File: 157190739813388182_00001.mp4 (1.33 MB, 768x768)

1.33 MB MP4

>Already got old install?
----Backup (cut) models and user folders somewhere (pref same drive's root where your comfy is)
----del old ComfyUI_windows_portable folder
----paste folders back into your fresh comfyui portable

>What do i need?
-Windows, RTX card, pref at least 8GB of VRAM, 32GB of RAM
-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
>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)
vol1 https://archived.moe/r/thread/19910536/
vol2 https://archived.moe/r/thread/19938024/
vol3 https://archived.moe/r/thread/19987607/
vol4 https://archived.moe/r/thread/20011173/
vol5 https://archived.moe/r/thread/20035627/
vol6 https://archived.moe/r/thread/20053516/
vol7 https://archived.moe/r/thread/20096113/

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

Thanks!

What's new from vol 7 ? Just updating ComfyUI or is there anything new besides that?

>>20137764
not much....just basic update (aside i downgraded torch back to 2.8 to get stuff working) Most changes to the batch file i made, it installs the nodes rather than i would pack them in. Also added options to download all needed files to play with sdxl and qwen workflows. If you got working setup now, no need to go for this

Still works like a charm. I still have issues finding useful loras to apply. They rarely work as advertised and many are t2v only and spit out horrible result. Some do seem like magic though. If you could post some examples of good ones that would be perfect!

man.. after updating comfyui, the workflow is so slow.. i shouldnt have updated it fuck. takes 5 minutes to gen instead of usual 2 minutes

>>20138393
> vol7 https://archived.moe/r/thread/20096113/
just get the old version if you have issues.

I have yet to try it out, but I do like the idea of playing with Qwen and SDXL. Thanks you again 1st, you're a real one for doing all this.

On another note, I have a request/question for other genners. Ive gone through the archive and main site for civ, but are there any other sites that do NSFW loras for wan? I am specifically looking for:
-hypnosis posing or being hypnotized
- bend over and spread ass to camera
- a decent spit roast Lora (i see vids floating on the board but I can't find it on civ
- horse/ dog depravity (!)
- humiliation/ body writing (anything in this genre really)
- absurdly, unrealistically large cocks (see horse above)

Ive had some fun with 4n4lpl4y by K3nk, but its hard to get it to be consistent sometimes. Anyways, just wanted to ask if anyone has any ideas for things that might work, or if they have recs for their fav loras. Ultimately, I think they just need to be trained and it hasn't happened yet (and im too broke to commission and too stupid to do it myself).

>>20138801
Also requesting

- bimboification /slutification
- lactation
- egg laying
- dollifying (like literally becoming a sex toy
- onahole /amputee (but not in a gore way)
- a decent rim job lora (pov or otherwise)
- mind control headset (kinda in the hypno vein still)

sorry for the novel, being a new genner has been like crazy hot and now I want to do it all

>>20138801
>>20138888

Pretty specific fetishes you got on your wishlist. An maybe this is the reason why there aren't loras out. You could commission some1 to make you one, but i don't think they are too keen trying to gather good quality dataset for amputee porn...

I saw a request fulfilled that used a LORA/promor that smashcut from a normal picture directly to a POV angle mid-blowjob

Any idea what it was? What Lora and prompt?

,@bag thanks for the update! have a question regarding 1.4 wf. how do i enable the reactor node also whats its use? Thanks!

File: Screen-Shot-2024-11-06-at(...).png (2.64 MB, 1900x1898)

2.64 MB PNG

>>20138888
>- egg laying
>- dollifying (like literally becoming a sex toy

>>20139101
>pretty specific fetishes

Yeah thats basically the issue, but I feel like once Lora makers do all they can with commonly requested stuff, eventually they will move closer to fetish content. Meh, vanilla sex is still hot, this was just icing on the cake

>>20139101
>Pretty specific fetishes you got on your wishlist.
im not as extreme a case but i notice this too when playing around with qwen. its decent at making 20 year old korean women but struggles with women in her late 40s wearing pantyhose

>>20139313
maybe one of the blink series:
https://civitai.com/user/iGoonHard
or just some other POV bj lora with prompt "scen suddenly cut to pov blowjob" or something like that

>>20139402
You can right click the group and set it to be on always. Honestly can remember what i tried there, maybe to do a face swap on next input image before run, but most likely that didn't work so well (that's why those nodes are bypassed ;) )

>>20138762

the thing is i am using the older workflow

File: asdfwe.png (675 KB, 3002x1821)

675 KB PNG

Having issues here on first time install, trying to use the default setup, 1stAldbag_WAN2.2 v1.5d

Im on Windows 11 64
NVIDIA GeForce RTX 3070

>>20140353
try with sage node disabled

>>20140353
oh, nvm.....you have qwen loras loaded on wan workflow.....load the wan lightning ones

>>20140353
hmmm wan 2.2 model + qwen lora? Maybe that causes the crash?

>>20140395
Oh sorry, yeah after running into the error I tried a full reinstall, so those got changed to gwen, but the first time I ran into this error I was running the wan ones. The error I got was the same though. After disabling the sage node it does seem to be working now, or at least has progressed past that stage.

>>20140486
that kinda means you you didn't install triton/sage/they didnt install....install vc_redist incl in the pack, then install sage/triton via bat

>>20140501
Thank you, I've tried uninstalling/reinstalling vc_redist and sage/triton but I still get the same error, but it seems to be working without sage.

>>20140725
>but it seems to be working without sage

it does, its tad slower without it, but good you got it running

Up

>>20140725
This is the easiest way to install triton and sage, just change the directory from python embedded to what the comfy app directory says exactly it's something slightly different https://www.patreon.com/posts/easy-guide-sage-124253103

Bumps

1st if you have time could you explain the interpolation node? I've seen that most nodes I'm not really supposed to mess with are pinned and toggled, but this one isn't. I tested with interpolation 6 and it just took longer to process and with 1 it just crashes cause out of range.

So is there any time I'd want to change it from 2?

>>20142625
Interpolation in simple terms; how many frames to generate between original frames. it estimates how transition between orig.frameX and orig.frameY should look like. it makes videos look smoother when using higher fps

example interpolation "2" make 2 new frames between original frames, aka doubles the frames you have, you can then set fps to 32 rather than wan's default 16.

if you wanna wanna go for 60fps, set interpolation to "4"

if you want slightly slowmotion in your video, set interpolation to 3 and fps to 32

setting it to 6 is tad overkill.....

>>20142672
thank you!

>>20137696
Got a 5070 coming tomorrow
currently using a 1080 ti.
Excited to work on faking myself into all sorts of scenes <3
Thanks for posting in advance!
Im sure youll see me posting my retarded questions as I figure this all out! <3

What's the different between the sageattention bat and the sageattentionFP16 bat?

Bump

Some observations for those who wants to run this on a potato like me (literally using half of the minimum needed specs on a 2080)

1. If you are not very tech savvy or just wanna goon right ahead and dont wanna find the lower models its actually generally ok, IF you have the patience. As usually my generations take 15 to 20 minutes (Do crunches or clean while generating) you will need a ton of it

2. Enable "sleep unused tabs". Don't overwork your gpu by having too many tabs open if you wanna generate it eats resource and can result to OOMs and absolutely DO NOT HAVE 3 TABS THAT USES YOUR GPU (YT, FB or IG or anything that uses gpu for video streaming)

3. When "lowering the resolution". You usually have 2 choices, resize or crop. Resize if you still want the full picture or want some leeway on the generations, crop meanwhile is better if the picture is too big anyways and you just wanna focus on the subject

e.g if you are generating your girl sucking a dick you dont wanna have it spawn too close to their face otherwise the generation either is gonna be them just playing with their mouth not sucking it or its gonna be a dick slap. If you like that no problem, if not then that's a problem.

Another note is that, and this worked for me, always have a picture with a ratio that is not 1:1 or close to it either its 9:16, 16:9 or 4:3. My theory is that because of the "output longest side" it doesn't know which side to choose, this usually result in an OOM and if not, a long ass generation like a 1 hr 30 minutes generation.

4. With the 4 steps LightX lora all you need is usually 8 or 10 steps. Duration at 6. Remember the final product is usually longer because Comfy's duration is kind of a lie, what do I mean? When you set the duration to 6 there's usually a 1 to 2 second delay before the actual thing you prompt generates, this is if you are not using blink or smash cut or whatever in the generation. This results in a 7 or 8 second duration product.

>>20143884

5. Prompt wise. Try to copy prompts present where you get the lora and just play around with it. Long ass prompts are good for T2V not I2V.

6. Lora wise, do not over stack it. Always find a lora that is already trained for the extras needed for what the lora does. For example. If you want a suck dick lora, its better if its also trained to generate the dick so that you wont need to add a new lora just for that. Mixing and matching is good IF you are going T2V or have better GPU which in this case I, and probably you, don't have.

7. Sage wise, its up to you to use it. It works sometimes it doesn't and results in an OOM cause I think because it makes things fast it overworks the GPU. BUT ABSOLUTELY USE THE BLOCKSWAP FEATURE. Also download Q5 not Q8, we low spec here after all.

8. Lastly As OP said before in one of the previous threads don't buy a 50xx just for this, have it because of other things too. A 40xx is good enough, just have good RAM and VRAM

>>20143099
Not much.... Fp16 flag gives little speed boost when using Fp16 models

>>20140725
I'm having the same issue. I think the problem is the libs and include folder are missing in the python_embeded folder. I compared my old install and it has those 2 folders. Doing some more testing

>>20144565
That could be..... I didn't clean install from the pack without the Triton libs and I got it working just fine. So I thought they were old depencies, but maybe you need them after all (I have cuda installed in the system as well maybe that's why I got it working without hickups)