okay I'm going to show you how to make an image of a consistent character like this using stable diffusion on automatic 111 so I'm going to explain how this workflow work and then after that I will show you how to set it up step by step so um you would need to use an sdxl model I found that not all the model work so uh the model I'm using is protovision XL so I recommend you to use the same model as I do uh to reproduce the result after that you can switch to other models and experiment with it so this workflow use two control Nets the first control net is control neet candy it fix the uh overall composition to generate a 3X3 grid of a consistent character image and then the second control net is the IP adapter so it is used to copy the face of an image that you provide and then put it into the final image and then you also need the ad teller extension so the reason you need this is because you are generating an IM image with 1,24 by 1,24 but with nine different faces so the faces are simply too small to be create uh accurately so you need to have this extension to do automatic in painting uh what it does is it will take out one of the face and then in paint it the high resolution and then resize it back down and put it back to the final image and then do this one by one nine times so with that said now I'm going to show you how to set it up on a clean uh automatic 111 so I will put in the prom and the negative prom and then the stable diffusion checkpoint need to select uh as this protovision XL model and then after that set the width and height to 1,24 and 1,24 so now we go down to the control NE session you should see three control next uh in your extension so if you don't see it you will only see one you need to go to setting and then control net here and then you will see an option called multi control net and control net unit number save to free and then click apply and then restart your guy so now I have it here so I'm going to uh put my image to the control n c canvas and then enabled it and then I'm going to cck cany so the processor should be cany and then I'm going to use the diffuser XL candy mid so um you want to tune in down a little bit so let's try 0.4 and also the ending control step is the number of steps um in fraction that uh this control net with n for example if I only use sampling step of 20 and then I say the ending control Step at 0.5 that means I only apply this control net from step one to step 10 and then in Step 11 to 20 so there would be no candy control net being applied the reason you want to do that is that uh sometime if you apply it uh too much uh it restrict the image generation so it would just follow the shape of this image so you want to free free the image generation from Cy control net at one point so that the image generation can just simply follow the prom and the IP adapter that we just uh set up so when I set control net I always want to review it um so you can click this uh run pre-processor button so you can see the uh pre-processing of this image you can so now it looks good because it accurately extract the outline of this control image so next um I'm going to set up the IP adapter so for this you would need to use an image that you like uh it can be generated or it can be something that um just like a photo of a face you want to copy so I put it in the control NE unit one and then enable it and then let's see I click IP [Music] adapter and then IP adapter pH ID plus and then the model is IP adapter face ID plus V2 sdxl and then for this one um maybe I can start with 0.8 so go back to Candy one more time um just check the setting and then okay okay I found that I didn't click the Pixel Perfect so this option is to make sure that the preprocess image is also the same size as the one that you specify here that 24 by 1,24 um so what I usually like to do is I don't enable the Adil first and then I would generate an image to see how it look because this is quicker so now you can see this image is generated um don't worry about the phase for now uh so you will be able to fix it with the a detailer so what we want to look at is the overall composition whether we get uh n gs uh whether we are getting something that resemble the image in the IP adapter that you provide so you like this one then we will fix the seat and then after that that uh we would go to the ad teror session and then enable ad teror and then generate again so Adil will detect the faces automatically on the image and then it would do in painting one by one and the reason why it work is that it would imp pain at a higher resolution and then scare it back down to the lower to the smaller size and then put it back uh to the image so now we see that um the image is much nicer because we fix the faces with a detailer yeah so uh that's how it worked so one thing I want to show you is that uh although we are using IP adapter to copy the face of this woman but um at the same time the image is also controlled by the prom so for example um let's say I want to uh change her hair color let's say uh it's a bone and then I can click generate and then we can see what happened okay we change her hair color yeah so um the image is influenced by both the IP adapter image and also the prom so this is the beauty of Ip adapter so it simply give you one more option to control the image but not taking away the option of controlling the image using text and then you can also change a short color U maybe I say uh black blouse and then this is something that uh you would need to experiment with uh especially if you're are putting two different colors um on the image a lot of time stable diffusion might confuse the color like where is where for example um in this lower corner um it may not seem to be gener everything correctly but um this is something you can simply change the seat and we generate and then try to get one that looks good okay so uh that's everything I want to show you today uh I'm going to put the um text description of this workflow in the link together with where to download this control EMA Imes um including uh this gr of image and also the IP adapter image I used uh as well as the link to download all the models uh and all the extension you need for this workflow all right thank you for watching
