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

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

Once all downloaded. Launch Comfyui, RESELECT ALL MODELS, to make sure
they are all there in your computer. Load image, type something in the prompt, run
If sage related error = you didn't install it correctly/at all --> install sage, or disable the nodes

>>19937473
>difference is pretty subtle, true.....main benefit is now i can get rid of one variable that could fuck up your gen =)

You can't get rid of the user, so it must be something else.

I'm sure I speak behalf of mouths of many; thanks bag! I don't know why other wizards in here won't help out newbs like you do. Question on those long videos and last frames. Is there a way to select better last frame. Aka if the last frame is blurry you could go back few frames?

1st Aid is a legend for sure.

I am having issues retaining facial details in photos that have the whole woman in frame. Any idea on what I can do to maintain facial details/structure to the original person?

>>19938473
I actually attempted to mess with settings myself and found better face retention with higher total steps

>>19938327
>>19938473
thx for kind words

For facial features changing....that will always happen as wan model don't have have 4 billion woman's faces baked in....it will pick up most look-a-like and go with it

loras change faces too, for the reasons ^^ ....to minimize this is to lower lora strengths and denoise (v1.13 wf)

>>19938317
i don't get what is your point, sorry?

>>19938615
>i don't get what is your point, sorry?
I was making a joke.

>.....main benefit is now i can get rid of one variable that could fuck up your gen =)

What fucks my gens is me, the user, but you can't change that, so it must be something else.

(you know a joke is funny if you havr to explain it)

>>19938327
>I don't know why other wizards in here won't help out newbs like you do.
I always have an initial reluctance, because
-I worked hard to figure this out, get gub like me
-if everyone can do what I do I'm not special

But at the end of the day I will share if asked because 3/4 of my learning came from playing with a workflow someone shared, so pass it on, and nothing I have figured out is actually special, someone else has already probably created a better version.

I kind of hope people point out ways it can be improved, but that hadn't happened yet.

File: 24d4afdde7f2ec3885db73a32(...).jpg (96 KB, 616x861)

96 KB JPG

>>19938641
was the hit-point in that joke word "GENES"?

otherwise i dont get the joke

File: a5llgo.jpg (71 KB, 500x625)

71 KB JPG

>>19938656

>>19938649
>-I worked hard to figure this out, get gub like me
>-if everyone can do what I do I'm not special

well do something "extra" or different from everyone else.....AI tools are same to all users, yet my, pisswizz, dg, rndm, bbcshop, sircummington has own flavour into their gens. Holding knowhwvw0ledge just to be a special snowflake is just......dumb

File: asdf.png (558 KB, 1278x680)

558 KB PNG

>>19938692
I was just mentioning some feelings for context. The rest of the post explains that I do share workflows, and hope someday somene will help me improve (as you have in certain contexts)

I have my own thing indeed, I just don't trip for personal reasons.

How easy is to switch Sage Attention 1.0.6 to newer version 2.2?

I've read somewhere that you need to use the command "pip uninstall sageattention" before and then do the process. Also, do you need to change the torch too or doesn't matter the version/etc?

>>19938024
is there a way to do it online?

>>19938024

where do i get split sigmas??

File: 157448169588835667_00001.mp4 (1000 KB, 448x576)

1000 KB MP4

fucking hottttt

thanks bag

>>19939398
there is installer in the pack...its for portable version. it updates torch, only thing you need to do manually is to download correct wheel file. For fresh install it's pretty easy job

>>19939434
look up runpod or mimicpc for example

>>19939597
you don't...the node does that automatically based on amount of steps and frames

File: A_succulent_Australian_De(...).jpg (23 KB, 387x258)

23 KB JPG

>>19939602
My gens are not that smooth, in relation to speed of the BJ/Deepthroat. What settings to I need to change to get a better fps/fast motion?
Interpolation? (current setting: 2)
frame_rate in the Video Combine? (current setting: 25)

>>19938024
Would you recommend 18gb vram(5070s) just for genning an undressing vid? will it be enough for high quality one?

>>19939669
more than enough.....i have 4070 super, 12gb card...and it does the job in reasonable time. 1sec=1min gen time

if updating your rig, get some ram also. Sticks are pretty cheap, 64gb for 150$ will also speed thing up

>>19939652
>frame_rate in the Video Combine? (current setting: 25)
I think this is the answer but it's dependent on how much action the lora in question is giving.
I personally run at least 32fps, for no reason in particular really. It seems ok to me. Your gen time will be slightly longer but if your prompt is working it's time to go all in.

>>19938024

If anyone has time for stupid questions:

If I have multiple 3070 TI, would that help on gen time?
If I am generating something nude, do I get to specifiy things like sizes, amount of hair, specifics?
Does somethign like this >>19939602
use a “LORA” style template? Are there other generic “Poses”? Or does Comfy use a prompt? Both?

Thanks and apologies for my limited understanding.

Thanks Bag! Got a few questions for you:

Why is the video combine framerate set to 24, Shouldn't it be 32? (16FPS with 2x interpolation)
Do you find keeping the lightning LORA's at 1 strength to work best?
Which schedulers/samplers do you like most for the Sigma split configuration?

BAAAAG, MY DICKS ARE COMING OUT LIKE BOLOGNA AGAIN. haha erect penis or dicks epoch pick one and they're both coming out like trash.
High sampler is .10 higher than the low sampler. I swear that was working fine last week.

>>19939935

1) not an expert on daisy chaining GPUs, but internet says no, because thay card does not support NVLink or SLI. Not sure if there are other ways or wlrkarounds.

https://www.reddit.com/r/nvidia/comments/1gv4u4z/

dual_graphic_card_3070_ti_8gb_and_4070_super_12gb/

2) Yes, your promlt helps build the image. When writingv a prompt "woman lifts her shirt to reveal her small/medium/large breast, it will influence tthe output.

3) Comfyui uses both. A prompt is a written text indicating what happens. A lora is an addon selection that helps the AI understand or execute a prompt. For example, the model (dictionary for lack of a better synonym) the AI uses might not understand how to make a good blowjob, so by adding a deepthroat lora, a specialized additonal dictionary, it can create a better blowjob.

File: 1757451782129377.jpg (3.91 MB, 3192x4829)

3.91 MB JPG

>>19939935
>If I am generating something nude, do I get to specifiy things like sizes, amount of hair, specifics?

>>19939631
>19939631
any way to downgrade torch? it says my torch ist a higher version than sage attention is made for

>>19939631

>you don't...the node does that automatically based on amount of steps and frames

but it tells me to install missing nodes.. do i just ignore that? the install missing nodes link is grayed out as well.

says im missing wanblockswap and the split sigmas

>>19940425
Go to manager and install them from there

>>19939968
I think I accidentally left it 25. I mainly use dmp++/sgm uni or euler/normal

>>19940481

ok thanks babe

>>19939652

this was my prompt and pretty much everything was default from bag's workflow.

the girl in the frame remains stationary in place. A cinematic video of a beautiful naked woman, she puts her hands down on her sides exposing her naked breasts. A naked man approaches the woman from the right side. his penis is large and erect. Woman looks up at him with shocked expression. He grabs her by the head with one hand and pulls it closer, then he thrusts his penis into her mouth in one clean move. his penis goes inside her mouth smoothly. He repeatedly thrusts back and forth with his hips, he is making wide movements, penetrating her mouth down to the base and smashing his hips against her nose. When her face smashes the man's groin the extended tongue of the woman wraps under the man's testicles. Woman's breasts jiggle with every movement. The video ends with the man pulling the woman's head against his hips, pushing his penis all the way inside the woman's mouth. Fast and intense motion. realistic jiggle physics on her body. The girl keep looking at the viewer as she keeps her eye's open. eye contact. Realistic animation.

>>19938024
hey bag question, i'm using your 1.1 workflow but how can i chain gens with other loras to extend the video length? seen others do it with tit reveal to bj loras for example

>>19941309
concatenate multiple generations using the last frame as the first frame for each subsequent video

>>19941321
makes sense ty!

also how detailed should your prompts be? for cumshots i've kept pretty simple but do i need to do more for combining gens?

>>19938024

what cumshot lora was this??

>>19937209

>>19941408
https://civitai.com/models/1922973/f4c3spl4sh-cumshot-i2v-wan-22-video-lora-k3nk

and

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

>>19941415

thanks bag

also

i dont find dpm++ 2m as an option . am i missing it

>>19941641
dpmpp_2m under the sampler

the p stands for +

sde and gpu have differences I forget, but minute. Read up about samplers and schedulers if you actually care.

bumping

>>19941321

I trued that before, but I always get some difference in the color temperature of the new gen.
Any tips?

>>19942425
Try different sampler. I notice Euler tends to add saturation

What’s our best Facefuck LORA for wan 2.2? Also any favorite 2.2 LORAs I should get?

>>19942428
Which one do you use for your multi-gen setup?

File: Which_Lora.mp4 (3.72 MB, 832x1056)

3.72 MB MP4

Anyone has a clue of which kind of lora is this one?

>>19942957

https://civitai.com/models/1923528/sex-fov-slider-wan-22

>>19942428
question - so i take the last frame and then use that to prompt for the second lora? so titty drop lora, take last frame of that, and use that to start second part with blowjob? what would be the best way to combine the two separate gens from there?

>>19943172
okay ty!!! this helps explain it better.

File: 459910155436434_00001.mp4 (593 KB, 480x672)

593 KB MP4

>>19943166
Yes. Exactly that. Gen 1 video, (Breast Expansion)...

File: 183390614030092_00003.mp4 (660 KB, 480x672)

660 KB MP4

>>19943169 (You) (Dead)
...so after the breast expansion lora is done, take the last frame, then run a cumshot video .

For joiner the files, you can use:
-Any online video joiner
-Any free video joiner program
-Certain ComfyUI nodes
https://www.reddit.com/r/comfyui/comments/1gezz1g/is_there_a_way_to_join_videos_in_comfyui/

https://www.runcomfy.com/comfyui-nodes/ComfyUI-Mana-Nodes/Combine-Video


>>19943174
helps if I upload the right videos in the right order

>>19943177
ty! where can you get that breast expansion lora btw? ones i found didn't seem to be for wan

>>19943184
I went back to 2.1 because it had better gen times for me.

Pretty sure I got it from here,
https://huggingface.co/ApacheOne/WAN_loRAs/tree/main

If not there,
https://civitaiarchive.com/?is_nsfw=true

and if not there
https://civitai.com/

>>19943177
huhuhu. man I love this thread.

>>19942959
Thaanks anon!

Can anyone upload or link a good workflow for Wan2.2 i2v fp16 without any optimizations? I don't mind the extra generation time in exchange for quality, and can't seem to figure out how to change bags workflow to work with Wan2.2-VAE.

>>19944420
Turn off all the optimizations in the box to the lower left and don't use the lightning lora?
Use single sampler instead of dual sampler?