can you do animatediff with 8gig of vram ?

>>27097286
>>27097454
>>27097468
Jesus these are some grotesque face filters. You have to have serious brain rot to find these women attractive.

File: upscale-4x-RIFE-RIFE4.0-48fps.webm (2.62 MB, 800x720)

2.62 MB WEBM

Following the discussion from the previous thread;
It's hard to say how far you could push this. I am having issues trying to get much more realism than the following.
granted I have only tried two so far but with the amount of controlnets I am having to use (tile, depth, normalmap, openpose) theres maybe just too much control over the final frame to really make a huge difference.
someone else might have better luck, I am running out of weekend so likely won't contribute again for a while.
This is definitely possible though, theres more than enough here to convince me that this is definitely a viable technique.

File: Potential2.webm (2.18 MB, 1280x720)

2.18 MB WEBM

>>27097229
Posting some webms with fake potential. If anyone wants to take a shot of them, go crazy.

File: Potential3.webm (1.91 MB, 720x1280)

1.91 MB WEBM

>>27099212
Potential #2

"The U.K. will criminalize the creation of sexually explicit deepfake images as part of plans to tackle violence against women.

People convicted of creating such deepfakes without consent, even if they don’t intend to share the images, will face prosecution and an unlimited fine under a new law, the Ministry of Justice said in a statement. Sharing the images could also result in jail."

"This new offence sends a crystal clear message that making this material is immoral, often misogynistic, and a crime,” Laura Farris, minister for victims and safeguarding, said in a statement."

An "unlimited fine"

>>27097229
nice

>>27099565
> The U.K. will criminalize

"The Soviet Kingdom" to be more precise

>>27098742
right side is the most accurate representation of lsd visuals i have ever seen in a video damn

>>27099215
>>27099212
Stop posting this gay shit.

>>27099786
Fuck off you gay retard don't you have anything better to do than troll threads?

>>27098742
yeah, i figured that's about where the ceiling was, because that's about where the ceiling is for starting from scratch, not including recent advances.
But the possibilities in my mind lie more with where the floor is. What can it start with and still end on "photorealistic porn, with some glitches" It can do SFM, can it do hentai or flash? When people on /(a)co/ ask for nude edits of like, disney scenes or something, can that now be automated, and if so for both 3D and 2D?
And that's not all. Even staying within the same medium and level of fidelity/abstraction, can it start with SFM but alter the nature of the contents not just aesthetic style? Could I make Tracer Goth, replace Elizabeth Comstock with Tifa Lockheart? If someone posts a Blacked webm and another guy pitches a fit about it, can I shitpost and respond with "and this is what the cock would look like if it were white or chinese?" Can I disarm traps?

There's so many possibilities here beyond mere x-ray specsing kpop idols and plastering the face of your ex onto pornstars.

>>27099565
>>27098742
Oh no, D.Va didn't fucking consent, guess she'll call the cops on him. Aaaaaaany second now.

>>27099623
because the neural nets are built like your visual neural net, but fucked up because they're not quite right, so it's replicating what your visual cortex sees when it's fucking up.

Just a matter of time until someone posts the perfect Abigail Shapiro edit. My dick is diamonds already

Is it normal that the "animate diff" mode with almost default parameters takes SO LONG?

I was testing with a clip of about 3 seconds and literally was like 20-25 minutes processing and did not finish, I had to stop it, the time only went up and barely saw him advance, that if the vram and graphics were being used therefore something was doing.

If this is so, I give up using that mode, it makes no sense that in a test clip of only 3 seconds it takes so long, I understand that my computer is not the best (3060ti, 32gb ram), but more than half an hour for a clip of only 3 seconds?

I don't know if I have to configure something, (I left everything by default because that mode has always given me problems), yesterday several extensions were updated and it seems that it was going well, I don't know if I have to configure something of the values so that it doesn't take so long in a simple 3 seconds clip.

Translated with DeepL.com (free version)

>>27100064
I would suggest (with the experience that I have so far) that the floor is extremely low. Like I said previously, I’ve dabbled with this idea before and the limitation was always how far you could push the changes before losing cohesion. This was before replacer had interactivity with animatediff however, and since then, the difference is significant.
The key here I think is differences between body shape. Taking someone like DVA and using an SFM to make something with Tifa instead would be really pushing the boundaries (both characters have completely different body shapes) however I don’t think the answer here lies with SFM or Blender but rather with VAM. VAM has a lot of potential but just lacks the graphics quality to really be capable of making realistic depictions.
However with VAM you could very easily make what every scene you wanted with what ever body shape you need and extremely easily slap a relatively low control controlnet animatediff over the result and get what every character your rotted out brain desires. Hell, I think vam can even output a proper normal map/depth map video output, which would make animate diff scenes a piece of cake.
There really is very little reason to manually make models, scenes and then key frame the animations when VAM could do all the hard work for you and animatediff could finish it off.
With a technique like that, and decent hardware with SD2.0 or even 3.0 models, real life pornography will have a serious contender in the AI Generated scope when someone like you or me could generate a several minute scene in less than a day.

The future is here, and contrary to what might be the mainstream consideration, humanity might well be better off for it.

Anyone has a fix for animatediff out of memory? I have 24gb vram and I can't even do anything past the masking phase

>>27100250
It can take a long time, especially with a 3060ti it will.
Try turning on xformers in your launch options if you haven’t already (add —xformers to your launch parameters) that should net you around a 25% increase, but yes, animatediff takes a very long time especially if you have less than like 12GB of vram.

>>27100478
I understand that it is a complex and new mode and it may take some time, but we are talking about a simple 3 seconds clip, in the "frame by frame" mode it takes me 1-2 minutes, but in this one I didn't know when it was going to finish, I don't know if it is wrongly configured or what, because I don't know very well how to configure the options and if maybe for a clip so short and with less frames I have to change something (like the frame lenght that is by default at 36, or some option like that).

If it is normal that for a 3 seconds clip it takes more than half an hour, I will continue using the other method, I'm sad.

>>27100471
I only have 8, but I have to drop the context batch (8) size and fragment length (12), also lower the fps and width/height. Lots of adjusting.

>>27100731
Perhaps you misunderstand.
This shit takes a long time.
Frame by frame is exactly that. It analyses a single frame and renders an img2img of it.
Animatediff references (depending on batch count) 12-16ish frames all at once and it does this for every frame.. you do the math.
Enable xformers and be thankful you can even generate them at all.

>>27097229
>detection prompt: bikini, white shorts, etc; simple is better. this will detect which areas to replace
Can it be made to detect robes and dresses?

When I use animatediff instead of frame by frame I get an error "please download models manually". Are there specific animatediff models I need to install? If so where do I put them?

anyone getting frame by frame batch inpainting to be as consistent as possible with generations? any controlnet settings i need to use? most of mine have very different inpaints compared to other smoother frame by frames on here

>>27100250
It's having to check each frame to make sure it makes sense with the ones before and after. Of course it takes longer than frame by frame. The results are smoother than having each frame drawn separately though. See example from last thread >>27081594

>>27100446
Mmmmm, a fair point.

For that matter, there have even been people piping Replika into VAM.

So someone could fully act out a scene, slap a voice replacement on it, slap a animatediff on it, and Bada bing you've a whole-ass custom porn scene in photorealistic style with the closely replicated waifu of your choice ready to publish and it only took a few hours. No casting and crew, no animation skills and days of sculpting, rigging, and key framing.

>>27101688
It can detect whatever you want, but the less it has to guess the better the results.

From the computer's perspective, the ideal replacement clothes input is 3 non-fleshtone bandaids. The worst case scenario is a niqaab (though by that point why not just let it start from scratch?)

File: e541c50e2b601d5d91cd99593(...).webm (3.68 MB, 576x1024)

3.68 MB WEBM

>>27102520
Can u guys help me pls

>>27102537
Gitgud and do it yourself faggot. You haven't even tried because you're not even posting errors.
The more you spam and demand handouts the less we want to help.

File: 1713325868949703.gif (1.94 MB, 396x560)

1.94 MB GIF

Can this be done in 3d?

>>27099215
hot af

>>27102829
Bmp

File: 1657680745612.webm (1.58 MB, 576x1024)

1.58 MB WEBM

Can someone do her pls

>>27100954
I have a 4060ti 8GB and I can use the default batch size, at full FPS at 512x512 without issue.

>>27102539
I did this one and it turned out great, thanks for passing. Best coom in a week.

>>27103967
Any chance for the result to share?

>>27103967
>thanks for passing
Don't tell me this tiktoker has a dick?!

>>27104068
Nah pure women

>>27104163
>Nah pure women
Good good.

File: dodgeroutw.webm (154 KB, 348x280)

154 KB WEBM

Oh lawd I'm still tryin'.

>>27100731
download a better graphics card then. you might as well be mining bitcoin here for the amount of calculations that need doing.

>>27102829
try it and find out.

>>27104068
do you want them to? AI can do that now.
do you not want them to? AI can do that too.

>>27102537
>>27102539
Bmp again. And yes i havent tryed my own. I hope someone is kind enough to help with these 3 girls. The other clips are not from me. Hopefully a future ty

>>27103304
u need to be checked

File: dodgewalk.webm (144 KB, 338x338)

144 KB WEBM

Matching skin tone seems to be a problem for me for some reason.

>>27105946
Can you try SparseCtrl RGB controlnet? Maybe it can help with skin color, but I don't have time to test it

https://github.com/light-and-ray/sd-webui-replacer/blob/master/docs%2Fvideo.md#sparsectrl

>>27105946
I've a stop motion one of this with some speed. https://rule34.paheal.net/post/view/5859399

anyone getting frame by frame batch inpainting to be as consistent as possible with generations? any controlnet settings i need to use? most of mine have very different inpaints compared to other smoother frame by frames on here

>>27102829
Is someone able to make this 1 in a 3d model?

>>27105901
>do you not want them to? AI can do that too.
The future is now

>>27107414
>>27103304
>>27102539
For the love of god fuck off with your shit requests. Learn to do it yourself or take it to the requests channel.

>>27107452
Not that Anon, but would you mind pointing me towards this requests thread?

Is it just me or has the latest update to a1111 (1.9.2) completely broken Animatediff?

>>27097229
>>27097265
>>27097267
>>27097272
>>27097286
>>27097451
>>27097454
>>27097468
>>27104139
>>27104141
>>27104144
>>27105437
goood thread

>>27099786
Found the gay fatso

>>27107527
>>>/r/

>>27107830
I believe it has. There was some discussion about this in the last thread. Reverting to the previous version will fix it.

>>27108562
... oh.

How would this work to for example switch faces
>face swap

would it be worth it? I kinda want to face swap
my wife to porn

>>27104144
Are there Miyeon ones?

>>27102619
how do you get started? Is there a program?

>>27110150
Use roop-unleashed
You can find it on github

>>27106080
I think my zluda install might not be jiving with this particular controlnet module - I get an error saying there's mixed utility service happening - that is to say, there are queries to use CPU and CUDA at the same time. Maybe someone with nvidia or a directml install might have better luck.

>>27097229
>At 4:30 pm on March 30, 2002, Israeli military forces took over Palestinian TV stations when they occupied Ramallah in the West Bank. Shortly after occupying the Al-Watan TV station, the Israeli forces began broadcasting pornography over its transmitter. The Palestinians were outraged and bewildered. “Why in the world,” one woman wondered, “should one do such a thing?” The answer is simple. The Israelis broadcast pornography because pornography is a weapon in the arsenal of psychological warfare.

>>27110150
face switching is literally the first thing people used it for. that's old tech by this point, should be relatively easy.
https://www.youtube.com/watch?v=f8g0tk1r9iM

that's also probably the single most wholesome use of deepfake that's ever occurred that's still pornographic. good on you for being so loyal.

>>27110583
Out of morbid curiosity, what kind of pornography? What did they consider the most destabilizing?

also, good thing we're making our own then, can't be hit with propaganda if the media you consume is self-made.

>>27110292
No it's all done by hand with colored pencils. Check the OP for tips on which pencils to buy.

>>27099565
Tell the British they can eat my turds

File: wa.webm (3.9 MB, 940x1920)

3.9 MB WEBM

first real attempt at a video using frame by frame (can't get AnimateDiff working probably because I'm retarded). Did this at 12 fps and interpolated up to 24 + manually fixed some frames in photoshop and vegas. really wonky but I'm surprised how relatively consistent this stuff is even doing it frame by frame. i didn't put alot of thought into it either.

>>27105946
>>27105437
>>27106323
Godspeed.

File: cristinysam.webm (173 KB, 540x960)

173 KB WEBM

bros I swear my computer was going to melt itself generating this. Computer just...shut off three times in the process. I had to turn up the fans to 100% manually and the GPU hotspot still pushed 112 °C. I don't know if I fucked over some thermal connections at some point through all this or what but that's pretty insane.

Also...yeah you can see the borders of her clothes. I tried a really tight detection interval given the stark difference between clothing and background to see just how form-fitting i could get.

>>27112203
I would stop immediately and remove your GPU and service it (clean heatsink and redo thermal paste).
112 degrees is well above what most GPU's would consider thermal limit and honestly I am suprised you were able to complete this at all.

>>27112203
Nice vid though.

I'm trying to get it to work by following the OP, but I just get an KeyError when I try to inpaint.

KeyError: 'uberRealisticPornMerge'

Doubt I will get help here, just wanted to complain

File: wa2.webm (3.93 MB, 940x1920)

3.93 MB WEBM

>>27111997
different attempt. only wanted her top to be masked out but i can't seem to find a balance between avoiding the skirt and not. think I'm better off just removing her clothes entirely and masking it back in in post.

>>27112572
i mean, there's been help with errors in the last 3 threads, even polite about it.

it's just people spamming requests begging for handouts instead of trying it on their own first that get told to take a long walk off a short pier with cinderblock shoes.

speaking of the last 3 threads, most of their webms are dropping from the archives, has anyone been saving the relatively successful attempts somewhere or are all those just lost to the aether?

>>27112203
>shut off three times in the process
>hotspot still pushed 112 °C
Bad thermal solution,
Airflow problems aren't to be expected to cause such high temps,
but different causes:
Missing one or more thermal pad,
Thermal paste on the die is either incomplete (It's to be spread manually, unlike with CPUs) or simply dry (for any reason, not just age of the application)

>>27097229
I want to install, but too many files to install

>>27105946
Found out if you have color correction enabled in ui settings it matches to the image more but in this case the image is mostly white so i'd try turning it off.

>>27110150
>>27110436
Download rope-opal, much faster standalone

>>27113823
>>27112333

I ended up pulling it apart to inspect. Thermal paste still seemed tacky but was definitely just center-beaded on initial assembly - pads looked okay too. Cleaned off the sink and the die and reapplied a full even layer across the whole die surface.

I also have a support arm inside the PC that keeps the GPU from sagging, and while I can't prove that it was out of sorts during disassembly, the rubber support pad was definitely out of joint on reassembly. I've had temperature/stability issues with sagging GPUs before that caused sudden PC shutoffs.

In any event I fixed both those things and the GPU hotspot has improved to 85 - 90 c during the 100% loading image inpainting phase so that's a plus.

>>27099786
Average cuck/tranny thread contributer
No verification required lmao even the website knows you a bitch

>>27112572
I have no idea but you could try a different model. There are many options: https://civitai.com/models

>>27102829
>>27102829
>>27103304

I believe you can do it bro. You have my full support

>>27104141
>>27104144
You guys are heroes, fighting the good fight on behalf of your fellow low status men who have been Alienated by feminism and modern dating culture

>>27104144
This one is pretty smooth.

>>27115386
God

>>27102537
>>27102539
>>27115386
Pls

I followed all the instructions but I only want to use it to "deepfake" photos.

I might be missing a step or something but it doesn't retain the original girl, it just replaces her with someone else.

Thanks for all the instructions.

>>27111997
>>27112592
Bro you should be able to get better results than that even without animatediff. What are your prompts? Can you post the original?

>>27105946
this usually occurs due to the mask missing a couple of pixels on the clothes leading the inpainter to draw the body whatever color those pixels are - see if there’s any settings to expand the mask border

>>27116917
deepfake as in replace the face, or deepnude, as in replace the clothes with nudity?

if you just want to replace the face, see >>27110436 and >>27114218

if you want to nudify and not replace face, then follow instruction in OP.
If you followed instructions in OP and it's also replacing the face, then you have done something wrong. But at the same time, are probably getting the best results possible that aren't potentially a lawsuit, so consider if it's worth it to you to fix it instead of stopping while you're ahead.

If you can't figure out the mistake, i suppose you could always run the nudifier, then faceswap the correct girl back into the ai generated girl.

>>27097229
Can someone for the love of Christ do a decent Dana Perino deepfake plz?

>>27098742
>>27095956
These are great.

>>27118212
>george W bush's 52 year old press secretary
..
Why??

>>27118453
She's sexy as fuck, that's why

>>27118678
the only explanation I see for this belief is that you too are also 52. actually, that would explain why you're not able to make it yourself, since boomers are allergic to advanced computer use. go watch fox news.

>>27118883

He's interesting and funny, you're lame and boring, IMHO.

>>27119376
i'm still not deepnuding a post-menopausal woman for you.

>>27119445

No one cares.

>>27119523
clearly you do, or you wouldn't have asked me to do it.
Either be productive or just lurk. basic social rule of the site.
Begging in a general thread that literally has step by step instructions on how to do it yourself, and where people have been giving advice on how to fix errors (at least when given enough info to assess the problem) if you mess up, is effectively shitposting. The recipe and stove have been supplied to you, you could make it yourself, to the exact specifications of your desires, but instead you're here demanding other people prechew and spoonfeed you, you poor helpless baby bird.

>>27119609

Not reading that.

>>27119610
didn't expect you to, we've already established that you can't fuckin' read

>>27119633

It's true, you've got me there.

File: New folder-5x-RIFE-RIFE4.(...).webm (3.89 MB, 1424x1096)

3.89 MB WEBM

Ok. Here is my first attempt at generating a scene with VAM and using animatediff to overlay it.
I tried to make this look as "realistic" as possible, in the sense that I used a realistic model (HelloRealistic) to really see how much potential lies in generating a pornographic scene solely with computer/ai generated data.
For a first crack I think this turned out pretty good.

>>27115386
more plz

>>27121519
this isnt animatediff or replacer. this is just img2img, and its a terrible one at that.

>>27119872
that looks pretty neat, my dude. doesn't look like a Replacer job like the rest, did you do this in the animatediff section of the img2img tab or did you use something else like comfyui?

golly i wish deforum's RIFE functionality wasn't fugged on my install

>>27097229
when I generate it kind of just places shit without any proper context, like it'll put the torso on backwards so she's got a front-butt and other bizarre shit

>>27119872
man what an improvement. I don't know how people use VAM, the before picture is uncanny as fuck, like one of those $2000 sex dolls come to life.
Excellent work.

though I see they've never quite ironed out that hand problem, lol.

These use very old models. Newest shit is pretty much 100% stable up to 60 seconds. Shame that it is not yet available to the masses.

>>27114218

Can you be more specific what setting? I checked the UI settings but I couldn't find this option.

>>27116917
OP's method is for video. It's way overkill for just photos.
Install automatic1111 and an inpainting model (if you've followed OP's instructions you've already done this I guess).

Now go to img2img -> inpaint
add your selected picture
draw a mask over the parts you want replaced
under "masked contents" select "fill"
further down set the dimensions to match your photo
type in prompts
hit run

short video for dumps?

a short video on how to install?

>>27122544
Yeah it's not avaliable for a reason sadly. /h/ have been waiting and begging for years and ended up creating their own model. "Pony", which does everything, not just poses and 'photo' gens.

XL + Pony + Realism model would be insane. You can hack it using LORA's tho

>>27123285 Read OP >>27097229
If you get any errors come back

>>27121845
>>27122390
It is actually replacer, just with a few controlnets, then passed through img2img batch with a low denoise to smooth it out a bit.

Ikr. VAM had so much potential but it just looks like a fucking fever dream most of the time, though, with so it might have it’s uses.

File: AnimationMP4_00016_waifu2(...).webm (1.72 MB, 1280x1920)

1.72 MB WEBM

i lost the orignal video kek

>>27124541
I really like what you've done here, anon. Good consistency frame to frame, high fidelity to the image. I'd be very interested to know more about the specific steps and modules used - to what degree replacement versus full prompt generation was used, that sort of thing.

Keep it up, very cool.

>>27124548
im confused. which video is the original?

blessed thread

>>27119872

Not bad at all--I'll see if I can replicate/improve your workflow. Which controlnets did you use to generate this webm?

>>27124789
I think this one was likely depth (depth-anything), probably normalmaps and a weak tile with a denoise around 80%. thats usually my go to workflow.
report back if you do please im curious.

File: AnimationMP4_00015_w.webm (816 KB, 1024x1820)

816 KB WEBM

>>27124605
>>27124587


i used that comfy setup

i dont know what is the source lol

https://civitai.com/models/330169/astraalianimatediff?modelVersionId=369955

The age of AI,
What a time to be alive, young faggits.

>>27124541
>>27124548
>>27124889
alright you convinced me.. Im gonna figure out how to do this shit and start doing requests here for a little while for shits and giggles...

I stopped making my own sdg gens becuase so many folks here were contributing... I want to do this shit and recontribute for anons...

if theres any quick advice please lmk, otherwise im gonna continue what I did last time which was basically piecing this shit together little by little until I fry my gpu .... anything helps thnx

>>27102539
>>27102537
>>27125611

>>27125611
>>27126203
I would HIGHLY recommend you don't cut your teeth on these videos. they are bad for multiple reasons. too much rapid movement, clothing too similar to background color, multiple subjects in the frame, reflections, etc etc. bad bad bad.
Start with a video with one subject, with little movement and good contrast between foreground and background.
Oh, and also, the more of the frame your subject takes up, the higher the quality will be without having to increase resolution.

somebody do neekolul's tiktoks

>>27121519
>>27124689
this tooo

>>27104141
>>27104144
proof women are just fuckmeat if you remove the clothes

Another bump

Second after another

>>27124605
None in that post that's for sure

>>27097229

just use clothingremover.app

getting this when i try to run replacer...

Traceback (most recent call last):
File "D:\DF\Stable Diffusion\modules\call_queue.py", line 57, in f
res = list(func(*args, **kwargs))
File "D:\DF\Stable Diffusion\modules\call_queue.py", line 36, in f
res = func(*args, **kwargs)
File "D:\DF\Stable Diffusion\extensions\sd-webui-replacer\replacer\ui\generate_ui.py", line 285, in generate_ui
animatediffGenerate(gArgs, video_output_dir, resultFrames, fps_out)
File "D:\DF\Stable Diffusion\extensions\sd-webui-replacer\replacer\video_animatediff.py", line 115, in animatediffGenerate
initImage: Image.Image = processedFirstImg.images[0]
IndexError: list index out of range

File: neekoniii.webm (393 KB, 432x768)

393 KB WEBM

>>27126393
>neeko
lol gave it a shot. keep getting these ghost tanktops appearing for a few frames.

It seems like her still images are very robustly AI nudified online already.

>>27125611
I just ordered a gaming PC with a RTX 4090 to start doing this shit and I don't even play games.

>>27099565
>men jerking themselves off alone in a room is hurting women
BAHAHAHAHA

>>27097229
Imagine people doing this on children lmao

>>27097229
Has anyone managed to do this for one of those cam whores who deliberately don't show nudes, then have it reposted to them? I'd love to see the reaction of them having their vagina shown to the public without their consent, it would be funny asf

>>27129833
this tooo

>>27131438
only inbreds would look at this shit and actually think its literally removed their clothes. its nothing more than realistic looking r34.

>>27131599
Then how come they're trying to make it illegal? If the nude parts aren't real then why care?

>>27131438

In the end that's probably what will kill this. It wont be their "right to their likeness", it'll be a commercial argument that "our businesses are being stolen from." Because feminists have convinced society that worthless whores who contribute nothing to society except the holes the were born with are somehow valid people who deserve resepect.

In telegram when I use bots to deepfake, using one picture with the Fake face is enough to replace the original face and have good quality. Is it the same with this software or do I need an extensive photo database?

>>27131805
there is an automatic11111 extension called reActor that I've had decent luck with using a single photo of a face to generate good quality work. You can give it an image to create a model with, then tell it to use that model for face replacements.

How "real" it looks depends on how much like your subject the original looks like, and it's a post processing step that replaces the face of an image rather than generating it as part of a prompt like a LoRA would...but it can get the job done.

>>27118883
Wut. You sound like a fag. Dude like a milf and you're here crying about it? Kek. Also, you sound like you're 10 years old.

>>27118883
He probably does watch Fox News, to fap to it, since she's an anchor there.

nope

File: output.webm (2.01 MB, 405x720)

2.01 MB WEBM

>>27097229
Tried it for the first time today and other than it being inconsistent due to limited prompts i think i can work with it to make it better

can someone post a screenshot of their controlnet settings in the replacer tab pls

>>27129833
I've gooned to her tiktoks before but now I can really put a dent in the ceiling thank you for this

File: MagicNippleNeeko.webm (3.65 MB, 576x1024)

3.65 MB WEBM

>>27134772
I've definitely found dynamic movements that occlude the regions you intend to inpaint does take some doing to prompt around.

>>27134817
magnificent effort

Can anyone make one of Maisie williams?

>>27136105
how about you follow the simple instructions in OP and make it yourself, then post it for all to enjoy?

>>27131685
>feminists have convinced society that worthless whores who contribute nothing to society except the holes the were born with are somehow valid people who deserve respect.
This is fucking hilarious and sad at the same time. They want people to see past their holes, but at the same time act like whores milking men for money. Quite sad how no one points out this blatant hypocrisy.

>>27103304

Why do you want to see his dick?

>>27099565
What makes tools like automatic1111 or comfyui safe to use? How do you know they won't rat on you?

>>27129833
Workflow?

>>27137356
Well most normies who download it won't be able to say for certain. But the downloads and programs are hosted on github in a visible way. To those with the proper knowledge, they can check the programs to make sure they're not reporting to the outside world in an inappropriate way, like uploading your images or reporting you for using certain prompts.

>>27099565
The pearl industry is taking off with sales through the room from so many being clutched by the ancient crypt keepers. Why can't we have leaders that aren't 3000 year old closet faggots?

>>27129833
basado , keep em up

>>27137199
over

bump

Another one

File: mika1 (2).webm (3.87 MB, 1920x1080)

3.87 MB WEBM

Not bad i guess (frame by frame method)

Shot in the dark.

When I try to run Replacer with AnimateDiff, first I get a bunch of these errors:
"C:\actions-runner\_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\IndexKernel.cu:92: block: [402,0,0], thread: [64,0,0] Assertion `-sizes[i] <= index && index < sizes[i] && "index out of bounds"` failed."

Then, I get these:
"Exception in thread MemMon:
Traceback (most recent call last):
File "threading.py", line 1016, in _bootstrap_inner
File "G:\Here\sd\sd.webui\webui\modules\memmon.py", line 53, in run
free, total = self.cuda_mem_get_info()
File "G:\Here\sd\sd.webui\webui\modules\memmon.py", line 34, in cuda_mem_get_info
return torch.cuda.mem_get_info(index)
File "G:\Here\sd\sd.webui\system\python\lib\site-packages\torch\cuda\memory.py", line 663, in mem_get_info
return torch.cuda.cudart().cudaMemGetInfo(device)
RuntimeError: CUDA error: device-side assert triggered
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions."

I've googled and tried a bunch of supposed fixes, but nothing works. Any help, before I give up?

>>27141440
Excellent actually.

>>27141607
try going to the UI settings -> Optimizations and check "pad prompt/negative prompt". then restart stable diffusion, and see if that works.
I think this is how I fixed the same problem, but it's been a little while so I'm not positive.

WTF bumpussy

anything i could add to remove black box censoring or should options from op suffice?

>>27137679
because you won't guillotine them

>>27144184
should be fine, but results might be slightly wonky because it doesn't have any info to work with from that area.

not sure there's anything to do about it though. Other than maybe run it back through a second time to fix the bugs of the first time, since it's working with something closer to the desired product. No idea if that would actually work though. Still, worth a try if the first pass isn't great.

Anyone ever run into issues with a file not found error after mask creation? Its looking in the output directory for a `<ouput_dircetory>\\fragment_1\\frames/image` and it generates the frames directory with images with `frame_<x>.png` in there, but nothing about image there.

I'm guessing there was a change in a naming convention somewhere and Animation Diff is looking for `image_<x>.png` instead of `frame_<x>.png`

>>27145057

Actually looking more at this, I think I may be hitting an issue with ControlNet not saving images further up. an error with `invalid load key, 'v'` digging into it more.

>>27104144
Did anyone save the Sunmi - Pporappippam one?

>>27145104
Okay, figured out the issue, I pulled the ControlNet files with git, which I guess left some pointers rather than pulling the actual file. Downloaded each file manually and seem to be moving again.

>>27145327
Oh, also I was using an older version of AnimateDiff (1.13) and now I just pull from master.

File: uglyray.webm (3.67 MB, 564x640)

3.67 MB WEBM

Sometimes it makes stuff that looks pretty good...sometimes it looks like trash like this. I've yet to achieve anything similar to what I've seen other people put out here, but maybe I'm just choosing challenging content to mimic.

>>27145501
Also the delineation at fragment boundaries is really annoying maybe I need to go single fragment.

anything good posted in the previous threads?

File: 1713610761891071.webm (3.88 MB, 1440x1024)

3.88 MB WEBM

>>27145603
this was the first replacer over animation test. looks pretty good to me imho.

you know what? fuck you
*throws a pic of you into sd and jacks off to a deepfake of you*

>>27145501
lol, there's something very meta about this. it's like she's aware how weird her breasts are acting and ends it with the 'Wow' at how absurd it was haha

>>27141440
extremely good

>>27097229
Tutorial on how to uninstall/delete if shit doesn't work well on pc?

>>27146284
Sauce?

>>27146294

https://www.tiktok.com/@daisybloomss/video/7242427277428968746

>>27134772
>>27134817
>>27141440
>>27146284
this tooo

>>27099565
Why is it always only about the women? Why not make the ban and its reasoning gender neutral? Or is it still perfectly fine to generate these for dudes?

>>27146262

If you can't figure out how to uninstall this, you aren't smart enough to be using this in the first place...

You literally just delete the stable diffusion folder.

File: output.webm (309 KB, 1280x720)

309 KB WEBM

>>27143080
It actually worked!
Thanks a lot dude!

>>27147495
you have reading comprehension issues

>>27148289
whats the original video?

>>27148340
How so? The entire reason they are doing this is to "tackle violence against women" because this material is "often misogynistic". There is zero gender neutral reasoning for this, it's all about protecting women. You can piss yourself in seething but that won't change what they said.

>>27148358
the ban is gender neutral

>>27145501
Oh please give me all you have of her

>>27147617
>>27145501
I love you anon
Pls more marish

>>27147617
>>27145501
Unexpected but HELL YEAH !

File: testrun.webm (272 KB, 1150x512)

272 KB WEBM

I'm trying to get a feel for when and how controlnet can be used to improve outputs. In this case I like that the depth map preserves the torso more accurately, but there are definitely artifacts that start to enter the image when certain creases of fabric alter the depth map. Maybe it's a function of tweaking its sensitivity, I dunno...

>>27149503
I have no idea of how this works but Im amazed by the fact it is her you are using...
Pls everything you have

>>27145640
Looks so good I love this style

>>27148348
https://www.youtube.com/watch?v=z9CRvCmJUnI

enjoy how music videos used to be, zoomer, before something like this would get you cancelled.

>>27149696
i think so too. if I ever free up enough ram I'll maybe try to do some.

>>27149503
try using a combination of openpose, normalmap, and depth

>>27146284
>>27146294
>>27146349
Theres already nude versions of all the dances she does on tiktok whats the point of this

File: marishfight.webm (120 KB, 416x280)

120 KB WEBM

>>27149923
Tried the three in combination on this image here; lots of movement and shadow, so it isn't perfect by any stretch

Alright, I've been fighting through errors for a week, and finally got it to output a file for the first few frames of a video, but the masked area is just a black/pink noise pattern.. Anyone experience anything like this before?

>>27097229
Does anyone have the hi-la kl-ein from h-3h3

>>27151087
This is the best marish thread ever
Keep it up

>>27147617
>>27149503
>>27151087
this tooo

bump for more naked bitches

>>27151089
Sounds like you might have the wrong checkpoint selected in one of the options. At least thats what happened to me.

>>27146284
Lmao, the REAL #fakebody

>>27152462
>>27152462
>>27152462

Would this help?

>>27151087
Any new stuff/progress on her ?

hello guys would like some suggestions on what settings to use for animatediff + reactor using a 2070 super 8vram for performance with still good quality?

File: rishboxing.webm (65 KB, 324x324)

65 KB WEBM

lol why did it think she needed braids. is it reading context clues from seeing a scene from a fighting match?

File: rishcupper.webm (403 KB, 400x518)

403 KB WEBM

tried this with multiple controlnets and without and each iteration seemed to get worse.

Single image inpainting seems to work great...controlnet inpainting seems worse. I can't get good nipples to save my life...or at least to save the gif.

Maybe it's a resolution thing, idk. but 512x512 is the SD1.5 native model resolution so...I dunno.

>>27156085
OP could easily share screenshots of his configurations

>>27097229
When you are guys going to start grafting big waggly cocks onto these bitches?
Just dont overdo the size. The futa bullshit is ridiculous.

File: notbad.webm (366 KB, 1080x1080)

366 KB WEBM

I'm fairly pleased with the results here. I think it's been done though, and better. But I think it looks alright. Didn't have time to render the whole video before bed though

>>27134817
>>27129833
please more neeko or pokimane

>>27156073
>>27156085
>>27156519

I fucking love you man
Please more of these redheads

>>27134772
Keep a consistent seed, change it from -1

>>27156519
Remarkable work, anon.

>>27156519
>>27158487
whats the og videos???

>>27154156
Alright, I'm guessing I'm dumb, so from the uber realistic porn merge, should I download both the inpainting model _as well_ as the base model?
In painting model I'm assuming goes in with ControlNet Models and I'll select it with AnimateDiff inpainting model.

There's other errors going on when I do that, but want to make sure I'm understanding the process correctly.

I've also tried just the inpaint model and just the safetensor base model with other errors, so trying to find the correct path.

>>27158937
I'm not an expert but I'll tell you what I've got:

I've got the inpaint and base models. I choose the inpaint as the overall model. I allow the controlnet inpaint to use the base model to override the overall model (this is recommended within the UI I use).

The controlnet folder is not where I place the inpainting version of my model - there is a specific controlnet model called 'inpaint' that I think you are obligated to use there, but I can't say I've done much searching for alternate controlnet inpaint stuff.

>>27158487
You're the best !!!
More old scully perhaps ?

>>27158516
Bump for this

>>27160273
Wow !
Holy fuck
Keep it going pls...so detailed

>>27158937
I've tried different combinations of where to put inpainting and non inpaintings as I didnt understand completely either. I believe, when using replacer with animatediff-, the very top overall needs to be an inpainting, there is another under advanced which I do not check for overriding, the controlnet inpainting which im using "control v11p sd15 inpaint" in the animatediff mode section, motion module "mm sd v15", and stable diffusion checkpoint under animatediff as the NON inpaint. So as to the first question, yes you need an inpaint and a non inpaint. As for controlnet im not sure as they ahve their own settings/models, the small bit i played with em i just let it choose its own. Someone suggested try using a combination of openpose, normalmap, and depth. It takes forever atm.

https://www.bbc.com/news/articles/c6py33gxk74o

they are fighting back against stuff like this. it's too powerful

OP will never post any configuration settings but here's what I've been using for frame by frame:

Detection: body, torso, chest
positive prompt: [use your own w/Loras]
Negative: [use your own]
Steps:30
Sampling: DDIM Karras
CFG:8
Inpainting mask blur:25
Inpainting Padding:80
latent noise w/denoising set at 1
Controlnet0:Normal Midas 0.5
Controlnet 1: Openpose 1.0
Controlnet 2: Depth 0.6

>>27161025
Fuck BBC, no cucks, plz. :p

>>27161657
>Big black cock news

>>27156085
>>27156519
>>27158487
>>27158837
>>27158842
>>27158848
>>27158853
>>27158859
>>27158884
>>27160273
this tooo

>>27099565
Australia is doing the same shit

bump

>>27097229
All my animatediff attempts come out slightly orange no matter what settings I change. Anyone have the same problem? Frame by frame works fine, so it's not an inpainting or mask blur issue or anything else, but animatediff is always orange. I've tried systematically changing practically every single setting and no dice.

>>27164245
I have found, occasionally, that I need to play with a setting called img2img_color_correction depending on the clip to improve odd colorizations. Sometimes it keys off the background to a significant extent when doing the inpaint with animatediff/controlnet, and women in RGB environments get RGB bodies for example. I've also seen people recommend adjusting your detection mask expansion, since it might include some pixels that are distinctly not meant to me masked and keying off it for color.

Grain of salt, random faggot on the internet, etc.

>>27141440
9/10
WOW amazing stuff

got moar like her? or even of hers

>>27160273
Pls more scully
Any younger 90s scully

>>27150750
not only that she also does porn. kind of a waste of time, really

File: 1714690645360199.webm (1.31 MB, 1328x1080)

1.31 MB WEBM

something for fa/tv/irgins

>>27161544
latent noise eh? now that's interesting. I've been doing fill predominately myself but I might try that out. might get rid of some bizarre artifacting.

>>27098742
>>27119872
>>27145640
Would EBSynth be better at this?

File: 1714587613431787.webm (2.62 MB, 720x900)

2.62 MB WEBM

>>27097229
pls do this webm senpai sama

>>27167288
Just skinify her clothes at this point

>>27115386
What have you done... IT'S FAPPERINO TIMEEEE

>>27097229
as the Figo Fago BFF brothers sing "there is nothing more exciting than a dressed woman" wert https://www.youtube.com/watch?v=KaOEVHQoHzM

More Marisha ???
Miss these already

>>27097229
Heads up for anons using automatic1111
The sd-webui-animatediff github page says you have to use his version of the motion module. Apparently it's different than the regular AnimateDiff motion module (the one linked in the OP instructions).
Go to the sd-webui-animatediff github, scroll down to "Model Zoo", click the link to his huggingface repo, click the motion module folder, and download "mm_sd15_v3.safetensors." That's the one you want to use. This solved my issues with weird colors and flickering.

>>27166701
Op of these here.
It can be, the TemporalKit workflow does have benefits over this technique however it struggles greatly with aggressive motion.
I’ve done many EBSynth conversions like these, it’s much faster and gives pretty good results, however generally animate diff gives far more consistency with the ability to really push the denoise much higher. It takes much longer, and requires much more powerful hardware, but it generally is better.
You can see examples on my civitai profile:
civitai.com/user/violet

File: lbail.webm (237 KB, 860x860)

237 KB WEBM

the weekend has arrived, lawd have mercy

>>27141440
Could you post the original? Would like to see the transformation.

>>27170807
Good catch, i'll test this out later

>>27158859
more EXID and AOA plz

WE DEMAND SEOLHYUN

>>27172258

can i suggest using this for source material

https://www.youtube.com/watch?v=KKxeZcSVVMo&list=PL6lWyxUovC1KEHut8v9ftfUUYZnSwhKGf

>>27172258
>>27172280
is this bitch still around even? whats she up to?

>>27171209

https://www.youtube.com/watch?v=nccU1SYs2zk

22:50

>>27172367
Thanks. AI was VERY generous with her boobs lol.

>>27171195
Fuck yes !
Any ashley too ?

>>27171195
Is her pits A.I too or real?

>>27172359
idk probably just rich and doesnt need to do that shit anymore

I think she acts in a k drama once in a while

File: ashj.webm (865 KB, 368x368)

865 KB WEBM

I'm not especially enthused about the tone of the skin. Tried using sparsectrl and the img2img color restoration at the same time and I think it keyed off the couch too much.

Also there are very few videos of Ashley out there not wearing baggy clothes and/or slouching, which makes it have to guess more. I generally prefer to run a leaner detection mask and fill in as little as possible to make the result more believable.

for the controlnet model and animatediff model should i use the non-inpainting model like epicphotogasm and override it, but the webui model being something like uberrealistic or should i use uber for both because animatediff is working but not really making the subject nude

>>27172258
YES

>>27174282
Damn thats good
Not as hot as marisha but I like the cartoony tits