I've been playing around with <a target="_blank" rel="ugc" href="https://civarchive.com/models/897413?modelVersionId=1147588">Big Love </a>to prepare for the fattening season. Version #1, the others don't work well for me at all. I share my workflow for running this model. Also runs well with pony/illustrious models. Here's the flowage:You write words in the big green box to tell it what you hope to see. There's a mix of LoRA stuff, two stacks for sliders, and some Manager loaders. If you enable Controlnet you can load a depth map or pose skeleton. If not, don't feel too bad about it. Three ksamplers, ah crap I forgot to put the steps back. Oh well. It's set to 34 steps overall, just add 10 or so steps to start_at and end_at widgets (except first ksamplers start_at, obviously) for bigger love. I turned it down a bit for cleaning up for export, as detailers take some time. Anyway, after the generation it goes into the face detailer, wherein you can detail a face. *Fixed in v2*<a target="_blank" rel="ugc" href="https://codeberg.org/Gourieff/comfyui-reactor-node">ReActor</a> comes after the face detailer. It's set up to be able to do two faces, left and right, but you can use one face in both. Here is where you must take care. If you have a nice HQ single source you want to use, plug it in on the bottom center (normally bypassed). If you have a face model, use that. ReActor does not handle garbage sources well. The masking is set up to take out the spots where it usually struggles. If your likeness is off, tweak the masking before making major changes to the swapping widgets. Last step of ReActor section is a refiner for the balls. ReActed balls suck eyes. The detailer fixes and masks them, <a target="_blank" rel="ugc" href="https://civarchive.com/models/334668/eye-detailersegmentation-adetailer">the detector model is on this site I believe</a>. There is a loader for a ball LoRA as well. You can also turn off the eyes in the segmenter to bypass reActor and detail the originals instead.Last stage is Ultimate SD Upscale and a model upscaler. You have to be careful with the denoise, as too high a setting will ruin your likeness if using reActor, if not, you can bump it well past .15.There is a switch on the fast bypasser that will subtract one from the last seed used, in case you want to rerun a swapped image without reActor, or vice versa. Just flick that and toggle ReActor for a redo. Or click the left arrow on the seed box... yeah that's not much of a shortcut. Stupid.Face and ball detailers have their own positive conditioning. You can add one to $ upscale if you like as well, and/or add a refining model. I generally use two or three models, but I kept it simple for upload. I have to constantly remind myself that nobody cares, and if they do they certainly don't tell me.Oh the JK Aspect Ratio node has a preset JSON file in its custom node folder. You edit that to populate the list with whatever you use regularly. It's really helpful. But you can always just type it in or replace the node with a regular one. There's a right-click option to swap W and H in all of these nodes btw. For anyone not aware of that.*I forgot to add a strength slider for ControlNet. If you need it, go into the subgroup and drag the strength input dot to the left to connect it to the subgroup input. Then get out of there and drag out the new input that will show up, and select 'slider.' In the properties panel, set the slider to min 0, max 1, 2 decimal places, 0.01 step. The closer to 1 you are, the more your image will be constrained to the input depth map/skeleton pose.

Files with SHA256: c38d602ec6d1ea195c592860aadd262dc5c11e24823fd9066e0a69dbad51dfce