Thread 19863970

Previous thread is gone:
>>19799683

OG 1.0 Post by 1st.AId.bag:
Many have asked, so;

https://www.mediafire.com/file/i1pslzko0hr7hy1/1stAIdbag_v1.json/file

It's my first workflow from scratch, so bare with me, gimme tips, fixes, let's revision it better. I assume you would know something about comfy, if say there are some errors/bugs/wtfs. i'll try to help (i'm kinda comfy-noob myself) where i can. I don't wanna guide how to install comfy, those tutorials are just one google search away

it's dual workflow in many ways, you could use diffusion or Q (gguf) models, it has option to use single sampler or dual sampler setup (later needs lightx's self force lora).

Maybe keep this in general comfy-faggot-q&a thread...

THIS IS NOT A REQUEST THREAD!!!!
posting your bitch in here will make i won't do it, ever
THIS IS NOT A REQUEST THREAD!!!!

>>19863970 (OP)
Updated Workflow Version 1.1 Post by 1st.AId.bag:
Update workflow little bit

-Dual sampler is now just single sampler using lightx self-force lora. Does the same job, same quality imho
-Replaced the NAG node with different one, since quite few had probs installing the old one
-Replaced random seed node with other one
-Replaced the clear vram node with other one, when swapping two sampling methods around, its good to clear model after run.
-Removed brigtness/contrast node
-Added blockswap to options for fast enable/disable
-Removed colormatch node from bypass list, i think there might a a bug on the node as when bypassed, it stops the process, i tried using any-switch between it and raw vae-images, but it would output input image as output video. Workaround for now is to set colormatch strenght to 0 if you dont wanna use it
-Added separate sliders for lightx/normal sampler steps, so you dont accidentally run normal sampler with low steps or lightx sampler with too high steps
-Added cfg into options for normal sampler method (lightx cfg must be "1")
-Added links for fp8 diffusion model and clip
-Updated lightX lora link. There now i2v based version. This new workflow uses it, it's important to get this lora in order self-force sampler to work correctly! Movement seems better now! So get it

Workflow 1.1: https://www.mediafire.com/file/p6atp5r6m5zjwx5/1stAIdbag_v1_1.json/file

>>19863978
Updated Workflow Version 1.1 Post by 1st.AId.bag (part 2)
Did a test and made a fresh install of comfyui_portable using this fork of it:
https://github.com/YanWenKun/ComfyUI-Windows-Portable/releases
Once all unzipped, rename folder "python_standalone" --> "python_embeded"
Then download:
https://www.patreon.com/file?h=124253103&m=441360653
Place the bat in root of comfyui_portable and run it, install triton/sage/and libs, no need for kijai nodes and wan stuff
Once done, rename python folder back to "python_standalone"
Launch comfyui, open my workflow, you will get popup window stating missing nodes, close that window, go to comfy manager instead, click install missing nodes, select them all and install / reboot comfy
All should be installed now, re-select model/clip/clipvision/vae/lightX lora file (so they point location in your computer)
Open image, load up your some loras in powerlora loader if needed, prompt it up and run
All went good, everything installed on first try without hickups

>>19863979
Updated Workflow Version 1.1 Post by 1st.AId.bag (part 3):
this is 8-step run with new i2v self-force lora, As you can see, movement is much better now even on deepthroat runs

Feel free to discuss more about the workflow, prompts, loras, improvements and results you had it.

Also, thanks so much for the workflow, Bag! I know more people will use it.

Is there a version of the bat file that is compatible with the non-portable comfyui?

>>19864606
I dont think the workflow works in the bom-portable version. Follow the instructions inside the json file after installing the custom models/nodes with comfyUI Manager.

For the people that are trying out 2.2:

Do you use CLIP Vision?
Do you connect CLIP in your Lora loader?

>>19865697
workflow works just fine on desktop/pinokio/portable versions of comfyui...installing triton/sage via that bat might be trickier...if it has python_embeded, then it should work

Up

anyone have the lora for titty drop?

>>19867125
i2v-tittydrop-e26
google knows

>>19867133

thanks babe

up

bump

Running some doggy loras in wan2.1 and the dicks are coming out like chodes without heads... anyone got any ideas on how to get a realistic looking penis? Thanks

Could someone please share how to make animated tributes aka blowjob through the iPad screen videos?
Would you need a tribute picture first or can your subject be edited into it

>>19867975
I know what you mean. I've some okay results and some hilarious that even a guy appears or a hand thats holding a dick. Here is my setup/loras that is kinda okay for now:

facefuck_epoch_80 [ weight 1.0]
jfj-deepthroat-v1 [ weight 1.0]

prompt:
Man out of the frame puts penis inside woman mouth. Mouth with tongue showing receiving penis inside. Fast thrusting penis inside woman mouth. Throbbing Penis. Woman looks up with desire. static camera

If anyone has better loras or prompts that work better, feel free to share.

I'm very new to all this I was previously using tensorart for gens after the request started flooding this board.
Tensor is dead now so I want to start looking into local gens, there's a tone of information to lean but before I go down that rabbit hole my PC is 8 gigs VRAM 16 RAM. I'm aware that's is not great for these programs but it it enough to get decent gens at 480p? I don't mind waiting a bit especially if I can queue multiple tasks (preferably not hours). I guess I'm mostly wondering if it's still worth looking into local with these specs or if I'd be wasting my time. Others have mentioned renting remote GPUs, I'd consider this as a last resort

>>19867975
>>19868612

Add the erect_penis_epoch_80 lora to it, that should fix it.

Up

So i recently tried out WAN 2.2 and i am using the same loras i had on 2.1 but the problem is that they just dont look right ... i tried the above mentioned ones as well facefuck_epoch_80
jfj-deepthroat-v1.

So does anyone have wan 2.2 lora recommendations ?
Thank you!

>>19868638
you might get it running on lower GGUF models like Q6 or Q4, but the the quality goes down a alot. RAM is you bottleneck.
You can get pretty good results pretty quick as acceleration loras need only 4-6 steps, but on low ram, but loading the model takes time on your 16gb ram.

>>19869813
Loras have to be loaded 2 x times, before high noise and then before low noise. Some old 2.1 loras work in 2.2 but not as good as they would in 2.1.
New 2.2 loras are trained and splitted in two this in mind. So far i havent seen anything good for 14B, gonna give it a good moth or two and try again.

I tried to run the v1.1 workflow but I got the following error:

Failed to get input node 0 for group node child 193:1 with slot 0

I would greatly appreciate any help

>>19869896
screenshots help diagnose it

>>19869900
Had the same issue after update. Seems the upscale node on the far right (little pink one labeled upscale) has something broken now.
I just deleted that and it started working. I couldn't figure out how to get inside it and fix it, so no upscale now, but it works.

>>19869954
you can unpin and then rightclick menu -> select expand. This option isn't present when it's still pinned

>>19870031
Thank you!

Up

bump

>>19869954
Thank you!!

Someone have this error ?

Error
Failed to get input node 0 for group node child 193:1 with slot 0

>>19872466

Yes , try this :

>>19869954
>>19870031

Could someone recommend LORAs to make realistic looking titties

Bump

I know it isn't your bag, but are there any wizards who have gotten WanGP to work?

I tried installing via Pinokio, and got a line 6 error, and tried via docker, and got several other errors.

Hell is there a portable version somewhere?

I got your workflow working, but it is slower than I would like, and wanted to try it.

Hi so i see a bunch of people on here making vids from cock tribs.
Does anyone here know how to do that ?
I am able to make blow job vids using the deep throat or ultimate blowjob lora but i cannot quite get tribs to animate correctly.
Thank you!!

>>19872681
From the looks of it, Civitai was nuked and models that was 18+ vanished. I hope a new site appear.

So here it's what I've:
https://www.mediafire.com/file/is4d2232h29330n/Loras_B%2A%2Abs.rar/file

>>19874741
>>19872681

github has some, find them via google, and there is a backupsite for civitai

https://civitaiarchive.com/

>>19873546
ask pisswizard, he uses wangp. Far as i know wangp is now SFW only, so you need to edit some files to get your runs

up

>>19874861
Thanks for sharing. I hope they last and more sites open.

What makes an image gen longer than another?

I am confused sometimes that an image that has smaller dimensions than another takes longer to img2vid than one with larger dimensions.

I mean, I know more loras mean more time, but sometimes the gen times I notice seem kind of arbitrary. Is it about pixel size or something else?

>>19877884
time comes from length of the video, output dimensions (size of your input image has no effect on gen time as it get resized before sampling process), loaded model/loras....and generally how much resources comfyui can use

Posting WAN 2.2 workflow later tonight. Preload your models: https://huggingface.co/QuantStack/Wan2.2-I2V-A14B-GGUF/tree/main

>>19878623
Thank you for sharing. Had a lot of fun tinkering with your 2.1 workflow

Up

>>19878623
Thank you Grand wiz! o7

Ok. My first 2.2 wf

https://www.mediafire.com/file/h3pg9iom48u5y95/1stAIdbag_WAN2.2_(v1.0).json/file

#Make sure your comfuyi is up to date (need to be > 0.3.46
#Strongly recommend installing sage version 2
#I added option to use wan 2.2 diffusion modes instead gguf, but from my test diffusion models output pretty blurry/ghostly videos. No problems with GGUF
#Important that you get LightX lora files for wan 2.2
#Links and more notes in the workflow
#People with cards with less than 8gb vram, should get Q6 or Q5 models. Q8s load ok on 12gb card (

>>19880482
Thank you!!!!

Does the workflow let you make longer videos too? I'm still figuring out how to make one over 6 seconds

>>19880482
Looks promising as hell. Thank you, Bag!

>>19880571
after 6 secs models and loras start to behave oddly...that is just how wan works, there is really no easy fix to bypass this limitation

My workflow saves the last frame of your gen, use that as next input (needs to be clear and sharp frame) and make another 5sec clip and then stitch them together.....You can do this few time before thing go too uncanny valley

>>19880482
Clean wf

up

>>19880482
In your opinion, how much WAN 2.2 has improved fromp previous version? Like 10-15% better quality and faster outputs?

>>19881976
Quality wise 2.2 is better for sure, as it is basicly 1080p model. Therefor is also slower than Wan 2.1 480/720p models. Wan 2.2 has better and more natural motions in my option.

Speed wise...... Well the thing is, you can do good looking gens with low as 4steps now thanks to acceleration loras. It doesn't really matter if step takes 30secs or 120secs if the quality pays at the end. But on Wan 2.2, what slows it down is the fact that it needs to load another model in the middle of your run. Unless you have nice amount of ram, this could take more time than the actual sampling.

Another downside of 2.2 is Lora managing, as you need two loaders, loading the right versions in right places needs some attention before hitting the run button.

But its been out for weeks, we will get better nodes and boosts for it for sure.

>>19880482
Great workflow. i am finding that it likes to change the face slightly immediately in 2.2 though. Didn't have this issue with the old 2.1 workflow so i assume it's the loras or something on my end. Great work!

>>19882342
most likely lora issue.....old 2.1 might have some other weird effects as well.....try running it again, but lower the lora strenghts

>>19881976
Wan 2.1 5sec-clip @ 624x720;
480p Q8 GGUF model
SAGE 2.1 in use
8-steps single sampler (euler/normal)
lightx + 2 other loras,
rife-interpolation ratio 2
"70 year old man appears in frame and rips woman's dress open, revealing her big nude breasts. Woman looks confused"
Prompt executed in 323.70 seconds (4070 super. 12gb / 64gb ram)

>>19882401
Wan 2.2 5sec-clip @ 624x720;
High+Low Q8 GGUF models
SAGE 2.1 in use
(4+4)-steps dual sampler (euler/normal)
lightx + 2 other loras on both loaders
rife-interpolation ratio 2
"70 year old man appears in frame and rips woman's dress open, revealing her big nude breasts. Woman looks confused"
Prompt executed in 335.34 seconds (4070 super. 12gb / 64gb ram)

>>19882400
Do you suggest making the high noise loras stronger than the low noise?

>>19882456
yes..what loras are you using? there are certain 2.1 loras that fill change the faces even on low strengths

>>19882462
I've tried them at different strengths but this is my latest. They are all 2.2 loras

>>19882489
i haven't get many 2.2 loras yet....but, try. Disable one of those cumshot loras, make short 2-3sec run, see if the face goes uncanny. If not, culprit is most likely the other lora

And on those strengths....lower the steps are --> higher the lora strengths. When steps > 10, i use 1.0 on high/end

>>19882529
Ok, i can try that. Thank for your help! Best workflows I've seen. Do you post these anywhere else?

thread deserves a bump

>>19882342
I went back to 2.1 lightx and it feels better for keeping the faces. But that's a feeling and not a full side by side study.

>>19880482
Clip vision doesn't seem to do anything (as you've noted in the wf). Also your lora output clip isn't connected (and from what I've heard others say it doesn't do anything in 2.2). gonna do some side by side with my own wf. I do really like the sliders for inputs, prob going to steal that.

>>19883431
Yeah I think clip v is baked into the model

wh UP ps

Up up.
Btw, jfj-deepthroat-v1 + facefuck_epoch_80 seems better than wan-ultimatedeepthroat-V2 for Loras. At least with the speed of the action.

Bumps

wanted to give this a go but files are huge lmao,maybe tomorrow when downloads finish

friendly bump

>>19880482
Hey,
Do you use a different workflow for the blowjob AI requests? I saw your screenshot there. You had a different looking workflow right? Would you be so kind to share?

>>19887061
i switch and swap all the time....either 2.1 wf or 2.2 wf, post posted here

>>19887075
Oh ok I got to try 2.1 too! Thank you so much. Do you know how to fix the error: "mat1 and mat2 shapes cannot be multiplied (154x768 and 4096x5120)"?
With the default settings and models of 2.2. Error message points towards High noise NAG and if that is disabled towards HighNoiseSampler

Anyone else notice that prompts react differently between WAN 2.1 and 2.2? I’ve found the motion and faces are a lot better in 2.2, but prompt adherence feels worse compared to 2.1

>>19887131
sounds likely wrong clip loaded....did you reselect all the models/clips/vae, etc etc when you open'd the workflow first time?

To everyone who faces errors....post screenshot of your whole workflow, and error you get....it's 1000 x time easier to help you

>>19887075
>>19887195
try selecting (or reselecting) this

>>19887149
My POV Blowjob and Deepthroat prompts are useless on 2.2. They never look good. Could anyone please share some prompts that work great on the new version?

Hi, im new on this and I would love to learn to do some of the AI edits that some wizards do around here, anyone knows of a good video or guide to learn the basics on this

>>19887297
well since you got comfy loaded, try loading my workflow, install missing nodes, download needed models and files, load image, type prompt and hit run

>>19880482
thanks bag, I got started with your 2.1 workflow, sticking with it for now and experimenting with different loras.
since a lot got taken down from civit i get weird gens not knowing how to prompt it exactly. but i wouldnt have gotten anywhere without your setup. do you use disc?

installed all the models but always get this

error CLIPLoaderGGUF 'NoneType' object has no attribute 'endswith

>>19888039
Reselect the models/clips/vae/loras EVERYTIME you open a new workflow so the flies point to your machine (not the wf maker's)

>>19887874
yes, leave yours if you need help on something

I think so. At least everything that I found. Here is a screenshot for testing purposes I didnt change anything except adding one picture.

>>19888807
>didn't change anything
....eeeeh, seems like you did...but anyways. Error is that you dont have right clip loaded (you are trying to load wan2.1 main model as your clip....didn't even know that is possible),

read the notes and save files in right places...you need this file as clip >>19887202

>>19888837
Now I see the issue in your annotation there is a link to download the 2.1 file that i tried to load there

>>19888850
It's 2.1 vae file.... Wan 2.2 uses that on 14B models....you need to download CLIP file into text_encoders, reboot, and run/have fun

>>19888609
thanks man,great workflow literal first time i can get this ai stuff to work lol

anons,any loras for general nfsw stuff you recommend?

>>19888943
make account on civitai (so you can see nsfw)...new loras almost daily. Depend what you want to make? more options for 2.1 now

>>19888889
thanks works now both versions

maybe my pc is not strong enough for this or some wrong setting idk,thanks anons

I could get her to show tits in another gen but cant upload it,got a fucked up first attempt when i was missing models but 4chan is not letting me upload tho it is less than 4mb
will check more models and stuff

>>19889215
https://www.freeconvert.com/video-compressor

good fast and free....click setting and set compression method to file size --> 4

Up because this workflow rocks

I don’t understand any of this shit holy fuck, someone point me in the right direction. I want to do AI stuff like this and yes I have a good PC. Where do I start

>>19890579

start with this: >>19863979 intall and get the models

How would the facefuck/deepthroat lora work on a regular laptop

>>19891091
it wont

>>19891145
How about a gaming laptop or highspeed laptop

>>19888837
I also seem to be having this same issue, I've restarted ComfyUI, made sure everything is downloaded in the right place but I don't have this proper clip file needed

>>19891242
Nevermind this I found the right file and put it in the right folder, had to look inside the 2.1 WAM for the text encoder

>>19863970 (OP)
>>19863978
Gonna finally try to install it. Just one question, Does deepthroats looks better in 1.0 or 1.1? I read on the previous thread that dual sampler was kinda iffy

first attempt doing these locally, didn't realize the Q2-8 implied quality/vram usage so this is with Q2 GGUF'S

>>19891360
The Q8 Models, honestly, didn't take that much longer yet looks 10 x better, can't wait to keep experimenting

>>19891380
I'm using Q6 and the results are good enough, in case anybody wants to speed a little or have low end GPU with 8GB only.

>>19891380
Also, the difference in the moments is astounding, lol

>>19891299
1.1...at those times lightx got updated, get the rank 128 one, pretty good movement when using ultimate deepthroat loara

>>19891425
>>19891380

i'd say Q6 is is the low point for i2v

Hey Bag. I've saw this kind of gen in another thread and was blown away with the results. Some user mentioned the wizard used WAN VACE. Did you have a trial with this method/models yet? Looks promising

Another mind blowing gen results with that "WAN VACE"

>>19891455
i've never used vace models actually, i know it keep reference input image (aka face don't change that much) intact more. wan 2.2 seems to do that pretty well on vanilla just as 1stFrame2LastFrame

One thing nice on VACE is that you could use reference movement from other video. But like said, i haven't playe'd around with it

>>19891458
dunno if if vace has 2.2. models yet, most likely so. ATM i try to add FLF2V into my wf, maybe vace version next

haha what the fugg

>>19891564
post screenshot of whole workflow.....you got something very very wrong =D

Thanks for pulling this together. Running into an issue where the image I'm using isn't being returned in the output? Its generating someone different. How can I fix that? Ideally that it completely uses the original image

>>19891569
yeah idk what changed,it was doing someting convincing before

>>19891653
did you install the missing nodes....some seem not loaded?

>>19891455
>>19891465
Hey Bag, I've successfully contacted one of the wizards that was using the methods mentioned and this was his answer about WAN VACE:
-----
WAN VACE is for Video2Video generation, like nidifying a tiktok video. For the gens you attached I was using this workflow:
https://civitai.com/models/1822764/wan-22-i2v-gguf-compact-wf-or-lightx2v-7-steps

Maybe try starting with this lora:
https://civitai.com/models/1874153/oral-insertion-wan-22
-----

The lora he mentioned has tons of great comments about being the best blowjob lora ever (and for some reason civitai is allowing nsfw again?). Too bad it's for the newer wan.

Just wanted to share, in case you want to analyse and incorporate in your workflow.
Seems like WAN 2.2 is a game changer. I hope gpus with 8GB can run it okay, lol.

>>19891465

This is what the maker of these vids (pisswizard said)

" This is WAN VACE, its for doing generations from a video source, rather than an image source. Video to video, here we are inpainting to remove her clothes but keep all the existing sound and movement. sir "

that means he uses WAN VACE for Video2Video

>>19891897

Also his lora and prompt that he uses, he shared in this thread:
>>19891604

"
A completely nude woman is on screen, she has big perky breasts, perfect breasts, perfect nipples, very perky breasts, very full breasts, full tits

nsfw_wan_14b_e15 - 1.0
genitals_helper_v1.0_e219 0.7
BoobSlider_WAN14_V1 0.15
"

>>19891911

for blowjobs:
"Try using oral-insertion-wan-22 lora sir"

>>19891897
>>19891911
>>19891915
I was read to paste the same stuff. Thanks anon.
>>19891773
Bag, picrel is his workflow

>>19891202

-What the internet says: Have minimum 6GB VRAM GPU

-It is more complicated than that. NVIDIA is better than AMD, CUDA, Cores, Tensor, and other electrical engineering shit matters, but the short and the long is more VRAM = Better, NVIDIA works better, but you can make AMD work.

-Normal RAM matters too, but the price barrier is lower than compared to a GPU

-Anecdotal experience:
Started with 8GB VRAM AMD, at the time, I had to make extra steps to make it work. It initially DIDN'T work, and I was genning/inpainting small simple images on CPU alone at 2-5 minutes a pop. When I sorted out the GPU situation, it went down to 30-90s for a 1024x1024 image and video was a garunteed failure.
Upgraded to 6GB NVIDIA, which had similar results, but video was no possible, but not practical, since 7 seconds of video took 10-30 minutes of gen time.
Upgraded again to 12GB NVIDIA, not 7 second gens clock in at 10 minutes flat.

The space may have changed and AMD might be more feasible or useful now, but I don't know. Also, I might be retarded and slow times are based on poor implementation of resources. But that's my experience.

>>19892137
>>19891911

He also shared his GPU details, uses Nvidia 5090 and his time required ranged between 110 to 200 secs

>>19892174
yeah, thats 32 GB of VRAM, and he also stated he runs 64GB RAM. I run 32.

So his rig is atleast 3 times, if not more, better than mine.

>>19891915
This good wan 2.2 lora, says insertion but does ok job on BJ

Up

>>19891983
dman this is insane

>>19891714
idk what happened but reset my pc and it does non abominations now,

>>19893169

look at this one of his. look at the movement of the table when she takes of her hands, movement of the liquid within the glass and also the reflection of her naked breasts onto the table on that bumppy surface, isnt that insane?

>>19893386
Pretty complex result for sure. Also, up

Come find me & degrade me on my pages, I fucking love it
Insta: @vanirahiomai
@dredlocksbyvanira
@sparklebyvanira
@tokeholders
@hiomaijewellery

what can I use to keep the face consistent?many times it will turn the girl into some generic face bimbo

Up

>>19894943
Workflow 1.1 seems better. But the WAN 2.2 is what you're looking for, if you want a more consistent face.

>>19894943
>>19895177
most common thing that change faces are the loras, either use other lora or lower the streght

Trying this out aswell, but my results are weird with pink on bra and the bra staying on even when I prompt to take it off.

I've been having a hard time getting anything running. I'm trying to use GGUF models with a single LORA. Anyone recommend a simple GGUF workflow for my 3090 and 32gb of sys ram?

>>19891458

holy shit this is fantastic. i need the workflow for this shit!!

>>19880482
so when my workflow gets to lownoisesampler, it just crashes with a pause. no matter what i do, i cannot get past that. any ideas? i can try increasing pagefile

if i can improve the resolution of the original pic would the definition of the video output improve:?>>19895455
yay lowering weights improved it

>>19896935
upscaling pictures always comes with a downside of making details look less like the original so you would most likely come out with a more defined video output but the people in it will look alot less like they actually do

>>19896368
You don't seem to have Lora that enables the undressing.... Try undressing/ripclothes/tittydrop lora

>>19896381
Shouldn't be issue running any gguf with those....

>>19896756
Get any errors? Post screenshot of your whole workflow, it's much easier to troubleshoot the issues. Also what card/ram amount

>>19897322

i was actually able to continue with increasing the page file. though, my results come out a bit blurry compared to most, anything i can change in your workflow to increase quality? i am on a 3090 as well. here is an example i did for a thread

>>19897744
doesn't look half bad.....it's pretty big image with very small face (not much to sample from). Post image of the WF so i can see the settings. Increasing steps is always option (note that this takes more time)

>>19897752

the only thing i changed are the loras. everything else is default. speaking of, these T2V loras dont work at all for cumshots or facial expressions, where are u getting i2v loras that work on wan?

>>19897765
there are some 2.2 cum loras at civitai....not as good as wan 2.1 ones in my opinion. Some t2v for pretty good on i2v as well

are you using gguf or diffusion models, on my test diffusion ones get blurry/ghostly

>>19897770

using the exact models u used in the workflow. sorry, i should post a pic of the workflow instead of being a dunce.
if i am using wan 2.1 loras, do i just put the same one for low/high noise?

>>19897774
>if i am using wan 2.1 loras, do i just put the same one for low/high noise?

yes....on high pass strengths can be bit higher. I also suggest that you crop your input a bit, focus more on the girl, but leave some room to man walk in. That bg is bit busy with all the trees and foliage wan tries to animate

>>19897803
thanks, started to get them a bit better. now just need to make these cumshots not so insane - u have any specific loras u use for more "normal" ones?

>>19897900
https://civitai.com/models/1858645?modelVersionId=2103700 ...this one pretty "tame" use @ 1.0, also note your prompt plays big part what you get

anyone tried running those models on rented gpu servers?

Know this is way below wizards, but for general newbies and ease of use https://unlucid(removethis).ai/r/eylkbe9r does great blow job edits, tit drops and cum shots.

Know this is way below wizards, but for general newbies and ease of use

https://unlucid(removethis).ai/r/eylkbe9r

does great blow job edits, tit drops and cum shots.

>>19898888
>>19898884

fuck off

what a waste of quads on a gay post with your shitty referral gay fuck

>>19898576
Someone has, and I thought it was this thread, but I can't find the post. Maybe the last thread?

any idea how to make fine tunes, what data is needed, how much, quality/preprocessing, how much compute/price is needed, what models are best to start from etc?

>>19898576

vast.ai are cheap and take crypto, and runs on people's home servers.

>>19898576
i have mimicpc account....prices range from 0.5-2.5$/hour. Most of the tools and models are preinstalled. Downside is that you cannot edit ai-tool's config/other files --> aka you cannot remove nsfw filters from facefusion or wangp for example. Comfyui has no nsfw detection so that is not an issue

Other good option is runpod, it's bit trickier to setup, but loads of tutorials avail how todo so on the net. you can install anything, edit anything etc

there is no difference if you run comfuy on rented cloud or your own local machine, except the speed

First of all I wanna say a huge thanks and shoutout to 1st.AId.bag for your transparency and support. Been following your posts as a fan and now finally found the time to learn about comfy, install everything, test and implement your 1.1 workflow. Now that everything is up and running I am having a really hard time finding good LoRas for the Wan 2_1 setup you showcased. Blowjob, Deepthroat etc. LoRas from your Screenshots are not downloadable/findable anymore. Any advive? Can you maybe upload your best of? Thanks again!

>>19899826
https://huggingface.co/ApacheOne/WAN_loRAs/tree/main

Or try civitaiarchive. Downside is you cant see the example prompts or trigger words.

>>19899826
You could try the normal Civitai as well. For some reason, the NSFW is back

Hey Bag, I'm on 1.0 still and I have two questions. Could a run 1.1 on a separate folder? Like "ComfyUI_windows_portable_1_1" If it's portable, there won't be a conflict with the 1.0 folder or needing to install sage two times, right?

Also, the same applies trying another folder with the pisswizard workflow he just posted? I want to try it out, but I don't want to lose your 10/10 workflow or have conflict models, etc.
Thanks in advance.

>>19900072
i dont really understand what you mean....but if you seem to think;

YOU DON'T NEED TO MAKE A NEW COPY OF COMFYUI FOR EVERY WORKFLOW YOU WANNA USE

just open pisswizard's workflow, open mine on next to it, open some other next to them if you wish. Workflows are just like blueprints, they share same models/settings like sage if available. My workflow, piss workflow, any other wan 2.x workflow use same models. My workflow needs some nodes pisswizard's wf don't and most likely same thing vice versa, but installing them won't affect how they work

And yeah i suggest using 1.1 or my wan 2.2 wf, loads of useless time-eating stuff on v.1.0

>>19900072

where is pisswizard's workflow??

>>19900126

>>19899802

But seems like piss is using my 2.2 workflow

>>19900139
Nvm, wrong file prefix, but basically same, mine have some extra stuff into it.

>>19900457

An amateur comparison of the workflows, though Pisswizard.

In my amateur opinion, FIrst Aid's is more funcitonal and works better overall, but Pisswiard's is admittedly preset for lower quality (in the expectation users with lower resources would be using it, and presumeably those with more resources would be smart enough to adjust it.)

Still not sure if it is worth switching from 2.1 to 2.2.

>>19900511
Like stated before....very low Q models are....well, very low

see:
>>19891360
>>19891380

I looked your thread. My suggestion is you set seed(noise) to fixed when you are comparing and testing stuff out, example how different strengths affect your output. My wf is most likely slower because interpolation and vram offload at the end (= force reloading model on every run)

bag i need your help i lost a lora can u help me out

>>19900901
what lora?

>>19900910
something like that

>>19900914
https://civitai.com/models/1434650/nsfwfemale-genitals-helper-for-wan-t2vi2v

>>19900918
thank you very much

Bag I need your help. My leg is stuck in the rocks and the water is closing in.

>>19900993
you can borrow this. you know what to do.....

>>19900993
>>19901003

hm it didn't work as I thought it would with the lora

>>19901168
https://civitai.com/models/1587277/ass-stretchgrab-wan-21-i2v-480p

>>19901195
didn't work either, thanks anyway, I'll try again later

>>19901354
what image are you trying and with what prompt?

>>19901365
I want to make this animation the same as before

I actually had the right prompt and lora then reinstalled comfyui and couldn't find it again

>>19901402
That is 100% the ass spread lora linked above

>>19901409
okay good to know then I must have had the wrong prompt

>>19901512
ok wait

>>19901512
wait wait wan pantiespulldown never seen that can u give me the link

>>19901534
>https://civitaiarchive.com/models/1420362?modelVersionId=1605460&is_nsfw=true&is_deleted=true

>>19901534
i dont think if that is even necessary, that ass spread or geni helper might also do that action

>>19901552
lets see where that goes

>>19901552
almost the same only yours looks much better

>>19901552
you're a sweetheart thank you very much once again :*

>>19901631
how many steps (and where you split) what sampler/scheduler are you using?

also dont use teacaches on low step

also also update sage if you haven't already

>>19901682
nta, what say you to just fucking baleeting everything under Q8? I tried most of them out and I'm not seeing much of a difference other than poor quality.
I'm still gonna mess with it a bit. And what did you mean by "switch sampler on this step"?

One of you anons ought to put a working copy in a Dockerfile for runpod/podman and use pierdat to download the models over bittorrent while it installs and share them while it's running.

>>19901682
sorry I just fell asleep

i think thats better

>>19902584
i dont think that lora works that great on 2.2, tried multiple times, with different strengths , but no good wide ass spread =/ gotta wait for 2.2 lora

>>19903253
I know, I have done a lot of tests but this was the best one

bumps

how do I fix this issues, sirs?

>>19900914
that's ycorrero's work. i recognize it.

Anyone have any method to run wan on colab, i dont care even if its colab premium

wheres the dick head lol
do i need an extra lora for that?
got genital heplers and deppthroat epoch

>>19904483
ye, dont know what lora hes using and prompt

>>19904727
>do i need an extra lora for that?
You do. erect_penis_epoch_80

>>19904428
one or multiple of your lora files is corrupt (due incomplete download) or you are using some non-WAN loras

>>19904911
dicks lora is better in my opinion

>>19904727
genitals helper is for vaginas, not penises. if you are on wan 2.2 try oral insertion lora, has pretty good penises baked in it (no black ones thou....well on my tries)

>>19905047

i used these lora

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

not sure if my download was corrupt will try again later


do you happen to have undress lora?? like titty drop and breast size helper, bag? thanks in advance

>>19902584
Does it start with pants on?
Can you share your workflow or loras?
I would love to create this type of videos

>>19905194
>>19901518

>>19905194
>>19901512

>>19905047
>>19904727
ahh I am in 2.1 idk if my pc can handle 2.2,st
added this >>19904911
fucking lmao

>>19905218
WHAT HAS SCIENCE DONE

is there a reason more often than not my videos end up trying to return to the 1st frame as if its trying to loop? using 2.2 I made sure loop is 0 and have added tags like "looping, looped video" to negative and it still happens.

>>19905204
Thank you Sir! can't seem to find to download the safetensors, any help or link?

>>19905218
pic?

>>19905252
my guess is; you are trying to make too long videos? try 4-5sec....

other thing that might cause this is that you have ping_pong enabled in the video combine

>>19905408
I was using 125 frames for length, took it down to 97 and it went away, cheers for that.

Also while I have you, do you recommend a particular cum lora, there's a couple I have found but they give weird results.

thank you for your help.

>>19905415
are you using 2.1 or 2.2 loras? 2.1 loras might work on 2.2 but not other way around

>>19905418
Using this one.

civitai.com/models/1350447/wan-cumshot-22-21?modelVersionId=2077123

both high and low loras, adjusted weights, used different prompts but the cum either looks like globby yogurt or it pours out of the girls mouth and onto the dick.

>>19905428
thats T2V lora (text to vid), they work sometime, but try getting I2V ones.....

https://civitai.com/models/1858645/facial-cumshot-wan-22-video-lora

https://civitai.com/models/1869475/wan-22-anime-cumshot-aesthetics-precision-load-i2v-beta-version

issues i'm having with 2.2 cum loras is that cum seems to "fade away" pretty quickly

>>19905443
ahhh, damn, i had it filtered for t2v, doh. thank you!

>>19905401
wan 2.1

>>19905218
wan 2.2

>>19905529

what lora you are using for this?

>>19905529
>>19905516
damn this is insane,thanks lol how do you do it?

>>19905636
yeah, kind of pointless just posting vids on their own, like "I can do it" .. real helpful

>>19905636
>>19905529

if you see file starting with "157"....like 157******_***.mp4 = was made with my wan 2.2 workflow

>save the file
>drag it on my 2.2 workflow
>it loads prompt, settings, lora weights etc (ofc it won't run unless you got same loras & models, and even you do, you need to re-select them). But anyways, you'll get some good info out from it

>>19905697
>>19905688

thanks babes

>>19905830
worked?

>>19905832

not sure yet but i just wanted to say thanks for posting steps

>>19905834
i tested today with 15 steps it took over 1 hour for 6 sec :D

>>19905840
do you get much motion range with 15 steps?

>>19905841
nothing has changed compared to 10 steps only it looked better

>>19905845

what are your PC specs?

guys is it possible to make a video like 20 sec long ? i keep trying but my comfyui seems to crash
i dont care waiting longer

>>19905963
better off making a workflow that takes the last frame of 1 vid and use it as the fram for a new vid, then stitch them all together at the end. there might be a better way tho. as far as I know weird things always happen after 6 seconds

>>19905963
not with wan, kinda limit stuff at 5secs....using last frame from previous gen to start the next gen is the way to go

>>19905972
nta I've been wondering how to pull that off. I mean it's already generating the last frame of these vids.

>>19905991
I havent doe it myself but I would just copy the workflow and paste it at the end and feed in the last frame, could allow you to add extra prompting for adding details, not sure if you can automate the merging of the vids tho. surely theres a node to do it.

ok thanks you guys

>>19905999
why the fuck you would copy paste the workflow? just save the damn last frame and open it up as new input =D

there are nodes to join videos together sure, but also plenty free lossless videotools that do it much simpler than comfy

>>19906017
im thinking more of a extended workflow that just flows onto each other automatically, i mean you could just do it manually but I kinda wanna give it a go, 1 click 2or 3 vid extension. maybe its impossible. ill find out lol

>>19906024
its not impossible but then you are putting a lot of trust of that last frame is; in focus, sharp, girl's eyes are open etc etc on the that last frame.......

>>19906037
could run it through a upscaler maybe, bah, gonna science this shit lol, gonna take a million years

>>19905218
How do I stop the mans face from spawning? Ive tried so may combinations when trying to bring a penis from the side
>>19905408

>>19906113
try adding "man's upper body stays out of the frame" into your prompt

>>19905047

hey bag. i still keep getting this error

>>19904428

is there something wrong with my workflow? im not using that picture btw. i just refresh it

>>19906192
have you re-selected all the models/clips/vae/loras?......so they point to your computer. I assume so as you have Q2s loaded (thats pretty low.....)

Thing to try; disable all loras and see if it runs....
if yes, enable one lora, test run
if yes, enable another, test run
....
until you hit that error again

If error without loras = Could be that your model files (GGUF) are fucked up (incomplete)

>>19906241

troubleshooting now.. i think i did fuck up my gguf when i downloaded so redownloading larger version

any decent anus loras?

>>19906249
start with Q8 if your machine can handle it....if you get out of memory (oom), move to lower Qs

>>19906269

thanks. everything works great. i did run into an error about sage node (it was on auto by default but I dont think i have it installed) so i disabled

>>19906279
i strongly suggests you install sage....30-40% faster gens

>>19906265
there's a general nsfw lora for 2.1 that may be what you need.

>>19906121
>man's upper body stays out of the frame
didnt work, im on lighting wan 2.1 though, so maybe its more rudimentary. the mans face keeps spawning every fucking time

>>19906291
how are you referring to the male figure in your prompt? are you saying "a man enters" ? if so try "a penis enters" or "a male frame" try some other variations .

>>19906283

idk if im blind but i cannot find it in list? can you please assist. this is what mine says in terminal what version i have

>>19906301
remember to rename the python folder before you run the sage bat

>>19906313

oh. so it's automatic install?

>>19906313

sorry what do you mean by rename folder. sorry for questions i am new to this

>>19906342
go to comfyui folder....rename python_standalone to python_embeded

save this:
https://github.com/HerrDehy/SharePublic/blob/main/sage-attention-install-helper-comfyui-portable_v1.2.bat
to you comfyui root and run it....follow the instructions in your terminal window

after done, rename python folder back to python_standalone

>>19906369
hey Ald, you have experience with hosting a discord and running commissions? seems like someone with your talents would make some scratch. or do you think the law will shut that shit down eventually?

>>19906402
>hey Ald, you have experience with hosting a discord and running commissions?

Both....but real life takes too much time atm to do commissions

>>19906284
I found one but doesn't seem to show anus , surely there's a anus specific lora.

>>19906369

my folder was already named python_embeded for some reason and i did follow instruction, but it's simply not working for me for some reason.. it makes the sage attention folder called "sage-attention-source" and i put in the .whl file in there then click to continue but then cmd window disappear and nothing happens... so confused.

>>19906369
>>19906911

i am using comfy portable btw idk if that is making difference

>>19906911
There are a few youtube vids that are great at explaining Triton/sage install, I remember using a bat install that did it all for me with no hassles, ill try find it. <3

>>19906946

hey babe. i found a tutorial and i believe it work for me now.. i appreciates it lot. thank you for helping me

>>19907108
<3 you're welcome boo

huh weird.. i feel like sage attention is barely making it go faster but maybe i am impatient?

>>19907165
there might be little toggles inside your workflow for sage attn, make sure they are sen to auto. otherwise yeah , let it cook <3

>>19907165
do tell us because I still haven't moved to portable yet. I don't get why there are so few folders for models compared to the desktop build.
It makes me feel limited. But I guess you could use portable for just vid gens.
A lot of desktop seems broken to me.

my gen time varias wildy from lesst than 5 to an hour,i dont really get why,even with the same loras/settings

Hi guys, total noob here.
I just installed comfyui on runpod, everything went smoothly.
How can I setup a workflow correctly?
I tried loading the one in the second post, but how can I load all the missing models correctly?

>>19906459

What does this mean?
KSamplerAdvanced
Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 32, 21, 40, 72] to have 36 channels, but got 32 channels instead

how to make video graphics from pisswizard smooth like this?

>>19899831

>>19907184

i will do a comparison later today. thanks babe

>>19907227


this is first time im using comfy outside of pinokio lol. not sure

>>19907615
I think an anon had a similar error in the pisswizard thread, but I forget the solution.

I think his Comfy was not up to date? Run the updater?

>>19907464
Did you install ComfyUI Manager? It really should be part of the program, not an addon.
https://github.com/Comfy-Org/ComfyUI-Manager

>>19907227
I have been on portable for a long time now and forget why anyone wouldn't run portable.

Is it a size thing? The portable selfinstallsa bunch of stuff that could be set up system wide?

I actually have two portables, one for image gen and one for vid gen, jist to keep the chocolate out of my peanut butter.

>>19907462
...I have speed variations when switching from image to vid, because I run them on sperate ports and have them open simultaneously. I might stop doing that, because, ANECDOTALY, after loading an SDXL model on image port, vid port runs at 800secs an ini. After running an SD1.5 model, the first gen takes time, but by thend, it is down to 60 and subsequent gens are speedy.

Might be the same for SDXL, but when I see 800, I jist reboot both servers.

damn i keep getting these frame 1 face changes,i have only the dick and blowjob loras lol any pointer?

>>19908022
is that erect penis one? it does that on higher strengths....so lower the lora powers some


if you are on 2.1 workflow, you could lower the denoise from 1-0 --> 0.97, it helps on some cases

I'm pretty ignorant, what are the PC requirements for this?

How can I adjust the vid length? Seems locked? Sorry I'm new to comfy

>>19908192
8gb rtx card and at least 16gb ram should get you going....but barely

>>19908206
you are missing some custom node....

>>19908211
Checked what custom nodes I was missing but doesn't say I am? Any other places to check?

>>19907682

please someone answer

>>19908255
not home atm so i cant check what node adds that...i get back to you later

>>19908283
Think I figured it out, it's using mxslider but may have been conflicting with mixlabs that I had installed

>>19908281
interpolation....aka add frames between and increase the framerate....default wan in only 16

OK....we are closing with bump limit again. Thx anon who made a sequel to original thread.

Don't make a new one! I will make new one tomorrow. I made my own comfy portable pack, that has all needed shit for 2.2 workflow pre-installed, including new sage 2.2 installation....Need to write somekind of install tutorial since pisswizz tutorial went to archives within 48hours =)........THATS WHY YOU SAY NO REQUESTS IN THIS THREAD ;). Stay tuned

>>19908309

why are you so angry at pisswizard

>>19908564
? i'm not angry at him =D

>>19908761

lol <3

>>19908309
Ok, Bag. Thanks for creating this workflow. I'll be waiting for your new thread!