example
md5: 8fea45768a0a6a409662417de92f489b
🔍
I have been trialling a workflow that combines controlnet and faceswap, because my attempts at using IPAdapter have been lack luster.
If you provide a source picture for the faceswap and an input picture for it to pasted onto, I will run it through the workflow.
It is more effective than a regular faceswap because I can play with the prompt to change hair color and other details, but since I am using canny controlnet, the overall picture structure will stay the same.
The closer the body and face shape of the source is to the input, the better the swap should go, but this means nothing on the face like sucking a dick, as it will interfere with the face swap.
Low resolution as I use SD1.5 and it seems to like empty latents under 1280, so be prepared for potentially blurry results.
Using this thread as a testing ground, so no garuntee on quality whatsoever, just tinkering because damned if I will actually learn something by reading how it works when I can brute force my way through experience.
>>19750556Still trying to find a sweet spot for output size, can take up to 3 minutes per gen, and these ones I am trying multiple sizes
>>19750574if you are using 1.5 models, they should be outputs of 512x768
for sdxl/pony, i usually use 832x1216
>>19750578does controlnet input image size have/require consideration with regards to the latent size?
Pic related is 50% of input image
>>19750548
>>19750548>>19750578>>19750589vs. reccommended 512x768, no other inputs changed.
Either way, I will keep those dimension in mind.
>>19750548>>19750578>>19750589 (You)
vs. reccommended 512x768, no other inputs changed.
Either way, I will keep those dimension in mind.
951test
md5: 66f19935266cba2a755ae66de6e7ca0e
🔍
>>19750594i booted up my old workflow and can it doesnt look like the controlnet image has an effect on the latent size
>>19750550>>19750556
>>19750635>>19750637goddamn, that's quite the quality swap.
832x1216 isn't that much larger that 512x768, maybe I'll get Biglust or something and give it a whirl in this.
>>19750632If the AI will cooperate, I was trying some ssbbw stuff the other day and it would not play nice, but fat seems to work okay.
>>19750641im not using canny as the controlnet model - im using the SDXL ControlNet Union Promax model
i am using novareality-Pony as the base checkpoint
for the reactor part - face restore model GPENBFR512, with faceboost.
Faceboost - GPENBFR1024
Facerestore - codeformer v0.1.0
TNPOCfJ0
md5: 2c382d0bdfdc2f1aed92b43205369732
🔍
>>19750659I'm not sure if it's straight face swaps your doing but this kind of body/pose is ideal
>>19750659Oops that's the edit someone did for me, here's the actual pic
>>19750662Got distracted investigating some sdxl stuff.
Straight faceswap
IMG_7401
md5: ba1a94f6e199043b33167acc92501152
🔍
Do anything to her please
>>19750713>>19750711I like them both, thank you!
What's the process like face swapping to an animated image?
>>19750736>What's the process like face swapping to an animated image?Beyond my means, and know-how
>>19750724
cover
md5: 0554562e8124b47d9cb0c797e965fd3c
🔍
>>19750664Would a face swap be possible on this?
Idk your workflow process handles creating similar images?
>>19750777 a different face pic for reference
>>19750777>>19750777made the workflow to take similar images and have a bit more control on the colors essentially.
Your riginal pic was giving me guff, so just threw in a cowgirl anorei pic, because those seem to work better than side pose for lack of a better term.
Although this image was giving coloration issues too
>>19751440might have a better one pop out later, trying something else to better merge the mouths
>>19751455Yeah mouths are always a tough one. They always end up looking like a monstrosity. Thanks for the time i cant edit anymore so i depend on other wizards
>>19751459Got a couple pics Ref coming up
>>19751471>>19751459These are coming out well! Props wizard.
>>19751490thanks anon, looks great taking dick. could you try to again and keep the bbc black? if not all good, this is hot!
>>19751478expect delays, isn't working as consistently as I would expect
>>19751485Any interracial, I'm not smart enough to prompt two seperate characters, so they are either both dark skinned or neither.
>>19751496
>>19751528thanks, no problem just curious
>>19751509other one was coming out abominations
>>19751533Love it, possible to add cum? Thanks so much!
>>19751486I thought it would be a easy slam dunk, but so close, yet so very far
>>19751558This pratical tryout is not going as well as expected, have a straight faceswap
>>19751560Seems like always, I can't gen complicated multi person poses.
18
md5: a8024487ed20f4aa0d212148368c6ae1
🔍
>>19751608thankyou anon! please try another with her?
IMG_0036
md5: 0b2bbd427f7442f34738bc0b77a1258d
🔍
>>19751625need this photos face pasted onto it
12
md5: 6c01b2bbf3e76ce8844f6a83835f3c43
🔍
>>19751626omg its so hot would this work if you want to try it!
thanks!!
>>19751637anyway to make it cleaner
>>19751653>>19751655Out of spite, I went and got some photos that I figured would have a higher chance of success
>>19750653not what was asked for, but it is something.
IMG_6832
md5: 7403398f38be6387a6d08d61f0d5e4b2
🔍
Please, it’s her birthday
>>19751702By any chance can you share the workflow?
>>19752140show me more of her to my discord. will fake her.
tserriednich.hui
>>19752140If you can make them work better let, me know
https://www.mediafire.com/file/fhkslljyqnamani/Set_Custom_Latent.json/file
>>19752140>>19752140https://www.mediafire.com/file/vs4fdfafa29s54u/Takes_Input_Image_Dimensions.json/file
>>19752165Awesome I’ll check it out
>>19752168Have a disc? I can send you my findings easily that way
>>19751481At least when I find one thing that works, it is easily reproducible, so the workflow can be reasonably predictable.
>>19752180I assume everyone on here is a pervert or a federal agent and attempt to minimize and direct links to my human being, so no, and no offense.
>>19752188weird how swapping between workflows, which essentially only changes the size, can have this effect.
>>19753096genning was abominations, have a faceswap
>>19753876These are absolutely incredible! Any gang bang related pics? Thanks!
>>19754022Posting input picture, thank you for trying
can you try her with the linked ref pic please?
bumping for the great tester to come back and try more requests
Could you do this hippie if you get a chance
bumping for more nekkidness
Strap
md5: 05fec023f712419a91d8e1a053c9406c
🔍
>>19755029to this picture...... pls
2
md5: 8b466431fd18ca3516275c2837d4c5f1
🔍
>>19754022I believe I can safely conclude that regular face swaps are generally better and easier than anything I can actually do with the resources I have.
1
md5: 92edb3f9e3e36ac49dd18f13a57f77f2
🔍
>>19754745Although I would have the ability to better match certain things like hair
4
md5: 1be88c32b37bcb865301e7697771cfbc
🔍
>>19754789Although if I wanted to do extreme swaps and the surface content is negligible, there are some positive trade off results
6
md5: 2ccca48b0c5a7eea37ae246402d7c9af
🔍
>>19755029Maybe by better incorporating loras, I could increase reality in certain aspects, but again, my rig limits this possiblity in general cases.
5
md5: 18ffe5219c26797c7fad33dece4a488a
🔍
>>19752119but again, extreme cases of plagarism are possible where regular faceswaps would not be
7
md5: 0d4d37c91cd4c94af90785083901bbed
🔍
>>19753119I still haven't extensively tested inpaint sketching
8
md5: ed0fc44d05dd5338928b0518c20fce38
🔍
>>19753359Inpainting is still funner overall though.
/end
>>19755498thanks anon, it looks great
