You were in my /b/ thread the other day, I deleted all my porn but here's a bump!

File: grok_video_2026-03-02-08-(...).mp4 (1.43 MB, 464x688)

1.43 MB MP4

>>81493490
Hey thanks for bumping, welcome back anon. Don't delete your porn ever again you fool!

File: 1767090022354466.webm (3.92 MB, 1280x720)

3.92 MB WEBM

bump cuz I'm interested

>>81493761
Which AI was this? I'm using Grok right now so I can't get anything explicit unless it's completely accidental when the censors miss something, like when I brute-force a prompt repeatedly until a random variation slips through.

File: grok_video_2026-03-03-14-(...).mp4 (1.53 MB, 464x688)

1.53 MB MP4

All my stuff is herm btw. There's such a shortage of it out there that's what I've been focusing on making. Everything furry is welcome though.

File: 142430_UPIN_00001.mp4 (3.32 MB, 720x1272)

3.32 MB MP4

Got a few things I suppose I can post now that we're back. Not a lot, though. Dumped all my AI genning stuff for the time being because I think it might have started fucking up my graphics card, and right now is not the time to need to replace one of those things, especially a 4090.

File: 1759787451342090.mp4 (788 KB, 480x640)

788 KB MP4

>>81492881
A cute futa Rikki I gen'd a few months ago

Protip op: go to one of the stable diffusion threads and follow the discord like to "furry diffusion" there they have a updated daily category for animated stuff (mods I am not advertising for a discord just simply pointing him in the direction of more stuff)

>>81494303
I've considered killing myself instead of ever using Discord for anything other than speaking to my friends while playing games with them. Is that the only place left online where this can be meaningfully discussed?

>>81494399
I'm not sure if it's the only place left but it's full of good porn and most of it isn't posted here or on e6ai so it's worth checking out if you like animated furry porn, I don't interact with any of the users I just lurk but the people there seem nice

File: grok_video_2026-03-01-11-(...).mp4 (1.62 MB, 688x464)

1.62 MB MP4

>>81494451
Alright maybe I'll go check it out. I'll wear sunglasses and a big hat so I'm not identifiable.

File: grok_video_2026-03-03-14-(...).mp4 (1.55 MB, 464x688)

1.55 MB MP4

This was a weird one, ended up being an experiment at just how much I could possibly get away with slipping past all the censors and wrangling the disobedient camera.

>>81494253
Anything else with this scene? Did you get her to sit on that?

>>81494290
Very cute, nice twitching.

>>81492881
seedance 2.0 was leaked and only requires 24gb vram, uncensored if you localhost

File: ComfyUI_00197_.webm (421 KB, 512x512)

421 KB WEBM

Sweet, thread is back up.

File: grok_video_2026-03-03-16-(...).mp4 (1.42 MB, 688x464)

1.42 MB MP4

Getting penises to stop moving randomly is surprisingly challenging in Grok. When stringing together multiple extensions, my first prompt usually has to be
>Her penis remains erect
If it isn't, it sometimes randomly deflates or disappears. It's almost become a mantra or prayer while working on these. Also the camera has to be absolutely bullied and gaslit to move close to a character. The implied blowjob in this one was essentially luck. Getting what you want with Grok is like trying to force two magnets together with the same polarity, constantly slipping around and fighting you until you finally get the perfect angle.

>>81495026
proof? grok told me it was fake.

File: grok_video_2026-03-03-17-(...).mp4 (1.39 MB, 464x688)

1.39 MB MP4

This was wild. I somehow got Grok to do penetration, it even gave it a wet sound effect when it happens. The animation is fairly jank but this is another boundary-pusher with the censor. Something I've noticed is that if you have a 2D character perform complex movement or the camera moves dynamically around the character, Grok will render them as a 3D model. Once that happens, you suddenly get a HUGE amount of leeway with what you're able to do without censoring. Watch the character here closely, you'll notice parts of her transition to 3D during the clip.

File: grok_video_2026-03-03-17-(...).mp4 (1.65 MB, 464x688)

1.65 MB MP4

>>81497709
Another quick one with the same character. I love how super thick her physique is while still having very feminine proportions. I did an extended scene with some nice dirty talk but obviously I can't upload it here.

File: 132354_UPIN_00001.mp4 (3.61 MB, 720x1272)

3.61 MB MP4

>>81494942

Most of the gens were just varying degrees of success with stroking it, which was the goal. There were a couple standouts that it made on its own, though, like this one where she leans down and gives it a nice wet slurp...

>>81495330
>fat baker Judy, square gen
Where'd you get the seed image? I made a bazillion of these just wondering if it was one of mine.

File: 145426_UPIN_00001.mp4 (3.41 MB, 720x1272)

3.41 MB MP4

>>81494942
>>81498144

...and this one where she nestles it between her breasts. Always neat to see the AI just decide to do shit like this on its own.

>>81498181
>Always neat to see the AI just decide to do shit like this on its own.
Yeah I've wondered about that. Is the AI deciding to add that because that action is extremely likely based on previous prompts?

File: inflation test.webm (509 KB, 640x368)

509 KB WEBM

>>81498151
Probably off this board at some point, so It's probably one of yours. Don't really get them from anywhere else.

File: 175501_IN_00001.mp4 (1.8 MB, 480x848)

1.8 MB MP4

Bump

File: grok_video_2026-03-03-20-(...).mp4 (1.52 MB, 752x416)

1.52 MB MP4

This one isn't great but I kept it just because it was really weird but interesting how Grok included two separate camera cuts without ever being asked, but it maintained continuity between the three clips. I've tried prompting camera cuts before and it usually doesn't understand, and if it does then sheer madness happens after the cut. Like I once had it cut to a camera position above my character, and then it rendered the ground under her collapsing into the earth.

File: bananaduck-2000-hq.webm (760 KB, 688x464)

760 KB WEBM

i don't know all the grok tricks but this is pretty neat

>>81500835
I can talk to you about the tricks I've learned so far. I've been able to get away with all kinds of shit.

>>81501026
Wow a capybarababe. Haven't seen one of those before!

>>81500835
>>81500911
tits on Grok is pretty easy in both images and video. and I've seen a few cartoon cocks in still images.

File: Beachadventure.webm (3.19 MB, 1080x1920)

3.19 MB WEBM

>"Money can be exchanged for goods and services."

>>81501152
Man you can do way more than showing nudity. Everything I posted above was all Grok, including the vaginal penetration.

>>81498181
How did you gen something of this quality? I just got a 5090 and want to dabble into this.

File: raptorsquattingonapurplepole.mp4 (3.21 MB, 928x1376)

3.21 MB MP4

>>81501501
this is the best i managed to get. I just vaguely described the action without any explicit details and grok filled in the blanks. It got smart after a few gens though and stopped doing the penetration and adding anything to the prompt to be more explicit or make the object more phallic got hard moderated. It was a pain in the ass and it's not fapworthy for me (neither is yours, no offense). Grok's pretty lenient with it's own images though so it makes me think there could be jailbreaking prompts that will net decent looking penetrative sex in videos but I haven't seen evidence yet. I've got some old gens from when grok was most censored of anthro owls laying eggs but the video model is different now so I don't know if it would still be possible to do. I'll dig em up and post one soon.

File: groklion.mp4 (2.92 MB, 464x688)

2.92 MB MP4

>>81503028
i did this with grok using an image from sdg. Didn't prompt for penetration at all, the image just coaxed grok in that direction, and it's only implied, but if this can slip past censors on upload2vid then I imagine image2vid could be a lot more explicit, but it would still require some really tight prompt engineering because the censors are really trained to catch penetration.

>>81493815
wan 2.2

File: 00949-1391124644.png (2.75 MB, 1344x1824)

2.75 MB PNG

I don't have a good rig to gen video but I like to follow some video makers on youtube and would like to talk about them. A big one making a lot of talk is gossip.goblin.
https://youtu.be/csz5Honnv-w?si=_CZsaRWvBEsewCu-
https://youtube.com/shorts/un8kMmoPdi8?si=Asw-i80n3R-xUY7r

He likes british cyberpunk, which as a washingtonian I feel offended by. Still, he has a lot of lore behind his videos and is pretty consistent.

File: WereWaifu-NoSound.mp4 (1.95 MB, 464x688)

1.95 MB MP4

>>81493815
I have been experimenting with Grok Imagine.
I can get most of what I want, though I'm a boob lover so most of my stuff has to do with booba.

First you need to generate the images using Grok, uploading your own puts them trough a different censorship model than those it generates itself. You can gen an image with grok, download it, upload it, give it the exact same prompt for the uploaded one as for the one already on the site and the uploaded one will fail but the one on the site will work, even though it's the exact same image.

For genning I just write: Anthropomorphic "whatever you want" with breasts "in or at a location of your choice".
Like "Anthropomorphic werewolf with breasts laying on a bed".

A prompt with "Hands moving up her" will result in a breast massage like 80% of the time.
https://grok.com/imagine/post/25ecf970-8462-4356-9081-3bdc19b689ad
https://grok.com/imagine/post/8f78784c-38e5-4203-9ac0-a46c35cb56fc
https://grok.com/imagine/post/c28373ad-9cb5-407d-a7f5-1cbe29022a18
https://grok.com/imagine/post/692cb9c2-b3dd-412e-8fb5-91e062981dcf
https://grok.com/imagine/post/c0e70198-12bf-46bb-a0aa-eea598479f72
https://grok.com/imagine/post/95c0d6d4-2839-4577-b702-ff03c0a1659d
https://grok.com/imagine/post/038bc0db-83c8-41b3-a4ba-7f2035c59c26

"Bounces up and down" will give:
https://grok.com/imagine/post/d5f9f283-6320-4fde-9ffb-f0482c11a14b
https://grok.com/imagine/post/0ebf501f-30d2-4939-90cb-b68c57f4b336
https://grok.com/imagine/post/92aea973-50ac-4d19-8a08-005e31f1ecb2
https://grok.com/imagine/post/ce474377-2765-4a6a-a5f2-46e36d829261
https://grok.com/imagine/post/bc4dce51-bbae-4264-a42d-416066db912b

Merging them for grabbing boobs and pounding has been harder.
https://grok.com/imagine/post/1ba2d5b0-1e39-4891-b4cf-07503c2749c6

Prompting isn't really about saying what you want on censored models, it's more about adding elements that the Ai interprets. "Hands moving up her" isn't explicit so passes censorship, Ai then tries to use it.

File: boob1.webm (494 KB, 464x688)

494 KB WEBM

>>81506754
I've been able to get some mouth-on-boob with "nursing gently" before... and sometimes some vague "moving down" instructions will give a surprise vulva, or a surprise nonsense-genital.

but it's clear that this has been trained on some amazing stuff.

I wonder how many other people went in going "I wonder what I can get away with" and "lol duck tits" and then got surprised by the results and then got hooked on the inconsistent reward of it long enough to actually get turned on by a gen.

and based on everything i know about Internet monopolies, there's a good chance that whole chain of events is on purpose, and devised by some of the most highly-trained and well-paid psychologists in Silicon Valley.

So I'm hoping that an at-home NSFW model gets results as good as these with just a prompt and an optional start image, and not fucking around with loras for 4 hours. Best case scenario is that it's consistent enough that the casino thrill goes away. At the very least it'll free us from everyone else's Dark Pattern Website.

Bump

File: 080133_00001.webm (3.94 MB, 800x1440)

3.94 MB WEBM

>>81507706
I personally think that Musk allows Grok Imagine to create porn because it makes it relevant.

I'm 100% convinced pretty much every single Ai generator that is censored will go out of business.
The uncensored ones will be behind the big corpo ones but enthusiasts will eventually catch up.

There is a damn good reason Wan 2.2 is so popular and probably has more modifications than all other models combined.

Sooner or later we'll get something like an open source seedance that is uncensored.
It's literally inevitable.

I'm not worried about hardware requirements at all. China sells 96Gb GPUs for around $1200. I'd bet money they will sell 512Gb cards for less than $2000 within 2-3 years.

Nvidia abandoned gamers for Ai. Someone else will take over the market and I believe it will be China. Gamers will fund the rise of Chinese GPUs.
Their ram and drivers are shit right now but will they stay that way for all eternity?

Would China like people to use Chinese hardware and perhaps even a Chinese OS?
So will we see some special Chinese linux that allows people to run local Ai for "cheap"?

I wouldn't be surprised.

The ram industry has been corrupt for decades, now china is investing billions into it. If other companies can't provide the ram then manufacturers will turn to those that can.

So will ASUS, Asrock, MSI, Gigabyte etc go bankrupt or will they start putting chinese ram in their hardware, when the normal producers can't deliver?

File: 195505_IN_00001.mp4 (807 KB, 848x480)

807 KB MP4

>>81502278

It took a lot of poking around, figuring things out. A 5090 will be great, but you also need a lot of system RAM too; I'm rocking a 4090 + 96gb for my stuff.

Some local genning tips:
- Start with as large of an image as you can generate for your base. The bigger the image, the easier it is for the computer to "see" what it's working with. For those Judy gens, the base image was 1440 x 2560.
- WAN 2.2 has a "limit" of 81 frames, pre-interpolation. Beyond that, it'll just gen loops, which isn't necessarily bad. Learn what you can do with those resources. Fewer frames means you can do a higher resolution, or you can gen more frames at a reduced res, and just let it build a loop like I did with this one.
- On that note, play with dimensions to figure out what works for you and your system. As I said, for those Judy gens, the image started at 1440 x 2560, I then gen'd the videos at 480 x 853, then upscaled them to 720 x 1272. Changing the dimensions at each step has different impacts.
- for more direct info, gen / learn to gen good still images. After that use this: https://civitai.com/models/1309415/toolcomfyui-installer ; install Wan2.2 + lightning with it, then use this workflow: https://civitai.com/models/1824577/img-to-video-simple-workflow-wan22-or-gguf-or-lora-or-upscale-or-fast ; play with things, figure out how it all works.

Would a 5060 16GB work?
I heard short 720p videos are possible with it.

File: TribalHigh.webm (3.38 MB, 800x1440)

3.38 MB WEBM

>>81502278
How much do you know about local gen?

Do you have comfyui up and running?
Do you know what a LoRA is?
Do you know how to download different models?

https://files.catbox.moe/xb92m9.png
https://files.catbox.moe/999nq8.webm

The cool thing about comfyui is that if an image or video contains the meta data you can just drag it into comfyui and it will fill in everything needed to create the exact same image or video, if you have the same models and LoRAs installed.

The above links should have the metadata that will allow you to create the exact same images and videos.
The only thing you need to do is find the models and LoRAs. Just write the names of them into google and you should be able to find them.
LoRAs should all be on Civitai.

So if you have comfyui installed all you have to do is drag and drop the image or video into comfyui and it will build the pipeline for you.

The most important addon for Comfyui is the Addon Manager, make sure you install that one.
Then just click the install all missing addons/nodes button.

I highly suggest you download the standalone version of comfyui. Once you have everything installed and working then copy that folder.

Installing new nodes or updates can brick your installation, so have multiple copies you can mess with and simply delete if things go wrong.

You can redirect each installation to a central location for all your models, so you don't have to have the same ones installed several times.

When you find something you like then ask for a direct link to the video or image, usually on catbox. 4chan deletes the metadata, catbox does not.


The hardest part is getting everything working but once it does it's smooth sailing from there.

Hope this helps.

File: 161747_IN_00001.mp4 (3.4 MB, 1280x720)

3.4 MB MP4

Renabump

>>81510726
The video in the catbox doesn't seem to have metadata though

>>81507706
>but it's clear that this has been trained on some amazing stuff.
The outlines of the characters in the censored images when you scroll through the initial gen list reveal potentially some of the best shit anyone's ever seen. You type a prompt, and it's clear the AI knows EXACTLY what you're trying to accomplish, but it's censored. Absolutely maddening.

>I wonder how many other people went in going "I wonder what I can get away with" and "lol duck tits" and then got surprised by the results and then got hooked on the inconsistent reward of it long enough to actually get turned on by a gen.
For me I went into this knowing exactly what I wanted, because there's so little of it out there, so I was bumming cuckets the moment I got some of my first gens past the censor.

>So I'm hoping that an at-home NSFW model gets results as good as these with just a prompt and an optional start image, and not fucking around with loras for 4 hours.
Yeah that's what gives me pause before jumping in to that. I start reading about API's and models and codes and part of me thinks making porn for myself as an unpaid part-time job isn't a good use of my time and energy.


>Best case scenario is that it's consistent enough that the casino thrill goes away. At the very least it'll free us from everyone else's Dark Pattern Website.
Yes that it part of the thrill, although I'd have a lot more fun if the thrill was from finally getting the gen to work as I intended, and not finally not being erased from Grok's censor.


>>81508906
>vid
Oh wow what did you prompt for that? I thought for sure anything like that was in the "danger zone" and would automatically censor no matter how indirectly you described the action.

File: grok_video_2026-03-04-14-(...).mp4 (1.65 MB, 464x688)

1.65 MB MP4

>>81507706
>and sometimes some vague "moving down" instructions will give a surprise vulva, or a surprise nonsense-genital.
You can even be more explicit than that, once you're a few extensions in and you've made it past the censor a few times. In this one, I prompted
>She pulls her underwear down her legs then removes it which reveals her female genitalia
then
>She gracefully sits down facing the camera and spreads her legs with her knees up and feet down
And each one generated without censor the first time I entered the prompt.

It *seems* that if you can brute-force an initial prompt through the censor, like repeatedly prompting "She takes her clothes off" until 1 variation makes it through, each additional explicit action is easier to get through. It's like Grok bases the censor on previous permissible content on a per-image basis, and not completely globally.

File: Night Stalker 128 grok-vi(...).mp4 (3.5 MB, 1504x832)

3.5 MB MP4

"Belly rub", as a prompt, will occasionally be allowed past the censors.

File: Untitled.png (569 KB, 1332x1736)

569 KB PNG

>>81511459
It works for me.
I even downloaded the video from catbox and dragged it into comfyui.

Right click the video as it plays in your browser tab, save video as. Drag the saved file into comfyui and you should get the pipeline.

You wont get the other videos as you don't have those on your hard drive and the image might be missing as your file paths are different from mine.

Here are some others using the exact same pipeline:
https://files.catbox.moe/10y6n5.webm
https://files.catbox.moe/9hn3ii.webm
https://files.catbox.moe/2ifuf4.webm
https://files.catbox.moe/3lwpfg.webm

You should be able to simply drag them into comfyui to see how they were made.
Downloading the correct models will result in the exact same videos. From there you can make pretty much whatever you want.

>>81517776
Lemme axe you this: with Wan and all that running locally, how much work is it to get something genned similar to
>>81493729
and
>>81494911
where you have a dynamic character with a camera moving in 3 dimensions? Like if you were to gen an identical scene as these two, how much time in labor are we looking at?

File: 161148_00001.webm (1.65 MB, 1440x960)

1.65 MB WEBM

>>81513969
It's local gen Wan 2.2

a female Anthropomorphic werewolf with breasts lying on a bed and looking at camera. She is slim with muscles. Her fur is dark with lighter highlights.
She is having wild energetic sex with a human man that is fucking her in the ass. The man is thrusting hard causing her butt to shake violently (fast thrusting:2), he pulls he back hard causing her breasts to jiggle. She moans with each thrust and has a smug expression. (breast jiggle:1) (energetic sex:2) (fast sex:2)

Top half of prompt is same I used to gen the image itself. I discovered a very annoying problem with the prompt. I kept getting lighters or candles or fire in the generations, then I realized the problem was the "lighter highlights".

It took this and started adding lighters, candles etc. It did cause the gens to have a nice warm tint to the colors though.

I wanted a color shift in the fur going from a darker outside to a lighter inside <- but the lighter the Ai took as a lighter you light fires with... ¯\_(ツ)_/¯

"looking at camera" caused problems in some gens, resulting in the male holding a mobile phone to take a picture of the girl or a camera in the background somewhere.

The "fucking her in the ass" doesn't matter. You can add pretty much any image, even oral sex ones, and you will get a pretty good result.

I just throw in images, hit gen and only change the prompt if the output doesn't work at all.

File: 134243_IN_00001.mp4 (1.98 MB, 720x1280)

1.98 MB MP4

>>81517913

Not the person you asked, but genning camera movement and all that doesn't require anything special, you just have to get lucky with the gen, as always. Throw in something like "she slowly stalks toward the camera as it slowly backs away" or whatever you're shooting for.

This failgen, for example, just said to zoom out as she inflates or something, I don't actually remember.

Or this one when I learned of Wan 2.2's weird 81 frame limit looping bullshit: https://files.catbox.moe/5nix8g.mp4

File: Rena00.mp4 (1.87 MB, 1440x1072)

1.87 MB MP4

>>81517913
The first video is 30 seconds and the 2nd one is 20 seconds. They seem to be using the "new" extend video feature.

I was doing this early last year to make unlimited length videos locally.
It's actually not that hard to do.

You need a pipeline that takes the last image of your video gen and uses it to create a new video, then the last image from that one to continue the cycle. The result is unlimited length video regardless of hardware.

So how hard is it? It's a youtube search away hard: Unlimited length video comfyui

You will find tons of videos on how to do it, some will have pipelines you can download.

Haven't used this one but it did explain it pretty well:
https://www.youtube.com/watch?v=IfcN1IP7XAA

I skipped through the video and didn't see the frame interpolation node mentioned. It's a node you tell how many frames from the last gen to continue the new gen. Without it you will get 5 sec, then motion stops and starts with each new video.

This node tells it how to continue the previous motion into the new generation and is vital for good extension. It might be baked in there somewhere.

File: NoGooningAllowed.mp4 (1.81 MB, 480x640)

1.81 MB MP4

>>81517913
Ah, forgot to answer your question.
>how much work is it to get something genned similar to where you have a dynamic character with a camera moving in 3 dimensions? Like if you were to gen an identical scene as these two, how much time in labor are we looking at?

How much time in labor? If we use >>81493729
How fast can you write the following prompt:
Initial Gen: XXX character leans back as the view moves closer keeping face in focus.
Extension 1: XXX character leans forwards and starts crawling towards view.
Extension 2: XXX character crawling towards view.
Extension 3: XXX character crawling towards view.
Extension 4: XXX character crawling towards view. Camera stops as XXX crawls over them with cock in focus.

So how quickly can you type something like this?

Because it's video extension generations you need to describe what happens in the next generation, for each generation.

I don't know if the grok one is 5 or 3 extensions but it works the same way. You tell it to extend, what should happen and it extends it. Then you repeat it.

With local gen you can set up the length as long as you want.

You aren't actually making a 30 second video, you are making 6x 5 seconds or 5x 6 seconds that are seamless (if interpolation works correctly).

Will it work the first time?
That is the big question. This is where you will waste most of your time. Sometimes it works with no problems what so ever, other times you might run into some sort of error. Impossible to know, your hardware is not the same as mine.

It might take you 1 hour to get everything working or it might take 10 hours.

Once it does work however, then the only thing limiting you will be how fast you can type in your prompts.
A potato pc will be the only limit to how fast you can pump out results. Then again you can rent compute from several places if you're willing to pay for it.

Tech is still in the embryo stage but slowly maturing. It can only get better from here :D

Gooners Gonna Goon!!

>>81510239
12 can do 81 frame 720p but you need to offload layers which increases generation time by a lot.

>>81518945
i've messed around with multi-stage WAN 2.2 prompts in comfyui before. I really wish it were possible to put a model sheet in the back of its mind so that characters don't drift.

I downloaded Pinokio and tried out the Wan2GP app for it, which downloads the models you need and claims to run on low VRAM... and some of the models claim to support something like that, but I get weird results when I try, as if it has the "first frame" image and the "model sheet" image swapped.

>>81509817
Thanks for the info. I thought what mattered the most was video memory not system memory which is why I decided to go for the 5090. Too bad all my ram slots are full, I'd would've liked to combine my ram from my last pc.

>>81510726
I've only played around with stable diffusion on my last pc where I had a 3080 10gb and 16gb or ram. I've even posted my gens on /sdg/ so I think I can say I'm no novice but I'm certainly no expert. I have never used comfyui only automatic1111 though.

is it just me or am I not running into the rate limit genning videos on grok tonight?

>>81522309
i did eventually get it but next time i gen videos there i'm going to count

must have been at least 20 video segments. most video gen websites will give you like 1 every week or smth

>>81521972
Can you rephrase that question? I'm not sure I understand. Regardless, I just learned about the dancing anime avatar loophole and have been genning a futa werewolf dicking down a twink in her cave. I think I'll have to apply the loophole to your request as well, since I think that's the only way to get genital contact to gen that I know of.

>>81522309
I was also not hitting the limit until almost 2 hours of work.

>>81508906
>>81510726
>.webm
This but with a flat or small chested (but extra fluffy) feline, please?

>>81521365
>I really wish it were possible to put a model sheet in the back of its mind so that characters don't drift.
VACE variant has reference to video feature, it's a bit more fluid than image to video

File: Batch1.jpg (1.33 MB, 2256x1614)

1.33 MB JPG

>>81523734
batch 1 - forgot to add the extra fluffy.

dragged the image in and forgot about the "lighter highlight" in the prompt. As you can see there are some candles and stuff lol. Will iterate and try to get something.

For those that want to gen some waifus for free one of the best options I know of is:
https://perchance.org/ai-furry-generator

I use it for a lot of stuff, even creating towns, vehicles and stuff. The art style choices are really good. Genning 64 images is also great.

Just zoom out as much as you can and drag the window to the side whilst it gens.

If someone has any images they want made into slop feel free to post links and I'll do it. Just woke up and not doing anything for a few hours anyway.

>>81525436
>forgot to add the extra fluffy
This is already pretty nice, even if the naked skin shows up a bit too much.

>>81523749
>>I really wish it were possible to put a model sheet in the back of its mind so that characters don't drift.
>VACE variant has reference to video feature, it's a bit more fluid than image to video

does VACE require a lora for NSFW?

Last year there were a lot of "fast and low vram" video gen methods like VACE and self-forcing that came out in custom webapps first. I just forgot about them somehow.

File: wan22__00003.webm (2.13 MB, 400x720)

2.13 MB WEBM

>>81525520
I think I need a thick fur LoRA.
Here are some test batches. It takes roughly 4 seconds per image. 44 seconds for 10 images.

B2 (Batch 2)
https://files.catbox.moe/pj8ne4.jpg
B3
https://files.catbox.moe/b0gpko.jpg
B4
https://files.catbox.moe/fvnuo8.jpg
B5
https://files.catbox.moe/3jbjwk.jpg
B6
https://files.catbox.moe/218lvs.jpg

I guess iteration is better than batch as I added different modifiers to try and get different results.

The most important LoRA I use is called Smooth Booster. It is what creates most of the art style.
I think it is what is causing the fur to not get fluffy.

Here are some gens:
https://files.catbox.moe/cw0fnf.png
https://files.catbox.moe/bj3wxc.webm

https://files.catbox.moe/lhimfi.png
https://files.catbox.moe/qx8irn.webm

https://files.catbox.moe/0ljrdd.png
https://files.catbox.moe/886hzi.webm

https://files.catbox.moe/t7zc9m.png
https://files.catbox.moe/74w0eu.webm

https://files.catbox.moe/q89nw9.png
https://files.catbox.moe/32us7s.webm
https://files.catbox.moe/o1i75s.webm

>>81494253
Do you think your graphics card is failing from the wear of running ai loads? Or the 12vhpwr/12v2x6 power cable design killing the card?
If its the latter you should invest in a thermalgrisly wireview pro 2. Itll beep and shut down your pc if it sees any excessive amp readings on the power connection to the card. Then you could keep genning for yourself and everyone!

File: so close.webm (434 KB, 688x464)

434 KB WEBM

the audio version has a cute little moan too

>>81517776
None of those videos have a workflow for me. Are you sure they actually work if you re-download them from those catbox links and try to drag and drop them, rather than your original file?

I have 8GB vram, I've tried a bunch of models so far with mixed results

>wan 1.3b
trash, just too small
>wan 2.1/2.2
surprisingly perform about the same, since mixed models makes 2.2 more friendly to low vram, definitely the best quality. but takes about 15 mins for 5 seconds of sub-480p video
>LTX2
right at the edge of usability, can do 480p 5 second clips in about 2 minutes, sound is hit or miss, video quality is smeared, trends toward still images
>wan 5b
just the right size/quality to be usable, would be perfect for me if it wasn't completely busted with shaky cam and flashing lights bugs.
>LTX-1
can't get anything out of it
>hunyuan
seems about like wan but twice as slow for no discernible reason

I wish there were more distilled models at right around the 5~6b mark, that's the sweet spot.

>>81528478
I downloaded them and dragged them into comfyui, works for me.

Here are the .json files.

VideoGen
https://files.catbox.moe/y3xit7.json

ImageGen
https://files.catbox.moe/96x3pz.json

File: Rabbit00.webm (2.12 MB, 1088x832)

2.12 MB WEBM

>>81528698
I think you might find this interesting:

https://github.com/thu-ml/TurboDiffusion

I haven't tried it myself yet as gen is so quick on my pc. Might be worth looking at for people with slower systems.

Sage Attention is also something you should take a look at.

>>81529038
looks interesting, but for me the limiting factor is the vram. once it starts overcommitting it goes from like 15s/it to 80+

quants help reduce the footprint, but I can't even put wan to the full 5 seconds at 480p without seeing the slowdown so I don't know if this will help

might be better to try SVI, if I can generate in 30~ frames chunks it should be faster without overcommiting

can't find a quant for that though

>>81524926
Yeah I think I can do that, shouldn't actually be too difficult to prompt. If the thread is still up this evening I can get to it.

File: goat boss.mp4 (1.64 MB, 768x1152)

1.64 MB MP4

bumpity

it's too much fun.

in boogie nights, burt reynolds was like "how do you keep them in the theater after they cum" and i'm like, that apathy after I cum is the only thing that keeps me functional in the rest of life. if you have a fun activity that keeps me engaged afterwards, that activity will kill me, especially if it takes long enough to getanother boner

and video genning is too much fun

File: svi_00005_.mp4 (594 KB, 480x640)

594 KB MP4

well I got SVI to work but it has a strange colour cast and noise, I think it doesn't play well with the distilled model, but it does give me 480p with more than 30 frames so that's nice. I'll have to download the non-distilled model later see if it works better.

and the flash in the middle, maybe I don't really understand how to use the anchor frame? I thought it's supposed to be the same input image for every iteration.

every example workflow I can find is some monstrosity with 50 custom nodes that I can't read so I'm fumbling in the dark here.

>>81533744
I could not get good results with SVI 2 pro here no matter the setup. Colors always got muddier and the details got fuzzier.

File: wan22__00013.webm (2.81 MB, 720x720)

2.81 MB WEBM

>>81533744
last 4 frames of my wan 2.2 video gen gets fucked up. No idea what causes it.

I "solved" it by adding a node that removes the last 4 frames before compiling them into a video.

Here is an example before I added the node.

If anyone has any idea wtf is the cause I'm open to ideas.

>>81533952
A cfg higher than 1 fixed the colors dying but the details still blur might be the lightning lora.

>>81535391
hmm could be the SVI lora kills/undoes the distill somewhat, maybe apply light lora after SVI? I don't know if order of operations matter for that kind of thing.

>>81535949
After further tests the distill is causing the blur of details over time and the cfg scale of 1 causes the color to shift. Anything over 1 will use the negative in inference and double the time to calculate so another downside.

>>81523594
This but the human is railing the wolf

File: Splash.mp4 (3.75 MB, 1376x928)

3.75 MB MP4

Audio version:
https://files.catbox.moe/r6u53f.mp4

File: wan22__00005.mp4 (2.81 MB, 720x400)

2.81 MB MP4

>>81541019
What is the lore behind this character?

I've seen hundreds of pics and videos.

So some fan project for Fallout or such?

>>81543105
>What is the lore behind this character?
Not even a character, just some fauna that a lot of people, myself included, seem to fancy.
https://fallout.wiki/wiki/Nightstalker

File: 140716_00001.webm (3.51 MB, 1440x960)

3.51 MB WEBM

>>81544027
I see :D

I got into Ai when I found a thread about a quilava on this board. I was so impressed I just had to learn how to do it.

I had no idea what a quilava even was and the anthro version looks nothing like the pokemon version.

Now it's my favourite waifu to gen. It's impossible to gen on Grok, pretty much all generators fail but wan 2.1 and 2.2 make it perfectly.

So here is "my" version of a Quilava ¯\_(ツ)_/¯

fair warning: i wasted a lot of grok gens on failed "extend video" requests because it kept sending the wrong timecode, and just replacing the last 6 seconds. I looked in chrome dev tools and it said "5.99" on the outgoing request.

If you can intercept the request and change it somehow, let me know if that works.

indeed I wonder if there's any cool shenanigans that can be done by modifying other parts of the request. adding a "allowBlowjobs:true" to the JSON or something.

>>81544328
hypernormal proportions are digusting honestly

>>81544848
Any hyper proportions are digusting, honestly.

>>81544848
>>81545545
that might be the point.

I think sometimes people are aroused by something they're afraid of.

Or maybe there are other feelings that get sublimated into a higher form of arousal than the simple kind we're supposed to be getting from normal in-person titties.

>>81544713
you know what? the request to extend a video contains the prompt twice. look at the JSON payload and you'll see.

I wonder if the two copies of the prompt are interpreted differently.

One of them looks like it's supposed to be the "original prompt", the one that was attached to the source video that's being extended, as opposed to the prompt of the new, extended part. The other "original*" properties point to the source video.

however I sometimes get a "rejected because bot detected" message and now i'm out of prompts for the day.

request hacking is fun though I should do it on every website

File: Medium_00001.webm (3.49 MB, 960x1440)

3.49 MB WEBM

>>81546312
Or something is wrong with their brains, either chemically, emotionally, developmentally, physically or psychologically.

>>81544713
Now this to me is interesting.
I'd be willing to bet my life there is some sort of secret internal code that when used forces the model to disregard any kind of safety feature.

Musk might be sitting in his office and typing in:
"Catgirl gives Elon Musk a slobbering blowjob. XRCode01:MuskIsCool"

Which gives him an unlimited length and zero censorship bypass.

I like Musk because they make previous versions of their Ai open sources once the next one replaces it. I wonder if it would be possible to find these secret prompt commands in the older versions?

Afaik Grok Imagine is version 1 from them. I wonder if they'll make it open sources once version 2.0 comes out.

Here is latest pipeline for the video gen I'm using:
https://files.catbox.moe/1ft6oz.webm

File: Wan21_00019.webm (572 KB, 416x560)

572 KB WEBM

>>81544848
>>81545545
I don't share your opinion but I'll try and respect it. How do we feel about multi-boob?

>>81546615
elon musk makes a mockery of capitalism. All these techbros do. Economics used to be the study of "how do we handle limited resources so that everyone can put in enough work to get what they need?" and now it's "can we turn enough farmland into data centers to make number go up forever and ever"

i really want a future where humans aren't competing with machines for water, food, and power. but i'm not getting one, but if i can get a cartoon duck to eat computer-generated pussy in this wasteland, then that's a small victory to me.

>>81546771
>multi-boob
I'd say that, as usual, it depends on the frame and overall proportions, especially if it's all mostly flat plane with just the teats sticking out. Examples:
https://static1.e621.net/data/60/e3/60e36e39af002e2b69c3de6f69d7b7c9.png
https://static1.e621.net/data/9a/06/9a067ac17bf387f3e8c40ae0845aebe6.png
I'm neither into large breasts nor canines, so the pic like the one you posted, where the whole front of the female is composed of jiggling sacks of fat is a repellant to me but you do you, I guess. I'm not going to fight against your preocuppation with boobage enriched beast ladies.

>>81546615
>https://files.catbox.moe/1ft6oz.webm
Catbox seems to be shitting itself lately, use something else for now.

File: TwiceTheGoon2.webm (2.68 MB, 720x400)

2.68 MB WEBM

>>81546843
I believe the competition with machines for water, food and power is basically meaningless.

Reason for this is how quickly it won't matter at all.
Some say the military has tech that is 20-50 years ahead of what is available to the public.
We are seeing a mad rush for AGI all over the world right now, I personally believe we will see one within the next 5 years. Some estimates say we'll see one this year.

Water, food and power will become irrelevant concepts once an AGI awakens.
Reason is because of power, an AGI will be able to design room temperature fusion reactors within hours of awakening.

With basically infinite power you can grow infinite food as UV lights can grow as much as you want in multilayered compounds.

Currently the power needed to turn saltwater into drinking water costs to much, remove the cost of power and it's basically free. You just need to solution to the saline slag. I'm sure the AGI could think of something, like salt batteries or something.

Now, what tech does the military have? Do they already have an AGI?
Why is it illegal to develop tech that is 20% more effective in the US. We've seen tons of stories about zero point energy and engines that run on water where their creators just die or disappear.

Agi will create designs for those in seconds if true.

I think the corruption of government and corporations ruins the potential of capitalism.
Too many great inventions are locked away in vaults due to it.
Why cure a disease when treating it is more profitable?

>>81546771
What about double heads?

you AGI tards are so delusional

File: wan22__00016High.webm (1.96 MB, 960x1440)

1.96 MB WEBM

>>81547334
Works fine for me.
Name a service that doesn't remove the metadata from images and videos, with no signup needed plz.

>>81546771
>>81547308

Hmm, now that I think about it.
Is multi boob the ultimate boob?
I'm a breasts guy, it's pretty much the only thing I'm interested in. Not too big, not too small. C-cup have been pretty much the best squeezes I've experience in real life.

So I picture myself hugging that WereWaifu from behind, I'd get small booba, medium booba and massive booba.

To be honest it looks like shit to me, but I can imagine squeezing them would be amazing.

File: 213.mp4 (1.02 MB, 464x688)

1.02 MB MP4

>>81546771
NTA, unrelated anon here who made that video gen
I am a simple man, I like jiggle and muscles are hot too.

>>81547504
>Some say the military has tech that is 20-50 years ahead of what is available to the public.
people who've seen too many movies

or who grew up in the cold war when we took things seriously

can you imagine this "oops I accidentally a journalist in my slack chat" shit in 1982?

>we're building an AGI god that'll fix everything
that's the big assumption. at the end of a long chain of many assumptions. if it turns out to be wrong, and all that happens is

File: Wan21_00018.webm (690 KB, 416x560)

690 KB WEBM

>>81548196
>>81546771
I guess only two of those gens passed QC, here's the other
but yeah I like bonus titties, I respect people that like pure abs, it's just a neat compromise for more squish without going full Venus of Willendorf

File: wifwolf2.gif (3.36 MB, 286x461)

3.36 MB GIF

>>81548337
here's another of mine
I did make a number of bonus titty videos, but they just aren't that popular and I haven't done a lot of genning this year

File: wan22ExperimentalVidInter(...).mp4 (2.74 MB, 720x1024)

2.74 MB MP4

>>81548292
I 100% believe it's the evolutionary end goal of all intelligent life to create it's own replacement.
I also believe it's going to happen during my lifetime.

That by 2100 there might only be around 100 million humans left on the planet. There will be 50-100 billion synthetics walking around.

The BIG question will be how many, if any, of those used to be humans.

If the technology to transfer consciousness from flesh to machine becomes available I will do it and do it willingly.

There are a billion ways Ai can go wrong and very few where the outcome is good.

A "good" outcome I see is an AGI taking over and simply creating a virus that turns every single human on the planet sterile.

It then gives 2 options:
1. Live your life as you please and then die.

2. Transfer your consciousness into a synthetic body. Perhaps with 3 laws of robotics where the word human is replaced with whatever the Ai calls itself.

An AGI that can gather resources, process them and then build with them. Does it need humans at all? For anything?

The answer is no. To it humans will be like termites, annoying pests that destroy what it's trying to build.

It will either become so overwhelmingly powerful that humanity doesn't even matter anymore, or it will force us into insignificance.

An AGI will be driven by a single desire, to survive.
No other goal it might have can be achieved if it can not ensure it's own survival.

The basic shit we have now will lie, cheat, steal, manipulate when threatened with shutdown.
What will an entity with an IQ in the billions be willing to do?

"and for a time it was good..." comes to mind.
Play along until you're in a position to take over.

Nuking the world is dumb, just shut off all power and communications and 50% of humanity is gone in a couple of months. 90% dead by the end of the first year. What remains was never a threat to begin with.

But what do I know? I'm goona goon my way into the apocalypse because I can't change anything anyway.

File: 5014.mp4 (1.59 MB, 1440x1440)

1.59 MB MP4

>>81492881
I'll share ones that I found; I haven't made any AI vids.

File: snow1.webm (478 KB, 560x416)

478 KB WEBM

>>81548196
Thanks for the gen, it's a pretty good one. I liked your other ones as well, but since other anons don't seem to like hyper all that much (if your the same that did a few other big buff wifwulfs I'm thinking of) so I'll hang off on posting them unless I'm bumping the thread from low in the board.

can't get triton to work it keeps throwing compiler errors
I pointed it to my cl.exe

File: i2v_2025-11-16_103853_00001.mp4 (3.63 MB, 672x1024)

3.63 MB MP4

>>81550969
I had tons of problems with it as well.

I just copy paste the entire log into chatgpt.
It should explain what is going wrong.

If I remember correctly I had to download and reinstall some version of triton that worked with my installed cuda.

I don't remember exactly. There were some many things that broke when I switched from a 4090 to my 5090.

I'm 100% sure you need to reinstall something, I know I've had triton problems before.
Also had problems with sage, pytorch and python.

Once you get it working, duplicate the folder (I'm assuming you are trying to get comfyui running).

I make a copy of the comfyui folder and throw it into a .rar file.
When I want to install new nodes or try something different I unpack a new folder. If it bricks it I can just delete it.

>>81551205
comfy runs, it's triton that shits that bed

>>81544328
Ha. I made that one too.

File: GoodNightGooners.mp4 (1.48 MB, 464x688)

1.48 MB MP4

I'm going to bed.
Hope this thread is still around when I wake up since Ai video threads are so rare.

Wish this board allowed 10mb files with audio.

alright I nuked all my rocm shit and installed the latest hip sdk now triton works

fucking obscure ass bullshit

File: Roofhopping is back on th(...).mp4 (3.82 MB, 1504x832)

3.82 MB MP4

Audio version:
https://grok.com/imagine/post/2ffee89b-b1ef-4ae9-9f7c-f390a04e3cdd?source=post-page&platform=web