiment and try out the different ways of doing this but generally 90% of the time I find myself using the reference only method but of course we want to give you all the controls possible so there are the other methods that if you want to experiment with them this is just sort of a brief overview of the reference control net model feel free to um play with it and try it out for yourself it's pretty fun and it can give you some really awesome results with minimal effort required

Channel: Supermachine

Views: 206

Keywords:

Id: ahdi-h8aOgk

Length: 4min 24sec (264 seconds)

Published: Thu Jun 06 2024