>How
Download the package and extract it on some fast drive that have good amount free space (100gb at least). Then
----Run 1stAIdBag_install.bat
--------Install ComfyUI and Manager. This is kinda bare minimum you need to install. You can install workflow's missing node via manager. Some user inputs needed during this.
--------Check if your ComfyUI runs after this. If it does....proceed
--------Install Custom nodes. This will install all custom nodes needed in my workflows, also other workflow's commonly used nodes (this might take some time)
--------Install Attentions. Highly recommend this as they give speed boost for generations.
--------Download essential WAN files and one of the WAN GGUF models (Q8, Q5 or Q2) Better your rig, bigger Q to get
------------NOTE: Downloading these models take time as those are several gigs
--------OPTIONAL: Download QWEN, GIT, or SDXL stuff if wanna play with image workflows

Run comfyui. On the first run it will take some time to load as some nodes like reactor will download some models needed. Once loaded, open the latest version of the workflow, re-select the all the models, vae, clip and loras. This is something you need to do every time you open up a new workflow, so the files point into locations in your computer. See the vid: https://screenrec.com/share/mkgZDHG5uI

After that, check other settings, type in prompt and hit run.

Buy me a beer: https://ko-fi.com/1staidbag

>Already got old install?
----Backup (cut) models and user folders somewhere
----del old ComfyUI_windows_portable folder. Then do as guided above
----paste model and user folders back into your fresh ComfyUI folder after install

>What do i need?
-Windows, RTX card, pref at least 8GB of VRAM, 32GB of RAM, drivers up2date
-You can run stuff with lower rig specs but then you would need lower models, lower resolution, lower clip duration, etc
-There are some info/help to run stuff with ATI cards in prev threads
-NOTE1: Major change on the pack, nothing is pre-installed, all nodes are getting installed via 1stAIdbag.bat now. Pack is now much smaller, 7megs

>If/when you need help....
A) take a screenshot of your whole workflow (and log console window if possible)
B) state what workflow are you using (i'm not going to try fix other's workflows for you)
C) what specs on your rig

>Where to get loras?
https://civitai.com/search/models?baseModel=Wan%20Video&baseModel=Wan%20Video%202.2%20I2V-A14B&baseModel=Wan%20Video%202.2%20T2V-A14B&baseModel=Wan%20Video%2014B%20t2v&modelType=LORA&sortBy=models_v9%3AcreatedAt%3Adesc
(you need to make account and enable show adult content)
https://civarchive.com/search?type=LORA&base_model=Wan+Video+14B+i2v+480p&is_nsfw=true&sort=newest&page=2
(old 2.1 loras that got removed from civitai due to their nsfw bullshit policy)

Old thread(s)
https://archived.moe/r/search/subject/ComfyUI%20portable%20for%20newbies/

Up

Anyone got Big love klein2 and could share it? 20k buzz/10 dollars is crazy

What method is that guy using to just adding black dicks to bitch's mouths? I want a workflow that I can just insert big dicks into pictures for easier and quick video generations.

>>20421791
could sound less like an incel lol?

>>20421729
literally just wait poorfag

>>20421729

read this >>20422445 .....can't wait 10days or donate 10bucks to the maker? you have any idea how much gpu power and time those model finetunes take to make?

This seems like the best deepthroat LoRA yet:

https://civitai.com/models/2413927/g4gging-deepthroat-i2v-wan-22-video-lora-k3nk?modelVersionId=2714048

Looking for a workflow and/or lora setup that will give me the intensity, quality, and drool effect of this BJ.

I was messing around with a multiscene bj work flow and tried adding a TAZ penis Lora and deepthroat Lora but it gave me poor results with the additional loras

File: 1774713598612830.mp4 (1.4 MB)

1.4 MB MP4

Oops forgot the video

>>20423839
i think its ment to be standalone lora...thou the author recommends putting some cumshot lora on the last part.....

you might add something like "saliva dripping" or drooling as she sucks the penis" in your prompt

File: 014721_157789330585114620_00001(1).mp4 (3.3 MB)

3.3 MB MP4

>>20424060
works ok in svi workflow....just need to set the durations bit higher

File: 2026-03-30 020508.png (567.8 KB)

567.8 KB PNG

>>20424171

File: 2026-03-30 020554.png (947 KB)

947 KB PNG

>>20424171
scene prompts

>>20424060

Seen one before that has it. Think it was just prompted "lots of spit and dribble from womans mouth when sucking" or similar. Can specify "white spit and dribble" if you want a cum/pre-cum effect.

>>20424176
What workflow is this? Could I get your work flow on a site like https://catbox.moe/ so I could play around with it?

I tried the pov and played around on the prompt with the drool but it didnt change too much. Could I get your prompt with how you produced that video?

here is mine if anyone wants to use it/play with it: https://files.catbox.moe/i1okgp.json

>>20424171
oops seen the prompt. disregard that, but still would like your workflow if you dont mine.

>>20424221
https://files.catbox.moe/bzoumz.mp4

>>20424233
God bless you. You are the man

>>20422464

what does big love klein2 do?

>>20424173
Question: Do you have the link or lora you used for wan_small_breasts.safetensors?

>>20424636
https://civarchive.com/models/1411545?modelVersionId=1595716

Any tips for facial consistency? I'm using a 4070ti with Q5 right now. Not sure if I can support Q8 with 12gb vram.

I'm guessing Q8 would help, or fiddling with weights on lightx. Faces just seem to contort in weird ways or instantly switch to something else.

Workflow is the provided wan 2.2 1.5, no changes in settings. Just lightx as a lora.

>>20425232

Have them move in super slow mo with face always pointed towards the camera. Aside from that not really, any motion, face looking away from camera there is a big chance face will change.

File: ComfyUI_00836_.mp4 (1.1 MB)

1.1 MB MP4

>>20425282
>>20425232

Also using good lighting and highest res your card can handle helps.

>>20424233
Ive been playing around with your workflow. Is there a way to maintain face accuracy? After the first output looks accurate and after that each output slowly becomes a different girl.

Also do you have detailers bypassed for a reason?

I was thinking of putting a FL RIFE Frame Interpolation before the video combine to get it to 30 fps but idk if thats a bad idea

>>20425580
>Ive been playing around with your workflow. Is there a way to maintain face accuracy? After the first output looks accurate and after that each output slowly becomes a different girl.

yes, this will happen...there is not much to avoid that. Use sharp input images, output higher resolution as output, avoid using loras that tend to alter the face a lot....even some pussy loras change the faces...each extend have blurrier previous latents to work with, so quaility dropping is given. My suggestion is to make small scene, maybe one or two extentsions, then join those clips together.

>Also do you have detailers bypassed for a reason?

Yes because they don't work.....as they should, haven't find time to fix them

>I was thinking of putting a FL RIFE Frame Interpolation before the video combine to get it to 30 fps but idk if thats a bad idea

feel free but there already is a Rife node for interpolation before video combine.

>>20424173
workflow json file?

>>20426037
>https://files.catbox.moe/bzoumz.mp4

drag the video on comfy

Bump

how do i stop or change the location of the copy of the input image comfy saves when you gen?

i can change the lastframe location easily, they all go to a folder, but not input

>>20427318
I dunno.... It's not input image, but a resized copy of the input and I think it defaults itself to same folder as vhs

>>20427337
hmm is there any way to stop it from being created?

>>20427345
on that resize v2 node, i don't think so (unless you can/want to edit it's python files)

alternative is to change the resize v2 node to some other scale/resize node.

i don't know why these saved images are issue thou?

just had my computer crash, anyone know anything about this? been using 1st Aids stuff for months with no issues

Faulting application name: python.exe, version: 3.12.10150.1013, time stamp: 0x67f515a7
Faulting module name: ucrtbase.dll, version: 10.0.26100.7623, time stamp: 0x53a0792e
Exception code: 0xc0000409
Fault offset: 0x00000000000a4ace
Faulting process id: 0x94D8
Faulting application start time: 0x1DCC15337E058D6
Faulting application path: D:\rbwu4v\ComfyUI_apprentice_portable\python_embeded\python.exe
Faulting module path: C:\Windows\System32\ucrtbase.dll
Report Id: 085a5a8c-5681-4300-b47c-7c14d457a8ee
Faulting package full name:
Faulting package-relative application ID:

>>20427672
did you install some custom nodes and then that occured?

>>20427674
nope i have not added anything to your workflow. the last thing i changed was deleting everything and doing your v4 installer, but i did that the week you put it up and have had no problems since then until this

>>20427672
enjoy your virus. thats what yall get for messing with strangers python files

>>20427682
dang you're right bro, rip, bank account being drained as we speak, bonzai buddy won't leave me alone. wish i was like you and didn't have a pc capable of doing this, then this wouldn't have happened :(

>>20427679
have you updated any other drivers/win updates in your rig? that seems system side problem rather than comfyui...that specific dll version seems to cause issue in many programs

you could try reinstalling VC redist drivers (ucrtbase.dll is part of it)

https://learn.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist?view=msvc-170#latest-supported-redistributable-version

>>20427686
i have not updated in Nvidia in a while, and according to Windows, the only thing that has updated in the last week and a half is

Security Intelligence Update for Microsoft Defender Antivirus - KB2267602 (Version 1.447.105.0) - Current Channel (Broad)

so idk. i use comfy every day so maybe that update did something? but it wasn't something that required a reboot or anything, i never told it to reboot, and I very rarely turn my PC off, so I would have noticed

>>20427691
judging what i gathered from net, issue is caused by win update on jan 2026 KB5078127....

uninstalling that update might help, won't recommend thou.....try reinstalling those vc redists first

my videos are pretty nsfw (no dicks at all) even when I add specific Loras
where does it pull the nsfw checkpoint for SVI2PRO?
am i possibly using the wrong "clip_name"
nsfw_umt5_xxl_fp8_e4m3fn_scaled
?

>>20427713
meant theyre sfw

>>20427713
i have no idea what you are talking about.....svi2pro is a lora. There is no special nsfw checkpoint for it, vanilla wan 2.2 models work ok. Some goes with clips (or text_encoders) there are some nsfw variants out there that say they understand nsfw lingo bit better, but again, they are not required, basic clip files for well on nsfw on wan

and AGAIN screenshot of your workflow and setting would help a LOT

So i would assume/guess your are doing something wrong in the lora or prompting side

>>20427615
the output folder has thousands of useless images in it which i have to keep deleting.

I know this probably sounds like a stupid question but if I only wanted to do image to image edits with minimal settings to see if it could work on rtx 3050 as I have heard is possible would I still follow all steps on setup or just skip some for video creation as I’m certain my rig wouldn’t support it.

>>20430328
6 or 8gb?

see >>20419606

--------Install ComfyUI and Manager. This is kinda bare minimum you need to install. You can install workflow's missing node via manager. Some user inputs needed during this.

(Honestly, jsut do this, and install Comfy Manager and find workflows and install the nodes you need. But if you are really unsure of what is going on...)

--------Install Custom nodes. This will install all custom nodes needed in my workflows, also other workflow's commonly used nodes (this might take some time)

(You'll get some nodes you don't need, but since you don't know what you need, ehn. As you get more familiar, you can just reinstall a portable and then just the nodes you need instead.)

Haven't used his installer in awhile, so not sure if

--------OPTIONAL: Download QWEN, GIT, or SDXL stuff if wanna play with image workflows

is a real option or soemthing you do after the fact.

>>20430336
8
And I’ll give it a shot. I figured getting qwen or sdxl was a must. But honestly being so new to this I’m sure I could just get comfy and find my own Lora’s or something but as a fresh newbie it’s all strange and seems like a lot. I’ll play around. I did everything in the package before but just using the preloaded settings crashed my computer at every attempt. When you have little know how on what to change in the nodes to bring down the demand it is tough. A few times I just attempted to half everything ie seeds and all that and it did nothing and someone told me that isn’t how it works lol. Thanks for the reply though

>>20430345
Yeah, download poratable, then

https://docs.comfy.org/manager/install

Then hit up civitai and look for SDXL workflows.

I'd start with SDXL with 8gb...QWEN might be usable, but I'd prefer Klein, since, anecdoatlly, it seems less demanding than QWEN.

>>20426041
i'm trying to set up your workflow exactly as is, i can't seem to find those cum aesthetics loras anywhere by searching by filename, do you happen to remember where you got them from?

>>20427672

Pretty sure its a Comfy issue. Just got off a reddit post where everyone is bitching about the new front end Comfy pushed out that broke everybody's saved workflows.

Try rolling Comfy back to over a month ago and may fix it. Also take off updates/unplug from internet.

>>20431250

Comfy has been busy.. breaking shit !

Pulled from
https://docs.comfy.org/manager/install#non-desktop-users

Switch between new and legacy UI
The following version updates only support pip installations. Versions installed via custom nodes do not support switching to the new UI.

Non-Desktop users
To use the new UI:
python main.py --enable-manager
To use the legacy UI:
python main.py --enable-manager --enable-manager-legacy-ui

Also
Portable users
For users running the Windows Portable version, the new ComfyUI-Manager is built into ComfyUI core but needs to be enabled.

Install the manager dependencies:
.\python_embeded\python.exe -m pip install -r ComfyUI\manager_requirements.txt
Launch ComfyUI with the manager enabled:
.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --enable-manager
pause

Recommendation for anyone struggling:

Download Claude AI and start with "Looking to improve the output of this workload"

Give it a very complicated workflow and answer when it asks this question: Q: What kind of output is this workflow generating?
A: Video / animation

Q: What aspect of the output are you looking to improve? (Select all that apply)
A: Image/video quality & detail, Prompt adherence, Consistency / coherence, Speed / performance

IMPORTANT: It should make it where its okay to help you with NSFW content. Dont ever show your image reference and output. It will stop working with you

I have learned alot doing this and have some very good T2I and I2V workflows because of it.

I like helping people since alot of wizards wont. For the wizards that do, lets stay strong.

>>20431439
Also I have fixed alot of my errors that other wizards and websites said couldn't be fixed using Claude.

>>20431444
>fixed alot of my errors that other wizards and websites said couldn't be fixed

there are no such errors.....everything can be fixed. My experience with asking AI help for comfyui issues.......it tells you to break it even more.

>>20431828
Claude has done wonders for me. Was using Grok and it is how you described.

I feel my T2I is really good. I am going to work on that a little big more tonight and then switch to I2V with Claude.

Also paid for early access for Apex V8 if anyone wants to use it:


https://civitai.com/models/573152/lustify-sdxl-nsfw-checkpoint

https://www.photonfile.com/dl#dp=d3NzOi8vdXMtZXZhLTAxLmFwcC5waG90b25maWxlLmNvbT90b2tlbj1VME80ZGdYZDViaE9nUVhScDQzWDU2MzBuYTNnMWFhVjY3

any idea why i can post once on /r and then i get "ip banned for abuse" for several days ? im not posting anything bad

>>20431861
Are you using a VPN?

>>20431861

i get the same shit, too! wtf 4chan!

>>20430742
https://civitai.com/models/1869475/wan-22-anime-cumshot-aesthetics-precision-load-i2v-beta-version

>>20432150
Despite telling me that my IP wasn't banned, I was "soft banned" for 3-4 days as well. I think I tripped some spam algorithm by posting too quickly or something.

File: 1774938599786464.jpg (352.3 KB)

352.3 KB JPG

Does anyone have an inpainting workflow that's perfect for high-quality cumshots?

I suspect this is Flux, but I have no clear source for it.

File: 157164006828060300_00005.mp4 (647.5 KB)

647.5 KB MP4

My gens often get stuck at 27% FLFV ... any reason why?

File: ComfyUI_00227_.png (1.6 MB)

1.6 MB PNG

>>20435611
sure is easy to troubleshoot this problem with that lovely workflow and terminal output you put there.

>>20424171
workflow isnt working.
first errors that come up are
Failed to validate prompt for output 585:
* (prompt):
- Required input is missing: images
* VHS_VideoCombine 585:
- Required input is missing: images
Output will be ignored
Failed to validate prompt for output 583:
* (prompt):
- Required input is missing: images
* VHS_VideoCombine 583:
- Required input is missing: images
Output will be ignored
Failed to validate prompt for output 584:
* (prompt):
- Required input is missing: images
* VHS_VideoCombine 584:
- Required input is missing: images
Output will be ignored
Failed to validate prompt for output 582:
* (prompt):
- Required input is missing: images
* VHS_VideoCombine 582:
- Required input is missing: images
Output will be ignored

threw the whole error message into ai and it says
The log you provided shows that your generation is failing before it even starts because of
missing connections in your workflow.
1. The Core Error: "Required input is missing: images"
This is the reason your generation isn't starting. The nodes numbered 581, 582, 583, 584, 585, and 695 are sitting empty.


2. The LoRA Warning: "lora key not loaded"
The long list of lora key not loaded messages suggests you are trying to use a WanVideo LoRA that is incompatible with the model you have loaded.

The Cause: You are using a GGUF or quantized version of the Wan model, but the LoRA you've selected was likely trained for the full FP8/BF16 version.
The Fix: Ensure your WanLoRAStack is actually receiving a valid LoRA file. If you aren't trying to use a LoRA, bypass (Ctrl+B) or delete the LoRA stack nodes to clean up the process.

3. "motion_range is EMPTY"
The [WanImageMotion] warning indicates that your motion settings are set to 0 or are not receiving a "previous sample" to work from.

>>20435758
>"Required input is missing: images"
>This is the reason your generation isn't starting. The nodes numbered 581, 582, 583, 584, 585, and 695 are sitting empty.

so look at those nodes as numbered and figure out why they aren't populating. Are you supposed to put an image in somehwere and haven't? I the workflow unable to pull these images from somewhere?

Not sure what workflow you are attempting to use

>>20435813
ive uploaded the image to every node that allows me to upload one. still doesnt work. the workflow comes directly from the video file you posted

>>20436027
huoh.......post the image of the workflow?! did you re-select all the models/clips/vae/loras?

and don't as ai for help because i would assume you didn't post a screenshot of the workflow either

* VHS_VideoCombine 584:
- Required input is missing: images
Output will be ignored

^^ these are normal errors when you got extensions bypassed (therefore preview nodes, on the most left side, get no images) has no effect on the your gen

>>20435611
if it would stuck on 27% then you would not get any output video.....

File: wf2.png (502.5 KB)

502.5 KB PNG

>>20436043

File: wf1.png (293.8 KB)

293.8 KB PNG

>>20436481

>>20436481
>>20436485
those look fine.....when you hit run, where it hungs up and what that the console say at that moment.....

and skip those
- Required input is missing: images
* VHS_VideoCombine 582:
- Required input is missing: images
Output will be ignored

or lora key not loaded errros....

>>20436542
this is the full error in the console before comfy boots me
it'll hang at like 2% and then the error occurs
https://pastebin.com/iFVc8iR4

>>20436652
Try changing the sage attention mode to "auto" in this node >>20436485. Did you install the attentions? If no, set it to disable

>>20430349
8gb 3050 works fine w sdxl and wan 2.2. Try some of the older aid bags with a single 5 second Gen + inpainter speed node. Haven't tested svi wan 2.2 I use my 16gb 5060 for those multi Gen wf

>>20430345
Reply
Meant for this

File: Screenshot 2026-04-05 204859.png (34.8 KB)

34.8 KB PNG

>>20419604
Having some issues with your new WF Bag.
Im continuingly getting a dissconect at 2% when trying to run. Logs are showing this
" 2026-04-05 20:40:04.809] [IAMCCS_HwSupporter] profile=balanced reserved_vram_gb=1.0 sage_attention=sageattn_qk_int8_pv_fp16_cuda torch_compile=False (mode=off requested=off) warnings=0 (2ms)
[2026-04-05 20:40:04.909] [IAMCCS_WanLoRAStack] Detected Wan 2.1 format - converted lora_down/up lora_A/B
[2026-04-05 20:40:04.942] [IAMCCS_WanLoRAStackModelIO] 'Wan22_BiggestCock_high_noise_V1.safetensors' strength=0.6
[2026-04-05 20:40:04.974] [IAMCCS_WanLoRAStackModelIO] 'W22_Multiscene_BJ_i2v_HN.safetensors' strength=0.85
[2026-04-05 20:40:05.018] [Wan-Cache] Optimization disabled, model restored
[2026-04-05 20:40:05.041] switching model at step 1
[2026-04-05 20:40:05.041] Running high noise model...
[2026-04-05 20:40:05.042] [MultiGPU Runtime] Using runtime device cuda:0 (comfy.sample.sample:GGUFModelPatcher)
[2026-04-05 20:40:05.043] Requested to load WAN21
[2026-04-05 20:40:18.277] loaded partially; 3077.65 MB usable, 2717.47 MB loaded, 12107.99 MB offloaded, 358.73 MB buffer reserved, lowvram patches: 0 "
I have a 5070ti and 64gb of ram.

>>20427729
I have to restart my computer after each gen with the SVI 2.3 workflow from your video or else it black screens and crashes. What’s weird is I can run WAN 2.2 for days on other workflows with no issues, so it doesn’t seem like a hardware problem

>>20439104
I forgot to mention that I have no issues with your SVI Pro 2.1 or 2.2 workflows. So its not a prompt or LoRA issue. I use everything listed in that video you shared

>>20439106
Then I would suggest sticking with prev version. That 2.3 version was something I was messing around when anon asked a workflow of certain Gen.

Hi
My 2060 6Gb is getting really old for AI.
I can grab a 5060 Ti 16Gb for $550-600 is it worth it, or should I bite the bullet and aim for a 5070/Ti ?
I get that I should go for the max VRAM my budget allows?
I want it both to be versatile and if possible ready for the 2-3 years coming. Any advice?

File: 1752537318616396.mp4 (2.2 MB)

2.2 MB MP4

Any idea how these were made? Like Same motion but switched out characters? There where a bunch more all of the same vid but always a different girl

File: 1752535367074608.mp4 (2.2 MB)

2.2 MB MP4

>>20440355

>>20440355
>>20440363
looks like wan animate

>>20440355
>>20440363
>>20440682

just simple bj deepthroat lora....same seed and prompt usually generates pretty similar output if the inputs are similar

>>20440355
>>20440861
its VACE or wanimate, the clips are in sync

>>20440881
vace maybe....wanimate is pia todo blowjobs. but still keep my vote on same seed/prompt/input

>>20440355
>>20440363
Imade them and this
>>20440682
is the right answer

>>20440950
have to say they look pretty good....its been ages since i played with that model, back then the masking was pretty terrible --> bjs looked pretty odd

>>20441121

ngl those still look like shit. too cartoony lol

File: 1-ComfyUI_00197__x264_001.mp4 (3.7 MB)

3.7 MB MP4

>>20441153

Speak for yourself. Some of like the cartoony stuff. I mean fuck, I didn't get AI to gen crap you can watch en masse on the internet. Gimme impossible IRL stuff any day.

File: 157732750245007188_00001.mp4 (711 KB)

711 KB MP4

Thanks for the tools Mr.bag, I owe you. Hopefully you upload that loopy workflow in a future release of your rar.

noob here, i still have no idea what i'm doing
i'm currently running this >>20424233 i2v workflow and installed the client using the OP bat
what the fuck am i doing wrong? why is it fucking impossible for me to proompt any kind of sex scene? i've gotten literally every fucking wan nsfw/sex position/doggy lora on civarchive, tried running each, running them together, fucking with the strength, the triple sampler, the starting image is already a subject in the correct position, the only way i could even get a dude next to her ass is with equally difficult qwen prompts of the same image, i got all the dick loras, insertion loras, fucking anything that is even remotely related, and have spent days just bashing shit together and i only get monstrous results; it works fine for blowjobs and titfucks and general undressing, but i cannot get this fucking thing to make any kind of sex scene, am i retarded? is this how it is?

>>20442463
>am i retarded? is this how it is?

Most likely...post a screenshot of the workflow so we don't have to guess your prompting and lora list

>>20442585
the workflow is exactly like the catbox file and the screenshots from it that aidbag has already posted
i have only one extend on, triple sampling on, CFG to 1.1, duration 5 seconds, motion amplification set to 1.00 and interpolation set to 3, those are the only differences, every other value is exactly the same, rechecked a dozen times

the actual lora list isn't actually a lora list, i've tried dozens of them on their own even with the provided triggers/suggested dependencies and sample prompts that the authors specified, 99% of the time the man never appears, the penis never appears, and the subject is just undulating back and forth, or it turns into body horror, and even the body horror is clearly not listening to my prompt (e.g. the man is to the left of the subject beside her hips, but it will only draw one on the right next to her face, yes any face-related lora is removed)
again, this only happens for explicit sex prompts, if i load pov titfuck or blowjob loras they work fine for that

File: 15745001950121169_00001.mp4 (716.7 KB)

716.7 KB MP4

>>20441747

>>20442613
you must be retarded, because you still have not posted a screenshot of your workflow

>>20442613
>99% of the time the man never appears, the penis never appears, and the subject is just undulating back and forth, or it turns into body horror, and even the body horror is clearly not listening to my prompt (e.g. the man is to the left of the subject beside her hips, but it will only draw one on the right next to her face, yes any face-related lora is removed)

loras are trained from 5sec clips usually. They usually just contain the actual "action", ie, doggystyle has the movement of man moving his hips behind the woman, maybe grab her hip and/or pull her hair....

getting woman bending down, lowering her pants, man walking into the scene, inserting the penis, doing the action is a tall ask to fit in 5secs

so either make image where the nsfw is happening already (missionary, doggystyle, cowgirl, etc) and then prompt it, or use so called blink loras, where the scene instantly cuts to that specific action (these have side effect of changing the faces thou)

As for body horrors....these usually occur when lora weights are off (too low or high) OR you have two loras that attempt to generate the same thing. ie; you use some dick lora and then some general nsfw (like dr34m nsfw lora).

increasing motion amplitude to 1.15-1.2 helps, also setting lightx loras bit higher might help with.....but like like statet many times before, workflow image would help alot to see the settings/prompts and input image you are using.

To all new wiz apprentices

If you think WAN will do the gen as YOU IMAGINE..ain't going to happen. If you get frustrated easily, AI is not for you. There is a certain prompt logic to all AI models. On WAN You kinda need to be very specific, but then again not too specific. If you overflow the prompt with too much info, model has problem cypher what is important in your prompt and what is not. I've been doing wan gen for ages now, and still hitting my head to brick wall many times.

on prompting you can set weights to certain things...ie ;

"woman dances, looks at the viewer, smiles and shows her tongue. She reveals her nude tits"
to
"woman dances, looks at the viewer, smiles and shows her tongue. (She reveals her nude tits:2)"

^^
this makes wan go....ok, i got 5secs timeline, he wants all this, but seems like revealing tits is pretty important...gonna try to make that in at least)
things that are frustrating on prompting; that left/right are not objects left/right but as seen on image.....man and woman gets confused most of the times (tip: use words like girl or lady on your prompt)....on SVI prompting is even more more tricky. there is loads of trial and errors.

and for some older question from prev thread; you don't have to prompt in english (prompting in chinese is the best case scenario as the model is chinese) but UMT5 clips (aka text-encoders) support almost all languages. (note that lora key-words/phrases should be in english)

>>20442463

Dude, just go to Civi, find a vid similar to what you are trying to do and copy their prompt / loras. It will at least get you in the right ballpark.

>>20444120

Also if you are trying to field a request off the /R board 90% of the photos posted are closeup face shots so you may need an Igoon/Blink/Smashcut lora or do a pan out shot.

>>20443312
yeah i'm not even thinking about that anymore because this piece of shit has just stopped working entirely

i originally had default qwen installed, and last evening i installed nunchaku and z-image to try them out, ever since then it can't do wan video renders anymore, i'm only mentioning that i installed these because i did not touch the thing otherwise; i was rendering a 1280 max length video previously, which took me ~30 minutes to complete, and then it just stopped working
i reloaded the same workflow and input, put it on overnight and it was at 17% when i woke up after 7 hours, i didn't fucking touch anything, i didn't update anything, what the fuck happened? persists after reboots obviously
i put it back to 640 and it's at 0/7% after 15 minutes on a render it should have finished in 200 seconds, what the fuck is happening?

>>20445049

Probably a Comfy update. Comfy will push shit you don't want that breaks shit whether you tell it to or not. Once you get it working how you want KEEP IT OFF THE INTERNET.

If you ever watch the prompt when you start Comfy it will reach out to the server and start installing updates.EVERY TIME you start comfy.

>>20445065

This is especially true with custom scripts/nodes. Less so with the default Comfy templates but still happens.

>>20445065
so what can i do now? shift delete and reinstall everything?

>>20445085

Thats outside my wheelhouse. Might be some Comfy nerds that can walk you through it.. in the meantime look through your startup script see if there are any issues/conflicts and put it in google/ChatGPT or Claude (haven't tried Claude but people keep raving about the dude)

>>20445085
Back up your model folder somewhere before you do. I do a fresh install every few months.

>>20442463
>>20442613
pretty easy to get sex from dr3ml4y alone, even with a standard pic with no nudity or anyone else in the scene, start with the example prompts and tweak.

>>20445049
>>20445085

Comfyui has been pushing out some pretty terrible updates recently. most of the big issues are ironed out now

this is why it seems that some people have random issues and why some people don't. it's whether you downloaded on one of the shit updates days.

I'm guessing you either, updated your GPU Drivers and they hate you, something failed to install and you didn't notice, you hit Update all in Comfy Manager, or you're running the wrong BAT file to launch comfyui, and it's falling back into CPU mode

My recommendation

Since Comfyui pack is running the portable version, just redownoad the portable version again, go into .comfyui\models\ and copy everything over from the old install to the new install, and then run the BAT file that 1st Aid Bag has included. that way you're installing fresh, and not downloading 100 gigs of models again.

As stated above
>>Already got old install?
>----Backup (cut) models and user folders somewhere
>----del old ComfyUI_windows_portable folder. Then do as guided above
>----paste model and user folders back into your fresh ComfyUI folder after install

Comfyui is not without its quirks. there are "easier" tools to generate content but nowhere near as flexible. keep in mind this is for "Newbies" not "Beginners", there's a huge difference, you're still expected to have some knowledge or know how to google or throw it into chatgpt or some shit

If you have a log file, i can take a look at it and see what's up

>>20446437

Even downloading from the source isn't reliable anymore. I downloaded from Comfy.org last weekend and installed the default Wan 2.2 that were friggin broken out of the box (quick dive into reddit found one of the nodes was not connected).

S M H @ Comfy lol

>>20446665
Yeah, they've been try to push subgraphs pretty hard and started converting all their workflows into half working/broken subgraphs.

Kind of a shame because I like to use groups and subgraphs all the time.

>>20446669

Was reading through the prompt before I found the reddit thread and saw a line that was something about Zuckerberg being a visionary or some shit. Not sure if that is new but was weird for sure.

Bumps

Would it be possible to make a workflow implementing FFGO? or even implementing it onto the 1.5d workflow? Im way too retarded to figure it out myself and it'd be greatly appreciated.

https://github.com/zli12321/FFGO-Video-Customization

>>20447720
looks pretty ugly and pretty major faceshifting on their examples.....and also loads of pre-work. You would need to make huge white canvas image and then look up some transparent pngs to paste on it, then prompt where everything is....just for wan to make a first frame from it. Why would you need this? Is there some hidden benefit i don't see?

(i mean if that would give you option to add certain objects in certain points at your gen....aka they would work as reference objects....they claim that they do, but looking at their samples, 1/5 on that)

i would use qwen or klein for to make 1st image featuring different objects and and backgrounds....might be slower, but 10x better quality than their examples.

Quick question. I'm using the built in wan2.2 template in comfyui with a few added loras. The videos I generate often seem to tend to "bounce" back towards the original image creating a sort of looping effect. There is no end frame set and it sometimes doesn't happen so not sure what to do about it

>>20447799
your gen duration is set too high...default is 81 frames (5secs)

>>20447783
I heard it would help with better face coheerence if using a second input image but im super nooby when it comes to comfyui, i pretty much just use your workflows lol

I am so confused I did the first steps..

Downloaded the file
Extracted it
Ran Install
Says its installed

But i have no clue where to open Comfy

I am so lost.....lol....I thought I was computer savy.

Thank you!!!!!!!!!!!!!!

File: ice_spice_looks_bigger_than_i_remember.jpg (105.6 KB)

105.6 KB JPG

i know its ai but is this inpainting or some sort of text prompt? the skin and the clothing is too perfect and consistent to the original to be inpainting and the quality is something i've yet to see from any workflows. and what checkpoint is even being used? is this flux or zimage, i cant seem to figure it out.

>>20449651
maybe qwen of flux...load og image as input 1 then (maybe the outfit or fabric as input 2) prompt it to be semi-transparent. stretched on the ass

Who has a link to any of iGoon’s new blink loras, the ones behind paywall?

what's the best image to image model for nsfw content? any standouts?

File: Video_00201.mp4 (1.3 MB)

1.3 MB MP4

up with ye

wanted to try running this image to image workflow - https://civitai.com/articles/24793/lightweight-image-to-image-workflow-for-z-image-turbo-no-controlnet-wip.

got clip error, turns out need to update comfy -
https://www.reddit.com/r/StableDiffusion/comments/1p9budf/getting_error_when_trying_to_run_zimage_error_in/

cant update comfy:

ComfyUI-Manager] Failed to checkout 'master' branch. repo_path=E:\User\Downloads\ComfyUI_Apprentice_portable_0360\ComfyUI_windows_portable\ComfyUI Available branches:

used chatgpt to fix this error, comfy updates, then i cant load comfy at all:

Traceback (most recent call last): File "E:\User\Downloads\ComfyUI_Apprentice_portable_0360\ComfyUI_windows_portable\ComfyUI\main.py", line 32, in <module> import comfy_aimdo.control ModuleNotFoundError: No module named 'comfy_aimdo

Should i just do a fresh install?

>>20451045
seem to have fixed this by doing

.\python_embeded\python.exe -m pip install comfy-aimdo

>>20451048
welp, doing this and everything else has broken the cliploader. looks like its a known issue but updating comfy doesn't seem to fix it...the ride never eds! good thing i kept a backup installation

CLIPLoaderMultiGPU 'NoneType' object has no attribute 'Params'

>>20451083
Instead of using Manager to update, try to use the update.bat in the update folder.

it will likely break reactor and anything that uses insiteface like ipadapter or pulid, but you can downgrade NumPy to fix that

>>20451090
this fixed it, thanks

Sent you some beers!

File: ice spice.jpg (90.6 KB)

90.6 KB JPG

>>20450591
yeah but this is the original. like its still keeping the context of the pose, background, etc. this is like some grok shit but on steroids

>>20451627
well that seems like just prompting job. On klein or qwen

"Make woman's ass bigger, her hand on her has. Keep rest of the image unchanged"

maybe add some lora like
https://civitai.com/models/1888846/big-asships-narrow-waist-or-flux-klein-qwen-image

to give some flair

>>20451893
any workflows for this in comfyui?

File: 6876ff.png (1013.9 KB)

1013.9 KB PNG

>>20451893
got home and tested....it werks, but maybe for this extreme >>20449651 ....might need some other big ass booty lora

>>20452224
sick, is the workflow in the image?

>>20452296
4chan erase metadata. So cant embed workflows in the image.

Look civitai for flux Klein workflows, my version is very barebone. As I don't play with images that much. There are workflows for inpainting as well if you want better control

>>20453233
okay thanks ill try, you could share the metadata through mediaflare file share if you'd like. i tried recreating it from the image but i dont understand how to make subgraphs

Hey 1st. Any chance Kofi beers can get a request?

File: Assjob Grind 04121741.mp4 (1.2 MB)

1.2 MB MP4

know of any assjob lora?? searched civitai but couldnt find one

File: 87654cccc.png (915.4 KB)

915.4 KB PNG

>>20453310
https://files.catbox.moe/xb2qsh.png

wf embedded ^^

>>20453516
no...kofi won't allow. its only for workflows and other shit i do....commissions some other way

>>20453889
if that was i2v...civit dels them in minutes

>>20454233

stupid question but how to you change the output width and length?

>>20454313
you can't....image1 determine the aspect ratio. That megapixel slider controls the size...higher == more height and width == more time to gen

surely you could make custom empty latent canvas with custom dimension....but like said, my version is very barebone, that wf works for my needs

>>20454233
thanks man

I am once again asking: anyone have all the new paywalled iGoon blink Loras?

>>20454487
"Get a job, you fucking slob" is all he replies

Anyone who has a shit computer and tried to run this through something like runpod? Im wondering how hard that is.

>>20455374
It's not that hard if you have basic computer knowledge, how consoles work etc

But my suggestion is that get cloud service that only fees your the actual GPU time (aka the time it is making your gen)... Model sizes are getting bigger and bigger and how most Ai service sites work, they load the models off from their networks rather than fast local ssd.

Example ltx2 model is over 40gb, clip file over 20gb....loading those into your cloud via network takes much more time than the actual Gen

>>20454233
real quick whats the pawwg lora that you're using here?

>>20455710
should be this

https://civitai.com/models/1888846/big-asships-narrow-waist-or-flux-klein-qwen-image

.....dunno why i had wan lora loaded in there

Bumps

>>20455729
1st.AId.bag! How's it going?
First off, you're a fucking wizard man. Like, that gets thrown around a lot on these parts but you are the real deal. Not only because of your knowledge but because you're willing to teach. What you're doing is fucking commendable. Massive respect.
Now, you likely don't remember me, but I was, a couple of months ago, really trying to make Comfy work with a Radeon. And I did, but it was a pain and because of diverse reasons I had to bail. Burned everything so to speak.

Well, Comfy has support for Radeon now, so I'm back here.
I'm interested specially on faces. My card (12GB) is not beefy enough being an AMD, so I'm interested in images atm.
Could you help me how to put a face from a picture into an existing image? or generate an image in order to put the face?
I remember there were some workflows someone posted that I was using, but I lost them all.
Thanks man

File: beforenafter_00002_.png (2 MB)

2 MB PNG

>>20455729
swapped to using qwen instead of flux 2 klein since there's no loras that go as exaggerated as I'd like. any idea how to get rid of the degradation/artifacts? im on a 6gb card using qwen 2511 q4.gguf with a 4step lightning lora.

>>20457881
20457881
>uding qwen with 6gb
Even with the 4 step lofa, how long is that taling you peer gen?

Also, ref your question, is your output resolution matched to your input resolution? Or is it scaling, and if so, scaling in a logical manner?

>>20457868
Try that Klein workflow posted above. Klein does ok face swaps..... Or if just plain and simple face swap, then use the simplereactor wf in the pack

>>20458312
its like 17mins per gen, the output i have no idea. the wf im using is doing some scaling stuff but the output isnt matching the input resolution.

>>20458312
its like 17mins per gen, the output i have no idea. the wf im using is doing some scaling stuff but the output isnt matching the input resolution. here's the wf https://files.catbox.moe/ga0ii8.png

File: error.png (26.8 KB)

26.8 KB PNG

>>20458920
>pic
>that's unfortunate

Yeah, you might want to locate what's doing the sizing that workflow (and post a non errorneous image/link, sometimes catbox does this).

I know I like speed over quality, and got a klein workflow, but when I'd drop in some higher quality images, it would take forever to gen, and by throwing in a node that limited resolution to max 1024x1024, I got quicker gens, and if I liked the output, I could just up the resolution when I had something worth upscaling.

Sometimes when the output is a different resolution than the input, or scaled non-linearly, it can have a negative effect overall.

>>20454233
hey are you releasing a new version of your video creating workflow for the new ComfyUI?

>>20459055
what new? old ones work ok far as i know?

is there a lora for a slow methodical blowjob? usually the loras i get are so super fast sucking

what happened with big love klein2? did the author make it not downloadable?

any idea why my rgthree power lora loader is broken? all of a sudden i cant add or see any loras, reinstalling node does not work

>>20459425
sensual blowjob lora

>>20459428
went behind paywall

>>20459443
is the node red-border? if so, chack the manager, filter import failed, click it for more info

>>20459341
not with 0.18.2

>>20459445
not red border, and reinstalling it has not fixed it for some reason. tried older versions too

>>20459448
oh a new update....what got broken?

>>20459451
well.....do you have loras in your /model/loras/ folder

it seems stupid question, but maybe you did fresh install and forgot to add the additional path yaml?

I get back once 4chan is fixed.... No posting is like dice roll. Can't post images or videos, even text replies take minutes

>>20459452
Can't load SageAttention

hey ald. can you help me understand your qwen workflows? i see qwen edit plus and qwen 2. my goal is to take a picture of a woman, prompt something like "she lifts up her shirt and shows off her breasts while smiling", but ive been getting pretty plastic looking skin and inconsistent faces. wondering if im missing something or using the workflow wrong

Bump

hey bag did you see that civitai has a porn alternative now called civitai.red ?

Would it be possible to make a workflow with mmaudio?

>>20460782
you could try this one

https://civitai.com/models/1847730/smooth-workflow-wan-22-i2vt2vfirst2last-framemmaudio

File: Untitled3.jpg (659.9 KB)

659.9 KB JPG

>>20459549
ANECDOTALLY, using klein, the more you request nsfw the more it gets plasticky, unless you use some loras to offset its natural aversion to nsfw.

Unless qwen has changed, it also had a natural aversion to nudity that was hard to overcome and when you did, it lowered the quality.

The suggestion would be to find qwen models that support nsfw or adequate loras that can make up for it.

Also, prompting is way easier in qwen, but you still need to know what you're asking for.

Going back to qwen, I find that the bigger the prompt, the more it strays from the orignal photo, but the more realistic it is, so it is finding a compromise between the two.

Prompt fora attached image is is:

"make image1 into a full body image of a woman standing in a bedroom lifting her white shirt to flash her naked breasts and nipples wearing pink short shorts, don't change her face"