Thread 19910536

ComfyUI "Apprentice" Portable Pack

https://getshared.com/sJXk2iw1

>What is it?
Modified version of comfy portable (used YanWenKun's portable fork)
Aimed for new apprentices who wanna jump in AI world.
----Kijai's WAN-Wrapper nodes pre-installed
----Pre-installed nodes needed for my wan2.2 workflow (wf included)
----Pre-installed nodes needed for pisswiz's wan2.2 workflow (wf included)
----Sage 1.0 installation bat for older RTX cards (by Black Mixture)
----Sage 2.2 installation bat for newer RTX 30xx and up cards (by HerrDehy)
----Simple batch file by me that downloads all needed files (vae/clip/lightx-lora/ggufs)
----Included program called LosslessCut to join your clips together
----Couple upscale models included

>How
----Download the package
----Extract it on some fast drive that have good amount free space (100gb at least)
----Run the RUN_Launcher.bat and see if you get your comfy running
----....if so, good, close it up. It won't do anything as you don't have models yet
----Install VC_redist.x64.exe and reboot (most likely you have this installed already)
----Run Sage2_2-attention-install_v1.2.bat
----Run DloadESSENTIALS.bat
----....Dload options 1&2 and pick one of the GGUF models (Q8 Q5 Q2)
----....Downloading these models take time as those are several gigs

Once all downloaded. Launch Comfyui, reselect all models, to make sure
they are all there in your computer. Load image, type something in the prompt, run

My WAN 2.2 workflow:
https://getshared.com/WvLLXIqJ

I'm new at this but I managed to install ComfyUI in my laptop but I'm still getting used to it, does it matter if I already installed it?

>>19910536 (OP)
May I inquire as to how and why you chose that to be the flagship image? I mean, I get the bag, but the remainder of the picture is lost on me.

>>19910546
i recommend fresh install (just remember to backup your models folder from your old one somewhere before you do

>>19910548
dunno....have been adding obese fat men in my gens for a while now. Seemed "natural". And got your attention didn't it? Thats the point =)

>>19910564
I was wondering if it was an edit from a film like Fight Club, because my mind was just saying "His name is Robert Paulson."

any remove clothes, titty drop, kiss, cum shot loras for wan 2.2?

>>19910570
Bob's tits are way bigger

is it possible to do something like this with the workflow? it automatically inpaints clothes and makes them nude. video 2 video thing.

https://tensor.art/template/892762790157149507

>>19910970
Bob's got bitch tits.

bump for effortposting

got a good one afrer many abominations lol

who got a nice list of prompts for nsw stuff,undressing,fuck and suck ,etc.my prompts are very short because idk how to write something that makes sense for the ai,specially making things happen simultaneously,5 or so seconds are too short

Thanks dor the effort man, mich appreciated.
Noob here, I got a few questions.
What settings do you use to obtain a good quality without stressing the gpu too much?
Frames, samples, ecc... This stuff is driving me crazy, lol
Also, is there a way to make clips longer than 5 seconds?
I noticed that if I change the frames I can get longer or shorter clips, but the video speeds or slows down accordingly...

>>19911249
it feels you need like a nasa computer to do longer than 5 without messing quality,you can make another gen from the last fram and glue them together

>>19910536 (OP)
Managed to make it work, you're the best, man!
Can you link me some good loras to make the magic happen?
Thanks!

if i try to use a pic with nudity,i get out of vram error,but normal pictures it works normally(??????????????????????????????)

malware detected

>>19911288
was wondering about this. proof ?

how can I queue several images?I owuld like to let them cook while i sleep but whenever I queue more than one i get ouf of vram,even if i have a clean up vram step

I spell you the Wiz Site!

motionmuse.ai/r/vop8zjl9

>>19911259
https://civitai.com/ for Loras. You could also try civitaiarchive.


>>19911288
Bag seems like a trustworthy person, if you can see the previous thread,
>>19863970
he helps other anons, and I used both 1.0 and 1.1 of his workflows with no apparent problem at all. He still a Jedi and I hope he never goes to the dark side to cause headaches for other people.
Also, I checked the rar files in VirusTotaldotcom and just one or two files flagged something (with 1 security vendor that is Microsoft) and could be a false flag like the Fit Girl games repack ones.

I think it's cool he streamlined the process for newbies. Personally I prefer to install everything step by step to learn a little bit.

>>19911300
I didn't have this problem with the 1.0/1.1/Wan 2.2 workflow. Which workflow are you using? I do have this problem with the pisswizard workflow and tried to connect the clean vram node with no success.

>>19910539
Sorry new to this. When you say pick one of the models whats the difference between Q8, Q5 and Q2 ? Should I only pick one or could I download all 3 ?

>>19911668
These are quality models. The higher, the more quality and also more time generating. It depends in which kind of GPU/RAM you have.
>12gb VRAM GPU? Q8
>8gb VRAM GPU/16 RAM? Q6 and below. But I heard you could ran Q8 too. But it would take more time.

>>19911711
i run q8 with 8 vram and 16 ram,gen times varies a lot,lets say ten minutes per gen(5-7 seconds)

>>19911734
Which GPU and workflow are you using?

where can i find good prompts or some sort of template for different scenes(blowjob/missionary/bukake/etc)

>>19911752
gtx3060 ,workflow idk i think its the firstone posted some threads ago with some parameters changed
>>19911240
made this

>>19911753
On Civitai, if you go in the Lora you would like to use, there is a section called "Gallery" below with the gens and the Information icon with the prompts/etc.

>>19911759
Rtx 3050 8GB here. My gens take like 30/40min, lol. But recently I discovered I'm running 1.0.6 Sage Attention and will try to upgrade. Optimize time. Thanks for answering.

Is there a way to download peoples workflow ? I think that would be best for learning for me.

>>19911860
See third reply by Bag, download link with the workflow
>>19910544
Also previou workflows in the first replies
>>19863970

>>19911875
I get this error when trying to run the workflow

KSamplerAdvanced

No module named 'sageattention'

>>19911904
I don't know which workflow are you using. But will probably need to install Triton and Sage Attention to work. Follow this blog post:
https://www.patreon.com/posts/easy-guide-sage-124253103

Also, read the notes that Bag leave inside each workflow. He give good instructions and tips.

>>19911904
brother the second post here>>19910539

has you install sage attention

just disable that node if you cant get it to work. it'll run fine, but might take a bit longer

I'm a total noob but I can't seem to find any NSFW loras on civitai?
Are they hidden somehow or am I searching wrong? Tried to look up oral-insertion-wan-22 as recommended by pisswiz

>>19911929

right here nigga

https://civitai.com/models/1874153/oral-insertion-wan-22

>>19911288
This might be the reason:
https://www.reddit.com/r/learnpython/comments/1mtmhbc/pyinstaller_making_malware_exe/

So it's false positive. But if you have doubts, don't install. Other suspect might be the sage installer as it makes a new launch bat when installed

>>19911249
Well good quality is kinda subjective. I'd say making 5sec 640p vid with 6-8 is good enough if the model is good enough.

Wan isn't designed to make long vid 4-6secs seems to be the safezone. If try longer, action might stop, go reverse do something else weird

>>19911929
After creating an account on Civitai, go to account gettins and enable "Show mature content" and all "Browsing Levels" too

>>19911929
You need account, and settings and enable show NSFW......... Thou after their porn policy update some nsfw loras are fucking pain to find in there

For old removed loras (undressing, tittydrop etc) go civitaiarchive

>>19910971
Not in my wf. You need VACE models and VACE workflow to make v2v gens. My makes just i2v

Hi Bag. Got a couple of questions after testing the workflows. Bear with me and I hope you can help if you have the time.

As you can notice that the volume of the cumshots change in version 1.1 in relation to 1.0 with single sampler. I like the less cum version (because of realism reasons). I tried changing the prompt, lowering the weights of the loras (and even running just one lora), but the I can't quite figure out how to bring down the volume of cumshots for a realistic personal preference.

I also tried using Normal Sampler in 1.1, but the outputs came with blurry/artifacts in the video.

I'm also using Piss workflow. But after a successful gen, I tried another pic and had this error:

"comfyui_windows_portable>pause press any key to continue...." and it closed/ComfyUI disconnected.

I read somewhere that it could be because I had too many tabs on chrome or using it while it was running. Or maybe It didn't cleaned the vram after the first gen. What do you think it could be? This never happened with your workflows.

Related to this, some other anon said he couldn't queue gens with Pisswizard wf. So I tried to copy and paste the vram node from yours and inserted after Video Combine, but it doens't seems to work like yours (when it shows cleaning in the cmd)
How do I insert a vram-cleanup in this workflow?

https://www.mediafire.com/file/qd6ilmpzq738zd6/Pisswizard_W2_BJ.json/file

Also, how could I insert the save last frame option in the same workflow? It would be nice to continue the gen onwards.

Trying this out, got it up and running but crashes when trying video.
Trying to learn image instead for the moment but need a workflow for image-image changes, preferably several or where to I can switch loras I guess?
Anyone got tips on where to find workflows?

About the outputs in Wan 2.2 (Pisswiz wf), what influences the velocity of the pumps/bj/dp? In the video, my gen(left) the girl just take the tip and the other one feels like a real deepthroat.

The penis pumps/velocity is influenced by the Lora strength? Q models? Total steps? CFG? Better GPU card? Or it could be that he's using a different lora that it's more effective? Wan 2.2 is kinda new and there aren't many loras around.

I want to understand, so I can adjust and have a better output.

I'm running on RTX 3050 8GB, Q6 models, 16 Ram.

Thanks for creating all these workflows. I'm testing your version of Wan 2.2 this weekend.

>>19912282
Well there are lots of variables; what lora(s) you, is lightx in use, if not what cfg value you use (if lightx on, cfg must be 1), what sampler, how you prompt it., and how long the gen is.

Same ^^ answer to; >>19912375
MODEL/STEPS/OUTPUT RESOLUTION affect quality (and the gen time). Getting more ram or better card only affects speed. If you use speedloras, like lightX, cfg must be "1".
On your leftside gen i assume you used only oral_insertion lora? On right side it looks like it is company by some deepthroat lora (or is prompted differently, different lora strength or just got lucky noise seed, most of the AI is just randomness anyways. Sure Wan2.2 is quite new, but are new loras coming out almost daily

>>19912282
If you like more realistic cum, there are couple cumshot loras for wan2.2 that make more realistic cum, i suggest you move to 2.2

>>19912329
Google can get you some examples if you try hard enough. For example:
https://www.reddit.com/r/comfyui/comments/18vdm02/basic_comfy_img2img_workflow/
First result in that threadeads to an article explaining the creation of a workflow. Reading these articles, though sometimes arduous, help you learn the bits and bobs of Comfy that can help you not only build your own, but ask better questions and see what nodes exist. That being said, that article doesn't cover masking, which is a key component to inpainting as opposed to img2img.

So I would advise googling and trying different workflows.

That being said, a good example is a good example, and I know the Generic Inpaint guy has
shared his workflow in the past for an example of inpainting.

>>19912438
>>19912329
https://archived.moe/r/thread/19896380/#q19896597


PS: GPU specs/VRAM total?

>>19912329
what does it say on your launcher log? Also you could try...

>>19912478
and/or enable lowm if you and smaller card

Does past Wan 2.1 loras work in Wan 2.2 without needing the high and low?

>>19912793
you need to load wan 2.1 loras in both lora-loader.....some loras work ok, some not so much. You need to trial them out. Prefer wan2.2 ones when possible

any good loras for tittydrop/shirt lift?

>>19912796
https://civitaiarchive.com/models/1423479?modelVersionId=1608935&is_nsfw=true

https://civitaiarchive.com/models/1387238?modelVersionId=1567740&is_nsfw=true

https://civitaiarchive.com/models/1597473?modelVersionId=1807726&is_nsfw=true


noticed this on civ-archive front page...
>HuggingFace has been on a deletion spree and has already removed 16TB worth of files. This means, despite our best efforts, more of our links are broken now.

thanks op.made a video of my boss sucking dick and coooooooooooooomed to it,not posting because i am paranoid lmao

Ah ok here's the thread. Thanks for coming back you're doing a lot of good work teaching people.
I hope to keep coming back to learn and teach others as well.
>>19912804
>has already removed 16TB worth of files
Yeah this started like two weeks ago. I got on the ai wagon barely two months ago. This is bullshit.
What we're gathering now is just a grain of sand compared to what was available.

>>19912804
do any of these unbutton shirts/unzip jackets?even if its like not perfect lol

i cant really get the girl to blow the guy in a picture,its always a random dick appears and she sucks it or sometimes the dick would spawn in the guys chest lol

>>19912804
Hey AId.bag. Not related to the post but any idea how to get a similar result as this?
Its done with an online page but I want to recreate it locally. To my unterstanding the current best way is to animate the still, then create a generation for the sex part. However I have no idea which type of lora could achieve such similarity form a single image as I know I2V only adds to the existing image. But the second part seems more deepfake than I2V. Are there loras doing something similar (as in keep the original face and likeness of clothes but change the scene completely?)

>>19913486
>Its done with an online page
Link?

>>19913486
that second part is just pov missionary/doggy lora used on some t2i ai image (with a faceswap)

>>19912804

these work for wan 2.2?

>>19914289
at least rip clothes do...haven't tested other two, most likely, might need higher strengths thou

>>19914316

thx babe

>>19914196
how would teh faceswap be achieved? Also obviously her tattoo is exact or pretty close as well

>>19914550
I don't think it's a faceswap, I get similar results with VACE

another example

>>19914560
VACE takes a lot of resources, doesn't it? The two tikes I have tries I got OOM with a 3060, but I might have had workflows with ambitious settings preset.

>>19914550
First result in google for video faceswap worked for me, but it is bottom of the barrel. Uses Reactor, so it just breaks the video down into individual frames and swaps the face on and recompiles. Anything that interrupts the face causes reality issues (like cocksucking)

>>19914578
>Uses Reactor, so it just breaks the video down into individual frames and swaps the face on and recompiles. Anything that interrupts the face causes reality issues (like cocksucking)
nta I've been trying to get that node running for weeks and that just burst my bubble. What are people using for faceswaps now? I mean reactor sounds faulty.

>>19914704

Depends on how you apply it, and you need to have the NSFW fix applied.

https://openart.ai/workflows/neofuturist/easy-video-faceswap/HWSPPeffnHTbrIX6FWE1

Produced this from..

>>19914709 (You)
...this pic and a still of the video. As you can see her bangs cause problems

>>19914716
Yeah you're right I was wondering what that shading was. I guess this isn't perfect yet.
Thanks for the resources though. At the least maybe I can get to some image swaps.

>>19914716
thats not looking like her at all though. in >>19914564 the face is very similar even keeping the nosering

>>19914704
On the lfip side, you get something close, it can work a lot better. Subway chick and Sonna Bella, but if you watch, you can see when she puts her fingers in her o=mouth in one of the vids, reality breaks down

https://files.catbox.moe/7uefgd.mp4
https://files.catbox.moe/g8q6bl.mp4
https://files.catbox.moe/w6vsp0.mp4

>>19914718
Well the eyes, nose and lips are all there. It's just the ring that's missing in that example.

>>19914722
That being siad, I'm too lay to search for pornstars that match what I want, download their full videos, and then extract sequences from the video to make faceswaps (like I did with Sonna/Subway)

It isn't hard, but I'm lazy.

So if anyone knows a searchable database of short (30s-90s) porn clips that is like Porn Pics, that would be appreciated

>>19914722
>>19914726
The last one turned out great. I guess it's a matter of not putting dicks in the mouth.
I'm wondering how censored vids would turn out. You know like a jav. I can only imagine the mouth would phase over the censorship and look like a nightmare.

>>19914730
Great question, and the next step would be finding a video workflow that incorporated controlnet, but I have been too busy to test, and videofaceswap has probably been left in the dust since workflows like this thread surpass it, and I have been fiddling with that moreso. And my entire other list of shit to try.

That being said, the faceswap ones are way faster to gen, and probably less resource intensive, but you get what you pay for.

>>19914717
https://github.com/leez7onefl/WorkflowFaceSwapComfyUI

https://www.reddit.com/r/comfyui/comments/1i43l3v/guidehow_to_disable_comfyuireactors_nsfw_filter/

>>19914704
>What are people using for faceswaps now?

Bump this question, is there better tech for face swapping or is it all WAN img2vid now?

>>19914783
VACE

>when the ai makes the girl prettier and now she doesnt look like treal one
:((

>>19914797
fucking kek iktf. I mean I knew she was a dog but they I didn't ask it to upscale her.

>>19912430
Thanks for the reply and the detailed explanations!

I've read somewhere in civitai that Lightx being turned off could provide a little better movement, but of course will take more time. I need to test to see how it goes.

You're correct about only using oral_insertion lora, I'll mix it up next time and experiment with other ones and strengths/prompts.

I saw the new Cumshot in Wan 2.2. Looks very good and realistic! Only problem is that the liquid is transparent/kinda invisible. But someone commented that if you pair with the Anime Cumshot, you could get good results. That's a tip for anyone who is already doing gens!

I'll test the 2.2 workflows tomorrow. Thanks again!

Anyone has a clue on what lora is used in the last 5s on her being bent over? It's amazing.

>>19914939
>Anime Cumshot
I ran a comparative test between
cumshot-54epoch
T2V Facial
Cumshot Aestheic (Anime one)
And I concur, the anime one was the one I personally prefer.

The only thing was I need to prompt better. I like the one I was using on 2.1, it was real good for just having a cumshot come from out of fram, this one always tries to put a spoon, a malformed penis, anything as a cum source, or if they're holding something, it'll shoot out of that.

Tentatively putting Dicks_Epoch100 in as well so at least it isn't a bowl.

"A huge load of thick white cum shoots from the right side of the frame and lands on her face and chest. Smiling, tongue."

If anyone will please help me set up ComfyUI and genuinely talk me through it I would appreciate it? I have kik/telegram if anyone can ? I have just installed comfyUI and no idea how or where to start

>>19915006
I'm up for a bit yet, but I don't use instant messaging apps with strangers.
Send a link to a non register chat room or ask here, because your question is too broad, because what do you want to do with comfy?

>>19915006
You might want to read up on what a checkpoint and lora is if you're literally just starting out. Maybe grab a workflow to make some images. Get a grasp on how this works.
You can also jump right into the videos too. Both are good. Just follow the guides above. The more you know the easier it will be to overcome roadblocks.

Guys anyone please help me with this error, i have been behind this for days..please y'all, you would literally be saving me. Thank you.

the error:
KSamplerAdvanced

PassManager::run failed

>>19914948
Any to that with this? Maybe even a split roast of her and two other african men, one doggy the other deep throating her mouth?
Or am i asking too much?

>>19915006
If you don't want to follow the first replies of the thread or even another thread mentioned with previous workflows, I would do what anon said here:
>>19915138
I know at first it seems complex, but that's why the Apprentice Portable Pack exist and it's all packaged nicely.

>>19915247
There is a mention of the problem here:
https://github.com/kijai/ComfyUI-HunyuanVideoWrapper/issues/45
Looks like a Triton+SageAttention problem.

>>19914991
Hey, that's a nice test! Looks like it mixes well with the anime one for sure. Thanks for sharing the prompt as well. I'm pretty sure the wizards are cooking a Dicks lora for Wan 2.2. But I'm glad some 2.1 loras works too.

When trying to download the models via Run DloadESSENTIALS.bat in a second it says
done! implying its already finished or they already exist but I did a fresh install on a new system so I doubt it. Can I check for the models in the folders somewhere?

>>19915568
In comfyui/models/unet

>>19915610
they seem to be there however get this error when I run your workflow. Google suggests the vae model is faulty though I got it from the portable package

whats teh best way to iterate fast in the 2.1 prompt,i am trying to get the prompt right so i dont mind quality but i want to get the movement/angles right,what is the best settings to lower to make it go faster?

>>19915568
Found out that the DloadESSENTIALS.bat only downloads 0kb models thus making the run fail.trying to get the models manually now but any idea why this happens?

>>19915714
No, idea..... You can download them manually. Links are in infobox and where to put them

>>19915735
infobox where? i know where to put them but I cant find the exact models by googleing

>>19915796
bottom right hand of the workflow itself

>>19915806
thx

Just to let you guys know these builds work with SD images.
wan2.2 lightning
jfj-deepthroat-w22
8 steps

So, I was running again the 1.0 workflow after trying out 1.1, but I had this problem happen after Ksampler is finishing. Tried to open the original 1.0 .json and reselected all the models but it happened again.

Anyone has encountered this problem before and know how to solve it?

>>19916135
Picrel is what appears on cmd.

>>19915646

you probably need to select the file again. the one I have says Wan_2.1_VAE.safetensors

notice the capitalization

>>19916140

Strange, the one listed inside the note has no capitalization and worked before. I even downloaded again from the link inside the note just to be sure.

https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/blob/main/split_files/vae/wan_2.1_vae.safetensors

>>19911240
this was a fluke and whenever i try to do something with two girls either one will dissapear/get pushed out of view or some weird shit will happen,like merging face,a second dick oming out of the first one,dicks manifesting in some girls mouth,idk lol how ,put some stuff in my negtative pormpt but still

>>19916156
Always always always remember to RESELECT all model/clip/vae/Lora when you open up a new workflow first time. Even thou they seem to be there in the workflow.... They point to wf maker computer locations, not yours

>>19914991
Using "normal" and anime cumshot loras work pretty OK, you might want to set up highloras pretty strong... 1.5-3 and use more steps on high side..... On 8 steps: 5high, 3low

>>19916156

use this one and reselect

https://huggingface.co/QuantStack/Wan2.2-I2V-A14B-GGUF/blob/main/VAE/Wan2.1_VAE.safetensors

>>19914316

for the rip clothes one. what's a good prompt to have it so they dont hold onto the shirts anymore? every time they keep holding onto the shirts after they reveal breasts lol

any good dick loras so it's not an abomination

>>19910536 (OP)
getshared link isn't working on my chrome for some reason. keeps saying it can't establish a secure connection. i looked it up and did a SSL and cache clear both for my system and chrome but some pajeet tutorial is saying to turn off the firewall to connect to the website because this is a error? can op provide another link or check it idk

>>19917101
dunno many stiles that accept 5gb+ uploads....site loads fine for me:

direct dl without ssl: http://getshared.com/handler/download?action=download&download_id=sJXk2iw1&private_id=0&url=http%253A%252F%252Fgetshared.com%252FsJXk2iw1

>>19917122
lol spectrum internet is blocking the site externally from my system through my network. dang didn't know they were dicks like that.

their exscuse is that the site contains a download that would harm my system. false positive right?

>>19917211
why isn't this on mega files instead

>>19917218
thought of that....but i think mega uses dload quotas...shit to wait hours for last gig

>>19917211
is the whole site blocked or just that file url? does the workflow link work?
>https://getshared.com/WvLLXIqJ

>>19917237
whole site blocked externally though network

>>19917249
oh...well that sucks....just because it "may" have some files that are unsafe. Does it let you go to torrent sites?....most trustworth places to get your softwares

>>19913486
>>19914196

>>19917315
Is that possible with this photo?

>>19917315
Or this one?

>>19917331
>>19917334
not a request thread

>>19917336
Got it

>>19917315
>>19917336
looks good. so you just made the second part as general animation with t2v and put her face on if with roop? Or did you somehow use the original image as reference? Sorry being absolutely new here and gut the firs gen running yesterday using your workflow.

>>19917315
This is so good

are the loras not included in this portable pack?

wasted all night trying to get this pose.idk it just refuses to do it,what lora can be useful for it?idk how the sex version is called lol

>>19917315

this explains nothing

>>19917480
its an example of how he would tackle it

When did Reactor become so hard to install?

up

not the pilllow cock lol so this girl has an of but only posted a bunch of solo stuff ,if i try to use a nude photo ,I get out of vram despite loading some loras that should give reference for naked people

any suggestion?I feel that if i try to so something more complicated than taking shirt off or blowjob my computer will inmolate lol

>>19918316
this was the prompt

blowjob, A sideview of a woman seen giving a deepthroat blowjob to a large, uncircumcised penis. She is quickly bobbing her head back and forth sucking the man's penis while looking at him, the penis is going deep into her mouth and throat. The woman is engaging in a deepthroat. The man's penis is visible as it enters her mouth and throat multiple times. The woman's mouth and throat are clearly visible, showing the depth and movement of her throat as she swallows. The man's hand is gripping her head, indicating control over the depth and rhythm of his thrusts. She smash her face against the man's hips. She swallows the whole penis. cum strings falling from her mouth to breast.

>>19918329
Is that oral insertion Lora? Add at beginning "man appears in frame from left side, woman sucks the penis."

>>19918363
it was 'wan ultimate deepthroat v2 224epo 3nk',gonna give that one a go later

>>19910544
>https://getshared.com/WvLLXIqJ
Sorry, got here late. Anyone got working download link?

>>19918423
mirror: https://www.mediafire.com/file/ulc14bnecr9m21s/1stAIdbag_WAN2.2_v1.11.json/file

>>19914196
>pov missionary/doggy
Any idea what Lora was used or something similar?

>>19918693
probably "pov missionary" one on civitai....don't you guys look anything up yourself?

i used wan2.1 anal lora on >>19917315 but there is pov doggy and pov anal (and insertion loras as well)

Are there any best practices for loras, like how many to use simultaneously or anything?

>>19918828
generally you don't want too many (specially if they are trained todo the same thing)....loras don't stack up....sometimes they compliment each other, sometimes they confuse each other. Other things is loras quite big and they are loaded into mem as you hit run....bigger model/lora load = more time needed

i suggest use one or two loras for one 5sec actions, then use last frame as new input, change loras if needed, make another 5sec, join clips together

>>19917464
using roop/facefusion/reactor on pre-existing image and do a faceswap works ok, sure, then load that as your 2nd clip input and prompt away. Sadly they are not so nsfw friendly nnowdays, so there is some fileediting needed. Also programs like forge and fooocus have faceswap (and image prompt) options

i did it in comfy with sdxl model using ipadapter using og image as reference.....ran batch of 50, maybe 10 are usuable, pick best one and load it up in wan. Been a while since i played with IPadapter, but getting ok results

Note that every faceswap method tends to alter the face, it's kinda impossible to get perfect one just by using one image as reference. And even if you do get good swap, wan will alter it again

Bump

bump
also any good prompt for blowjob in a pic with a guy next to a girl?i get a comically big dick coming out of the dudes belly lol

Quick question from a noob: I have a Geforce GTX 1066 Ti (I know, quite old, but I'm stuck with this for the meantime) would I get overheating/crashing issues if I try to use the PissWizard workflow or do I just need to lower the settings?

>>19920560
>Geforce GTX 1066 Ti

Maybe with Q2s and blockswap you might get it running, but it will be a long wait for mediocre results

>>19920796
>a long wait for mediocre results

How long does it usually take with a decent GPU?

>>19920560
Huh, I just pieced together a shitty backup PC with my old 1660ti, I could give it a try, but
>>19920844
In my own anecdotal experience, with a 3060, I have 10 minute gen times using First Aid's 2.1 workflow and 20 minutes with his 2.2

I might not have the workflow optimized, but Sage is working at least.

>>19921149
What kind of CPU does that PC has?

>>19921158
Older, xeon e5-1620 v2 3.7

If you mean the older one with the 1660ti, Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz

Yeah, I'm using old hardware, get over it.

>>19921175
>i7-4790K

Oh boy, mine is a i5-4460 3.20GHz...

I know...it's a very old PC

>>19921184
20+gbs of files download, lowest quality (Q2-K) ggufs downloaded, umt5 scaled, and all other necessary/reccommended files (lightning 2.2 4 step lora)

RUnning one test before a formal test of Pisswizard's workflow on >>19921175
(1660ti 6gn VRAM, Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz, 16gb system ram)

Estimation of initial test is 6-10 minutes, though no idea how long it will take, then we'll see what happens when I try the pissflow.

>>19921290
>Estimation of initial test is 6-10 minutes

I would be fine waiting longer as long as my PC doesn't overheats or crashes

Bruh how can I stop the 'cock sprouting out of a cock' gens lol
I have like three synonyms in my negative prompt

>>19921290
>>19921358


Seems like no vid gen for you, see image related. 53m.40 seconds of loading then a crash when
"Patching comfy attention to use sageattn"
I was hoping at least it would give me a time estimate, but crashed right at that step.

Full error log below if someone is smart enough to see if there was something on user end making a mistake.
https://files.catbox.moe/m0cf3z.txt

Image of workflow to see if there are any obvious errors there.
https://files.catbox.moe/3y0ucw.jpg

Second attempt is underway, but without meaningful change to the workflow or pointers about what to improve, I expect another failure, but who knows.

Unsure if there is a startup command I could add to ComfyUI to make it more stable.

Unsure if First Aid's workflow would be better, but I doubt it, since they are similar in nature, but I am not expert.

>>19921290
>>19921358
>>19921569
That being said, that was Test 2. Test 1 was accomplished first, because I knew it would be quicker, and could act as a consolation prize. It does not use Piss workflow, but a simple video faceswap workflow.

I took the following Pisswizard gen:
https://files.catbox.moe/5tgeh9.mp4

And faceswapped the following image:
https://files.catbox.moe/8afqrz.png

For the following result. 11 minutes for 11 seconds of video, so presumeably 1min/sec, but further testing would be required to confirm. I didn't watch the whole video oddly enough and didn't know it had a blowjob portion, so you can see its limitations.

>>19921569
>Second attempt is underway, but without meaningful change to the workflow or pointers about what to improve, I expect another failure, but who knows.

Second test is confusing. WebUI has it as unconnected, but terminal has it as running and no sign of error. Not sure if I should restart the server and reattempt let run.

>>19921587
Turns out terminal crashed, so I am going to chalk that up as a definite no without meaningful modification.

You can still do Stable Diffusion 1.5 image generation and basic faceswapping though.

SDXL is a little much for the 1660ti in my experience.

>>19921569
>>19921575
>>19921625
Aw man, though the quick video was OK until the jaw of the girl got disjointed in that glitch.

>but a simple video faceswap workflow.

Which one did you use?

Also, thanks for doing the test.

>>19921569
did you enable low-vram mode in launcher? also like said, blockswap might help ya

>>19921654
also im not sure if sage works on so old rtx cards....if s, it would be sage 1.0

>>19921652
https://openart.ai/workflows/neofuturist/easy-video-faceswap/HWSPPeffnHTbrIX6FWE1

>>19921654
No, I was going to ask about that launcher, I have never seen it before and don't know how to activate it. I always modify the start up .bat with notepad. How do you access the launcher?

>>19921663
I just followed the Patreon instructions, and yeah
>RTX 20xx (Turing) and older: Limited compatibility – you may experience issues.

>>19921652
>>19921666
PS, that's the way faceswap works. No large differences between the two faces. So blowjobs are out, but it is workable with a little of thought and reasonably quick for what it is.

>>19921666
so you are not using my package?

>>19921654
Also, I don't know how to loop in the blockswap into a workflow. Is it on one of yours? I think I have an old version of yours with iti, the 2.1 one?

Your link in the OP is expired, so I can't swap it it without transferring it, and I forget which version of yours I use.

>>19921679
Dude requested Pisswizard's, I tried Pisswizard's.

>>19921683
i can add blockswap into 2.2, gimme few

>>19921679
Oh, what you meant. I had ComfyUI freshly installed for image generation and modification, and just popped Sage ontop when he asked. I DID NOT use your package in the OP, because I just assumed video wouldn't work since SDXL wouldn't. I'm procrastinating, so I figured why not give it a try when you indicated it MIGHT work.

Going to your workflows now, transferred from my main computer.

https://www.mediafire.com/file/3sdeojepdg0fywa/new_fa.json/file

https://www.mediafire.com/file/ywd08jfemqjzit3/old_fa.json/file

>>19921693
>>19921715
https://www.mediafire.com/file/zz4u7gducz1kgqm/1stAIdbag_WAN2.2_(v1.12).json/file

added blockswap + switches to turn it off

reselect models (you should use Q2s)/vae/clip/loras....try 3sec 360p clip first

>>19921766
ACknowledged, and while you're here, I noticed in your 2.2 workflow, Teacache high/low is disabled by default. I thought Teacache was an accelerator, not an upscaler, am I misinformed.

>>19921770
it is accelator....but it does the acceleration by skipping steps....now with lightx you only need few steps (so there are not that steps there to skip)

you can enable it and see the results, but i suggest don't use it when you have less than 10 steps

>>19921766
Results of running your normal 2.2 workflow.

I really hope that isn't a Triton install error, but I don't have the time to do a fresh install following your initial package to ensure Triton/Sage was properly installed. It seemed to have installed correctly.

I'll give the blockswap one a try, and that's it for me tonight.

A useless aside, it is easy to enable in the workflow, but I'm in the minority, I hate nicely organized workflows like yours, I like have a space between all the nodes, easier to trace and navigate. I was trying to reverse engineer where the blockswap is and its functionality, because I know it was modifiable in your old workflow (how many blocks swapped), but I cannot find it trying to trace the lines. Unless it just plugs into the stuff above the disabled Teacache nodes and goes into the background and is no longer modifiable.

>>19921847
Aw shit, I didn't do step 5
https://www.patreon.com/posts/easy-guide-sage-124253103

>>19921847
Just try it without sage for now...no bother struggle with it if the shit won't run anyways

if you want to edit blockswap just unpin/expand the node .....oh and i accidently left seed to fixed on that workflow

>>19921358
>>19921908
Last post for the night.
No idea what this error is, but trying a run with no sage before bed, and maybe a no blockswap and no sage, but it consistently crashes after one run. So we'll see in the morning

>>19922012
I don't remember well, but it could be you don't have the latest ComfyUI version. Use the bat file inside folder "update" - update_comfyui.bat

>>19922040
wan 2.2 needs comfy > 0.3.46

>>19922044
Fresh install, running 0.3.56

The brutal part is it takes an hour for it to hit the actual generation stage to see if is functional or not, so I doubt the actual generation time will be any better.

Stupid love of science to see how low I can go when I got a decent computer over there that can do it in a reasonable time frame.

>>19922049
so you on 1060 or 1660? also how much ram?

>>19921175

>1660ti, Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz

Bought a used computer to put my old parts into. Only has two DDR3 slots, so 2x8 for 16GB ram.

I mean, I could buy 2x16, but again, this is a backup or going to get sold to recoup investment. At best, it was going to be the image generation computer and then the other could become video generation, but honestly, I don't need to generate that much anyways (and it being unable to reasonably do SDXL).

Just a weird coincidence that it was near specced to this guy
>>19921184
who was wondering if it were possible, and like I said, ran with the procrastination excuse.

>>19922063
>>19922079
Maybe I did install sage wrong, but we'll see in the morning.

>>19922012
try disabling both sage attention models. worked for me.

how did I do?proably can be improved a lotbut i am satisfied with the face

>>19922705
welcome to the playground =)

>>19922108
Wow... So it works!? Post what you got on output...... 3mins per step.... You can run it 3+3.....it's reasonable if qlity is decent

>>19922705
Which loras you used for that droolinga in the mouth? Or it was just a prompt?

>>19922794
woke up this morning and the terminal crashed and the UI reset, so I have no times, and only one gen in the output folder.
Them Q2 models are shit.
Oh well, like I said, this was for science.

>>19922814
>>19918376
this lora but also in the prompt idk which caysed it

>>19922883
>>19920560

Works, but total shit fest :)

up

>>19921666
For some reason I'm getting an node error when I try downloading the YoutubeVideoLoader, ReActorFaceRestore and ReActorFaceSwap

Any ideas on how to fix it?

>>19924056
I just went through a fresh install and Reactor can be a bitch to get going (not sure about the youtubevideoloader, that one installed fine using ComfyUI Manager->Install Missing Custom Nodes)

https://github.com/Gourieff/ComfyUI-ReActor?tab=readme-ov-file#installation

It probably is having trouble building Insightface. I went through the three options listed in Step 1,
and it didn't work until the third (not sure if this was a cumulative result, or if the third worked by itself)

Then the NSFW fix didn't work until I transferred the original fix from my old install.

>>19924084
>reactor
Let me tell you something about that motherfucker.
Whatever you're trying to install, check the readme first and get the EXACT version of whatever extensions it tells you to.
I personally had to force windows to correctly identify python before even getting started on all that.
AND IT STILL DON'T WORK BECAUSE I'M LAZY AND DON'T WANT TO USE PORTABLE.

up

Leaving aside the videomaking, is there a decent workflow for removing clothes? Be it for IRL or Anime

>>19924728
Any inpainting workflow would work, but I think there are automatic ones that identify clothes and remove sans inpainting, but I haven't looked hard for one or stumbled across one. Something about Segs?

Google has this for results I haven't tried:

https://openart.ai/workflows/datou/remove-clothesflux-kontext/9tLLRdzuI0UT57QezPva

https://mimicpc-application-data.mimicpc.com/user-assets/f1315060-cfad-41dc-8a7b-65107ad7bc19/ce0aba4b7f6e4531a1179d32b2dfc2be/-5ea607b7-ede4-40cc-9a4a-4c98650d4770.workflow-49png

>>19924798
I actually tried the first one, but I only got a duplicate of the sample image

I don't know if I did something wrong, with both a real woman and an anime one, just in case

>>19924812
The second one doesn't seem to have a download, but with the image, you could rebuild it yourself.

Here's Generic Inpaint guy's workflow, but it is one of the manual inpaint ones.
https://www.mediafire.com/file/ulk3pjenatu6pyf/Generic.json/file

>>19924812
The second one is actually in the image, so if you drop it into Comfy you could try it.
https://mimicpc-application-data.mimicpc.com/user-assets/f1315060-cfad-41dc-8a7b-65107ad7bc19/ce0aba4b7f6e4531a1179d32b2dfc2be/-5ea607b7-ede4-40cc-9a4a-4c98650d4770.workflow-49png

Here's Generic Inpaint guy's workflow, but it is one of the manual inpaint ones.
https://www.mediafire.com/file/ulk3pjenatu6pyf/Generic.json/file

how do you minimize the freaky glitchy ass eyes

>>19924884
From Wan 2.1 to Wan 2.2, the eyes and face definitely got better. Besides that, better loras, adjusting weights and prompts, I guess.

who did it better?

>>19925439
tho how can i get them both in the video,idc if each sucks a dick or both hte same one

>>19925442
https://civitai.com/models/1395313/wan-dr34mjob-doublesinglehandy-blowjob

You could try that after dick is on frame. It's Wan 2.1 Lora but worth a try.

bump

>>19924084
I managed to fix the error with Reactor, I did have to install a few stuff in the PC and the python_embeded folder to fix it tho.

I'll leave them here if anyone is having issues with Reactor:

Python_embded:

litellm
GPUtil
cython
insightface
sageattention
timm

PC:

Build Tools for Visual Studio

I got simple inpainting kind of working, but how do I make fakes in different positions like that one dude on simpcity who makes all the sabrina carpenter fakes? Just talking img2img I guess, not interested in the video ones

>>19926631
examples?

>>19926640
Like these, there are a bunch of sex ones too and I have no idea how they make it on comfyui

>>19926657
I assume you read the guides from the realistic parody threads on \b\?

I am not well versed in it since I haven't tried generating fakes of specific people, but I did play around with the DMD lora as per the settings therein.

https://rentry.co/biglust-training-and-loras

There are specific loras for generating specific people, and there are workflows you can try. Again, haven't tried them myself though because I assumed you'd need to make a lora of the taregt and the instructions say that takes more GPU than I have. But in retrospect, I should at least check out the workflow to see if I can incorporate a shortcut or jerry rig something.

>>19926657
its either
a) trained lora and text2image
b) text2image with reactor node to do the faceswap
c) ipadapter/instant id reference image + text2image

>>19926657
>>19926705
Yeah, the workflows in that guide are 100% dependant on a lora of the person in question and are txt2img.

>>19926714
The one in your pic is using IPA adapter properly and is img2img, isn't it?
I have never succeeded in finding an IPAdapter workflow that does good face, is that yours or from somewhere (probably yours looking at the color theme.)

Any chance you'd share or point to a similar one to play with?

I updated from within comfyui today and now the workflows that did make i2v are no longer working.
Its giving me errors on loading that I ran out of memory.
Is this actually because I updated or something else?

Dropping this in here as I discovered it yesterday. The motherlode of WAN NSFW LORAs:

https://huggingface DOT co/ApacheOne/WAN_loRAs/tree/main?not-for-all-audiences=true

>>19926705
Get BigLust 16 checkpoint. It has the most celeb LORAs our there. BigLust 17 is a bit better quality but not many LORAs.

Get the dmd2 SDXL 4step LORA and add to prompt, no trigger word.

Steps set to 8 max
CFG set to 1.5 max
Sampling method must be LCM
Schedule type set to Karras.

It gens super fast as a result.

Get celeb LORAs here: bt4gprx DOT com and search for BigLust16

That's it. Then just prompt using the celeb LORA. Set at about 0.85 so the celeb LORA doesn't limit the main prompt too much.

Then take txt2img gen to Img2Img InPaint, mask over face and run the same settings but just the celeb LORA in the prompt set at 1.0 and the dmd2 LORA - no other prompt info. This will inpaint a much better version of face.

Adetailer or HiRes might help but I can't use them as my rig can't handle it.

>>19924798
This is the best one I have used by far, and free. Just use a burner email. You get a lot of gens before you have to make a new account.

promeai DOT pro

Take the output into InPainting to then get better results after

Just a quick question to Comfyui experts.
I've generated a custom lora with a certain girl's face.
Great result, can't complain.
I'm trying to make her pose in NSFW stuff, using Wan 2.1, text to image.
What kind of NSFW loras should I search for?
I downloaded a bunch but none of them seems to give the desired result when I stack it in the lora node...
Any suggestion?

Promising cumshot lora is coming. Still looks a little bit cartoony compared to the 2.1 versions.

>>19926783
Yeah that is my wf, been trying to find good settings for it but.....kinda meh. The way ipa works it's basicly t2i but it injects the ipa model/reference into it while generating. You can get good resembles with it...depends on input images. Going to play it around some before posting it

reactor method is more i2i, it makes better swaps but has its limitations, like blowjob images

>>19927561
Great find! Only thing missing would be a single image to see the effect. Prompt to trigger is important too.

>>19928203
The files alongside the loras should have a writeup about them including trigger words (The .md file accompanying?)

>>19928194
Your results being kind of meh is nice to hear, because yeah, I have had such limited results and just thought it was lack of skill/familiarity.

Still would be interesting to see bbcshopper's workflow, his gens are still plastic shiny but have pretty good face detail.

>>19928227
i think bbc use flux models...they tend to have that "waxy" feel to them...dunno how he adds faces

>>19928194
Went back to the basic IPAdapter workflow to play around some more now that I figured a thing or two, and holy moly, does the IPAdapter model matter.

Downloaded the base four from hugging face, and three of them gave abominations, but the fourth, especially with DMD, is doing things, txt2img with base face.

But without DMD, it does things too, but I'd have to fiddle with the KSampler settings more to optimize.

>>19928691
using faceid v2?

>>19928736
Not even, went back to basics and somehow stumbled into not shit.

>>19928739
pretty good on 0.5 weight, hair is bit off and missing nosering, but those might come in just by prompting.....also lower the try lowering the "end_at" to give more deeper bj

>>19928691
Damn, thats a nice gen

>>19928739
Nice number of tabs, anon

>>19929103
one day maybe I'll read a comfy guide and actually understand the details of the ui, but until then, right->load workflow, number of tabs be damned.

(I have three browser tabs above with three seperate workflows for testing purposes, probably with just as many tabs open in each)

>>19928193
Tried it out with the example prompt and the outcomes are pretty realistic. Used no other loras with it besides lightning

Bump

can someone reupload this thing to megafiles. still blocked on spectrum internet

>>19929934
i make account and mirror it

>>19929948
medal of honor to you

>>19916135
>>19916236
Sorry for the late lateeeeee reply.

But I have a theory that the v1.0 Workflow from Wan 2.1 doesn't work anymore because of the ComfyUI being updated to the latest version.

Anyone had success using the very first workflow with the ComfyUI updated here?

>>19929971
it should work...does wan 2.2 one open ok?....i can modify wan 2.2 workflow into single sampler setup so you can use use wan2.1 on it

MIRRORS


::::Apprentice pack:
::::::::https://mega.nz/file/z5QDlQbI#U3GRmFugXHCbuRQakyLKLH0MZ0lZtSTfX-91luHo8cU


::::WAN 2.2 v1.12 workflow (blockswap added):
::::::::https://mega.nz/file/fsZVAapK#T1g8SUlgELHpSitWr6iVqfLXTZpjIadmsX32ExL8aHo


::::WAN 2.1 v1.13 workflow (based on 2.2 one but single sampler):
::::::::https://mega.nz/file/ilhmARTL#ZvV93hRQA7XojDhpcTBSkC70BClHI2ARd8auQTe3GNE

>>19930001
All wfs in Wan 2.2 work okay, even v.1.1 from Wan 2.1 works from what I remember.

The reason I came back to use a little bit of v1.0 is because of the looks of cumshot in single sampler. They seem better. Maybe because the author of the lora spent more time adjusting.

But the new loras coming up in Wan 2.2 are promising and the motion/look are getting there.

>>19930041
Awesome stuff! Will test it later

Just wanted to share this kinda gen with a zoom effect using this lora:

WAN 2.2 Anime Cumshot Aesthetics

Prompt
"a man stroking his penis appears from the side and slides his penis into her mouth, holding his huge and wet penis. as girl open her mouth, the penis drips and ejaculates a huge stream of semen or cum into her mouth. (the camera travels into a detailed view of her inner mouth, revealing cum or semen inside her mouth), she is choking and her eyes are sad or tearing, (the cum accumulates in her mouth, her mouth is flooded with semen or cum)"

Tried a couple of pics and looks pretty good. Only some bugs with the dude appearing close to face and going up. Maybe a dick lora will improve/fix that.

>>19930205
been using that prompt for a while, works great
make
1st clip with oral insertion on, save last frame
>Man appears in frame, woman sucks the penis
2nd clip cum aesthetics on.
>girl sucking and giving a deepthroat blowjob to a man's penis, the penis thrust into her mouth. suddenly, the penis leaves her mouth and the man starts stroking his penis. as girl open her mouth and stick tongue out, the penis drips and ejaculates huge streams of semen or cum into her mouth. ((the camera travels into a detailed view of her face, showing her sad eyes, wet skin and inner mouth, revealing cum or semen inside her mouth)), (the penis keep ejaculating massive streams of cum or semen into her mouth and face), (the cum accumulates in her mouth, her mouth is flooded with semen or cum) cum on face, cum on forehead, cum dripping down

oral insertion usually does better job aligning the penis

>>19924824
I've been doing some testings, but for some reason Python just dies without error when it reaches the Alimama Node

After loading an image, writing a prompt and clicking run, generation always gets stuck at this step, why??

>>19930393
always post the pic of the whole workflow (pref. with log window visible

also whats your rig, what models are you using?

>>19910536 (OP)
Hi 1st.AId.bag!! I've already use ur first workflow for wan 2.1, and i got consistently very good results with my 5060ti 16vram and 32gb of ram.

I just wanted to thank u for giving this pack for wan 2.2 and ask u about the performance in both cases, in my case i saw that 2.2 is more time consuming (i got sage installed on 2.2 only, in 2.1 didnt installed it) but i dont know if im missing something because in 2.1 i was getting a 5 sec video in 5/6 min aprox but now for 2.2 for the same 5 sec video is like getting 12/13 mins but deppending on the image i got one video in 6 min, is this workflow more sensitive to resolution maybe?

Just giving some thoughts, this is an amazing job thank you again =)

>>19928739
could you share that wf as well please??

>>19930900
I am travellingg with work until Sunday eve, but that is the complete workflow in the pic if you want to rebuild it.

>>19930900
>>19931475
Ibguess itnisnjust a slightly modified version of one of these, I forget which

https://github.com/cubiq/ComfyUI_IPAdapter_plus/tree/main/examples

Is Sage Attention 2.0 recommended and functional in RTX 30xx or it's better for 40xx and above?

>>19932759
Sage 2 should work on 30xx cards

up

what loras and strengths are we using for these facefuck gens? im trying with facefuck_epoch_80 and oral insertion but i cant get it as clean as some people here have been

>>19934410
looks fine to me....aside that it is 16fps. What do you mean by clean

>>19934452
great example here, i cannot get this effect, i usually interpolate to 60fps but i skipped that since im doing tests

>>19910536 (OP)
I hope the next Xbox has W12 so I can use these portable programs with hardware for less than $800.

>>19932759
Speaking of the 30xx cards, is the 3060 12GB any good when it comes to gen wan2.2 vids?

>>19934478
Woops, I meant to quote >>19933315

>>19934478
usingc 2.1, 10min per 7 sec gen, Q8.
Wan 2.2 takes 20.
Anecdotal evidence and not sure how optimized my setting are.

>>19934458
well....you dont have to keep oral insertion weights that high, it does the insertion at very low strength, more weight on the actual facefuck. Also some facefuck loras have insertion trained in, no need to have 2 loras that do same thing

then there are prompting....if deep action you could try;

"penis all the way in her mouth and throat, her big lips are touching his crotch, completely covering his penis."

"woman giving a deepthroat blowjob to a large, circumcised penis with extremely fast back and forth movement. she swallows the penis all the way. Her lips touches the man's groin. Her nose smashes against the man's hips."

....just try stuff out, thats half of the fun

>>19934494
yeah, thats why i was asking which loras ppl were using. i think my prompt is perfectly fine, pretty similar to what u have written

>>19934503
Try one without oral insertion, just deepthroat one.... BTW there is new deepthroat lora out on civitai, jft one... Was one of the best ones on 2.1. Haven't tested it yet

>>19934489
20 minutes for 7 seconds, geez

>>19928739
>>19930352
Does anyone have a screenshot of what a WAN 2.2 workflow looks like using it with nsfw Loras? I have no idea where to add Loras or what it needs to be linked to in order to load them and work correctly.

>>19934522
high loras and low loras on their own loaders....if using wan2.1 loras load same lora in both loaders

>>19934512
i did see those before u posted, not bad at all
for some reason i have it on low strength, .9 and .8, but its effecting her facial features too much. could just be the pic idk

>>19934543
lower the LOW lora strength more, that helps on some loras to keep facial features

>>19934535
Thanks! I wanted to see this, but it seems like it's going to be difficult for me to understand and add those two loaders without generating an error. I had no idea two loaders were needed with WAN 2.2.

>>19934543
What did you use in the prompt to achieve that brief disgust reaction?

>>19930815
Not AI d-bag but yes the resolution keeps unlike Wan 2.1 that automatically scales to 480p or 720p. Also tried 2.2 on a 5060ti last night results are 4 to 500 seconds for 8 step and 4 half, or try lowering to 4 steps and 2 half for faster gens.Took me forever to get Sage installed I manually input the patreon guide itt codes albeit changing the python standalone folder directory name to python embedded, like the default and notes in the opening of the workflow says.

Just got a pc running capable of this fun. 5090 9950x3d 4tb nvme 64gb of ram.

I've got comfy up and running locally, not portable. Seemed like most recommend local.

Is there a recommended guide for next steps? Need to get Wan 2.2 running, Loras, etc.

I tried to get Sage 2.2 but couldn't manage to finish the process using a venv environment, and at this point don't mind missing out on efficiency gains.

>>19934654
you listed all the reason why i made this pack...

Im very very very new , I have some logic , I can hadle and upgrade if I could learn but where and how can l learn and start to use comfyUI step by step ?