Breaking Image Making into Stages using Depthmaps in ComfyUI

hi this is a follow-up video really taking uh my last video on using depth maps just a little bit further so this is a setup where I'm separating out the making the background putting the figures in and then comping the whole lot together and applying the style so what it's doing is splitting up the job of making an image so we have the first module here making our background Zoom it so you can see that and there's I'm not attempting to make these images very pretty they've just got to have the right stuff in the right place and then this group here puts the people in and then this module here refines and does The Styling and and final rendering so if we go back to this and we'll have a better look at the depth maps involved so this is the background depth map which is uh just made from a a scene I'm not quite sure where Italy I think and it's just a simple depth map nothing nothing fancy at all and this goes with a prompt that is entirely about the background and there's one Laura here to um strengthen the architecture and here we we don't have very much I do have face in which uh tends to uh stop Generations being too symmetrical is not really needed in this case because it's being controlled by the depth map so here is another depth map this one has a mask so it's a PNG with a a transparency which is easy to make in Photoshop now the transparency is is uh the wrong way around for compfy so the um the the mask goes through an invert mask and into the encode for in painting So this essentially is an in painting module here and we are in painting the figures there is a reason for the figur is um being rendered in this way if you see I've put um I've put illustration in there because I don't want the lighting too decided on them uh the uh final generation o in the next section will will take care of that but you can pretty much see how this all works the uh control net is set very strong because we want those figures to be really quite crisp so the next layer or the next module whatever we want to call it grouping we have another depth map which is essentially the first two dep dep Maps combined together the only thing you have to really watch is to get a good contrast between the figures and the background otherwise they'll it'll decide it'll decide the bit of this is a bit of leg or something you how it is or it'll put a whole extra person in to the left there so here we have uh on on here our prompt was entirely about the people there's nothing said about um the background at all but here we have both prompts from the first model module about the background and also the prompts about people as well and then we have two lauras which to do with styling so the land EO one makes very nice um people and I'm aiming for sort of 50s 50s um romance illustration really but you can obviously set this section to be any style you particularly want the reason for these images being very plain vanilla is that they those sorts of images uh translate better than than an image that is is too decided so the settings here again the again the control net is set very strong it starts at zero ends at six the luras are not set terribly strong but they're quite they're very strong luras so Doo is set at 0.6 because we want to keep quite a lot of the original image or certainly the layout of it so if you go you go up to seven it'll start moving the people around which I don't mind in some circumstances that's fine but and I would almost certainly do a few generations with the doo higher just to see how what if I got lucky you know got a few glitches this is the final um result we have a few glitches the um background could be a little bit darker which would push this away so we got some unknown piece of trash there someone's left a chocolate a CLA on the on the on the roof there so uh the background could go uh a little bit darker relative to the Fig and then uh it would push it away but the figur are quite nice again you could do as many generations as you want so so I would soon find a generation that that didn't have any of those fault and then I I would just F to shot the two together so I think that pretty much gos it there's nothing unusual in the setup at all this is straight image to image this is straight in painting and this is straight just generation okay so I hope that's of Interest I'll post the work flow in the blur so if you want to try it out you can okay thank you very much

Channel: Rob Adams

Views: 1,916

Keywords:

Id: EUc5BJjkGFk

Length: 5min 12sec (312 seconds)

Published: Wed Jan 24 2024