Thread #30207647 | Image & Video Expansion | Click to Play
File: 1769872495824397.webm (4 MB)
4 MB WEBM
Thread for AI generated videos.
Thread Rules:
>No prominent AI glitches
>No spamming
>No blacked
see
>>>/g/ldg for guides and technical support
>High end cards: https://rentry.org/wan21kjguide
https://rentry.org/wan22ldgguide
Previous: >>30190089
245 RepliesView Thread
>>
>>
>>
File: 1769895894567601.mp4 (2.5 MB)
2.5 MB MP4
>>
File: Smooth_SVI_Pro_00002_RESIZED.mp4 (3.5 MB)
3.5 MB MP4
>>
File: GtfsyIlWsAACBCC 1.mp4 (1.9 MB)
1.9 MB MP4
>>
>>
File: popping 2.3.mp4 (2.9 MB)
2.9 MB MP4
>>30207803
We require more giant goth tits. as heroes you see.
>>
File: 932889476.mp4 (3.9 MB)
3.9 MB MP4
>>
File: 1765226414922002.webm (3.9 MB)
3.9 MB WEBM
have no clue what comfy did, nvidia did, or what i fucked up with but my setup is completely borked. can't do any renders now rip
>>
File: 607704209.mp4 (3.9 MB)
3.9 MB MP4
>>30208610
Look at the console for errors? Could be something simple.
>>
>>
>>
>>30208610
Reason number xxx why I keep my gen system offline mainly. Got tired of starting it up to find my system vandalized not knowing if it was Windows Update of Comfy or what. At least now I can schedule it on my time. Will do a backup before and every 4 mos or so update my Win/Comfy/Drivers.
>>
>>
File: 1AnimateDiff_01279_x264_001.mp4 (1.6 MB)
1.6 MB MP4
>>30208823
>>
File: 340907065.mp4 (3.7 MB)
3.7 MB MP4
>>30208812
Windows updates shouldn't be an issue, it's basicaly just comfy/python and driver updates, but those shouldn't be happening automatically.
>>
>>
>>30208812
>>30208844
I made the mistake of updating Linux a few months ago. Lemme tell ya, windows bros, you got it easy.
Just added a second GPU last night. Windows (my main system, NSFW not allowed there) recognized it with exactly zero effort. Linux? 1.5 hours of copy-pasting sudo this and that from ChatGPT.
>>
>>
File: 752314929.mp4 (3.6 MB)
3.6 MB MP4
>>30208905
>NSFW not allowed there
Why?
>>30208907
Unfortunately it loses too much of the likeness of the character, otherwise it's a neat transition.
>>
File: 125956_00001.webm (704.3 KB)
704.3 KB WEBM
>>
>>30208928
Windows machine is...
>unsecured - wife (and kids as they get older) uses it on occasion.
>linked to onedrive in insidious ways - don't need AI porn showing up amongst our family photos.
>too important day to day - don't want to brick it screwing around with drivers etc
>bonus - at the outset I thought I might do some gens if my wife. Never did, but imagine her finding that shit.
When I got started with this I just happened to have a spare HDD in the machine, so I figured, heck, let's Linux. Because of how Windows and Linux format their drives, you can't even see the Linux drives in the same machine from Windows without software utilities. And it's set to go to Windows unless you know to F12 into the boot menu on startup.
I'm pretty meticulous about keeping things safe from my wife's knowledge ever since I accidentally saved a pic of a woman sitting on a huge dildo to my phone and she found it in our shared Dropbox camera backup folder about 10 years ago. Don't need to go through that again.
>>
>>
File: asuka_rei_beach_walk_extra_thick_04.mp4 (3.3 MB)
3.3 MB MP4
Anyone have any more gens like this, with thick waifus in a beachy setting?
>>
File: asuka_rei_beach_walk_extra_thick_05.mp4 (3.2 MB)
3.2 MB MP4
A bunch like these were made and posted a while ago, but I only saved these two.
>>
>>
File: AnimateDiff_00110.webm (3.8 MB)
3.8 MB WEBM
>>
File: widowmaker01_TR.webm (3.8 MB)
3.8 MB WEBM
>>
File: 03 -WanVideo2_2_I2V_00001.mp4 (466.9 KB)
466.9 KB MP4
Ok so I'm using https://rentry.org/wan22ldgguide guide with ldg_2_2_i2v work flow but I cant make the pic related to undress, the farthest I've got is her showing underboobs
>>
>>
>>
>>30209512
download this one: https://civitai.com/models/1944129/slop-bounce-wan-22-i2v and check the description for some prompt help. if you don't have a civitai account you can make one in 30 seconds with a throwaway email.
>>
>>30209543
>https://civitai.com/models/1944129/slop-bounce-wan-22-i2v
but i want to make more lewds than just bouncing
>>
>>
>>
>>30209543
>>30209641
Also do I download Lora and substitute my current lora? or do I download training data?
>>
>>30209400
With the normal WAN checkpoints you have to explicitly describe the undressing in the prompt. Not just something like "she undresses", more like "she takes her top off revealing her breasts and nipples".
I'd recommend using a checkpoint that is explicitly geared towards NSFW.
Be somewhat prepared to be disappointed. If WAN has to generate nipples or vaginas itself, it usually doesn't do a very good job imo
>>
>>
File: test.webm (3.8 MB)
3.8 MB WEBM
>>30209689
>If WAN has to generate nipples or vaginas itself, it usually doesn't do a very good job imo
Eh, they look fine to me.
>>
>>
>>
>>
>>30209826
https://civitai.com/models/1981116/dasiwa-wan-22-i2v-14b-or-lightspeed -or-safetensors?modelVersionId=2388 548
>>
>>
>>
File: 119811039.mp4 (1.5 MB)
1.5 MB MP4
>>
>>30209849
I see. DaSiWa is definitely better at this than standard WAN. Should maybe give some undressing gens a try again. Tried it back with Smoothmix v1 and it wasn't very good. I'm mostly doing anime stuff, so maybe it's just bad at that stuff.
>>
File: 155058_00001.webm (3.2 MB)
3.2 MB WEBM
>>
File: 174410_00001.webm (3.1 MB)
3.1 MB WEBM
>>
Deep thoughts incoming.
Yesterday we talked about the limits of SVI, and I said I preferred FLF and generating key frames with qwen then VACE for joining things up.
The SVI workflow I grabbed just repeats the anchor image across 6 gens. Anyone try changing up the anchor image for each gen segment?
>>
>>
File: wanvideo_00037.mp4 (3.2 MB)
3.2 MB MP4
>>
>>
>>30210303
More thinking about trying to introduce new elements in a controlled manner. Say my first frame is lacking an element I want to appear in the second gen of the chain, not the first. I can describe it, but if I can qwen it into the image used for the anchor on the second gen...
I'm going to test it later. We'll see.
>>
>>
>>
File: 305594952.mp4 (1.9 MB)
1.9 MB MP4
>>30210116
this one?
>>
File: WanVideo2_2_I2V_00025.mp4 (826.8 KB)
826.8 KB MP4
Ok so I've tinkered with wan2.2 a bit more and started to get results like pic rel, but how do I create videos that resemble >>30209719 or >>30207652, for now I'm using
Wan2_2-I2V-A14B-HIGH_fp8_e4m3fn_scaled_KJ
Wan2_2-I2V-A14B-LOW_fp8_e4m3fn_scaled_KJ
and I kinda plugged 2 loras together
Wan_2_2_I2V_A14B_HIGH_lightx2v_MoE_distill_lora_rank_64_bf16
bounce_test_HighNoise-000005
and some other node below
lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16
all plugged into VanVideo Block Swap then to WanVideo set LoRas etc.
I've got little idea what I'm doing, is there some type of YouTube video or something that goes step by step how to create I2V and T2V images using Kijai models? I just want to create some porn.
>>
>>
File: 004338 00001.mp4 (3.2 MB)
3.2 MB MP4
>>
>>
File: AnimateDiff_00002-4.mp4 (616.9 KB)
616.9 KB MP4
>>
File: AnimateDiff_00003-4.mp4 (1.6 MB)
1.6 MB MP4
>>
File: AnimateDiff_00007.mp4 (1008.8 KB)
1008.8 KB MP4
>>
File: 042114_00001.webm (448 KB)
448 KB WEBM
>>
>>
Does anyone use OneTrainer?
I downloaded it earlier and i'm going to train a lora for ZIT. Under the model settings i'm choosing to download the model instead of it relying on HF, It didn't seem to recognize the files saying that it only accepts diffusion models.
From what I've gathered it looks like I have to download every single file from ZITs HF one by one and then point that folder to OneTrainer, does that sound correct? I have to download them individually unless I choose to setup HF-CLI but I'd rather not use multiple programs if i don't need to. Annoying but whatever.
https://www.reddit.com/r/StableDiffusion/comments/1f93un3/onetrainer_f lux_training_setup_mystery_solved/
>>
>>
>>
File: Aqua_02322_CRF_00001.mp4 (3.6 MB)
3.6 MB MP4
Had this one laying around. I think I got the pieces out of order.
>>
>>
>>30208627
>>30208812
the only thing it shows is a popup halfway through that tells me to either update nvidia or install c+ v14 redist which is already there,
it was working well a couple weeks ago before i updated my drivers, it's definitely an nvidia issue
>>
for me it's BE+MG that's the ticket
>>
>>
>>
>>
>>
>>
>>30211863
https://civitai.com/models/1817671/wan-video-22?modelVersionId=2057465
https://civitai.com/models/1307155/wan-22-experimental-wan-general-nsf w-model?modelVersionId=2073605
https://civitai.com/models/1922973/f4c3spl4sh-cumshot-i2v-wan-22-video -lora-k3nk?modelVersionId=2178869
https://civitai.com/models/1585622/lightning-lora-massive-speed-up-for -wan21-wan22-made-by-lightx2v-kijai ?modelVersionId=2337890
https://civitai.com/models/1811313/wan-dr34ml4y-all-in-one-nsfw?modelV ersionId=2553151
>>
File: Lucia ZZZ 23346.mp4 (1.3 MB)
1.3 MB MP4
>>
>>
File: 1765078016286156.webm (3.8 MB)
3.8 MB WEBM
>>
File: midna.webm (3.8 MB)
3.8 MB WEBM
>>
File: Dva Brt beach 4mb.webm (3.5 MB)
3.5 MB WEBM
For some reason character refuses to lipsync the word Beach/ Bitch. "Your ass is myne, bitch"
The voice is generated but the character does not move her lips.
4 of 5 gens no lips move. 1 says "Your ass is at my beach"
>>
>>
>>30210949
There isn't a whole lot of AI-VR options out right now. Maybe in the future. It's a pain just to gen a 6 min movie with AI without losing too much detail. I suppose if you try to re-upscale the image every gen it could work but would be a PIA even with character loras. With WAN the rule is after 3 extensions the details start deteriorating pretty bad. There are other programs that do a lot better with longer gens but they are pricey and SFW only. Disney probably has something that does full length movies by now.
>>
>>30210949
>>30212644
i mean you can, there are 360 vr loras
you'll be pumping out low resolution which ruins the point of vr and 5 seconds clips unless you have a data center
>>
File: wanvideo_00012.webm (3.8 MB)
3.8 MB WEBM
>>
>>
>>
>>
>>
>>
File: Zelda_02337_CRF.mp4 (3.5 MB)
3.5 MB MP4
Ok, I admit it was nice to set up a themed SVI gen, toss a few similar images at it, and then find all this treasure when I came back later.
>>
File: 18_01137_CRF.mp4 (3.7 MB)
3.7 MB MP4
>>
File: Kim_Possible_01199_SVI_CRF.mp4 (3.9 MB)
3.9 MB MP4
Wonder how much trouble it'll be to make this loop automatically. Just have to add a regular Wan2.2 FLF to the end and then if I'm being honest I'll want to VACE that as well.
>>
File: charlotte fire emblem.webm (3.9 MB)
3.9 MB WEBM
>>
>>
>>
>>
>>
File: tracer.webm (3.8 MB)
3.8 MB WEBM
>>
File: upscaled_00017.webm (3.8 MB)
3.8 MB WEBM
>>
>>
File: interpolated_00001.mp4 (2.9 MB)
2.9 MB MP4
>>
>>
>>
>>
File: outputraw_00012.webm (3.7 MB)
3.7 MB WEBM
>>
>>
File: Screenshot from 2026-02-04 11-43-03.gif (28.1 KB)
28.1 KB GIF
Here's a fun one for you. I've gotten my 4060 installed and working alongside of the 5060ti. Using ComfyUI-MultiGPU to offload the CLIP to the 4060 during Wan2.2 gens made...
That's right, no real difference.
From my reading, this was kind of the result I expected. My real use for the dual GPU is probably going to be to have 2 instances of ComfyUI running so I can set video gens to run on the bigger card and then fart around in Stable Diffusion etc on the smaller. I'll live with this for a while, make sure I'm happy with it and I have all the bugs worked out, with the end goal being to swap the 4060 for whatever 24GB+ card I can get ahold of without selling a kidney.
>>
>>30215281
Pretty interesting.
I'm looking into running a ComfyUI instance on two of my PCs for the exact same reason. 5090 PC will handle video stuff, while I can use my 7900xtx PC to try to squeeze some decent base images out of SDXL. AMD is complete ass at video generation, but stable diffusion usually works well enough. Will have to look into netdist for that, but that looks simple enough.
>>
>>30215307
I was able to mix cards on one machine because they're both NVIDIA, and honestly the 5060ti was what I replaced the 4060 with so I just had it on hand. I think in your case you're definitely stuck with separate machines, and my thought had been to go that route with a laptop sitting next to the desktop until I realized I had an open PCIe slot.
>>
>>30215342
I know. I already have both PCs, so I'm fine with going with separate machines.
I don't know much about multi GPU on ComfyUI, but would it be possible to have one GPU run the high pass and clip for WAN2.2 while the other one handles the low pass and vae? Maybe even in a way that allow them to run simultaneously when you're generating more than one video with two instances with one picking up the persisted latents from the other.
>>
>>
>>30215404
I think yes, but with some possible caveats. The MultiGPU nodes let you pin models to whichever card you want, so you could in theory have everything set up to balance that use and then models wouldn't have to unload as the gen progresses. You'll have to have a big chunk of VRAM headroom somewhere for the actual frames to be computed, though. Depending on card sizes, it may be better to have the models all on one card and the other card wide open for compute. I kind of think, if I get a 3090 for instance, the 5060ti would be faster on the compute, and a 24gb card would just about hold both the high and low Q6 ggufs.
One of the issues with all of this is the speed at which things get passed from card to card. ChatGPT assures me, and I've been able to verify, that even though my second slot is only PCIe x4, it's fast enough.
The reason I didn't see a speedup with the CLIP on the 4060 is that it loads once at the start of a gen then fucks off, and VAE is small potatoes.
I think this all a long winded way of getting around to saying, for me, if I had the 3090 and the 5060ti, I think the fastest result would be from letting all models load in and out of the 3090 (CLIP, VAE, High and Low, all Loras), and reserving the entire 5060ti for compute with it's faster cores.
This is all spitballing, though. I just barely got this multi-gpu going so far.
>>
File: PikaPooper.mp4 (3.4 MB)
3.4 MB MP4
Sound
>>
File: PikaHole.mp4 (1.2 MB)
1.2 MB MP4
Sound .
>>
File: FireTits.mp4 (2.5 MB)
2.5 MB MP4
Sound..
>>
File: FuckingKermit.mp4 (2.2 MB)
2.2 MB MP4
Sound. .
>>
File: TooMuchPP.mp4 (3.4 MB)
3.4 MB MP4
Sound
>>
>>30215561
>>30215568
>>30215573
>>30215585
Raising the bar on us here.
>>
>>
>>
>>
>>
>>
>>
File: rando_01032_SVI_CRF.mp4 (3.6 MB)
3.6 MB MP4
Some more SVI laziness out of old pictures. I think I'm coming around to this. If I could wire up a little "reasonable" randomness in the prompts I might could be happy with this.
>>
File: saber_02248_SVI_CRF.mp4 (3.3 MB)
3.3 MB MP4
>>
File: 204553 00001.mp4 (3.2 MB)
3.2 MB MP4
>>30215947
love you too
>>30216022
You'd lose out against some chad with 21 GeForce GT 710s, though.
>>
>>
>>
File: Ladybug_02348_Mega_Loop_CRF.mp4 (3.7 MB)
3.7 MB MP4
Some of the stuff I was using to test getting the 4060 running. It was all batches of FLF loops, and honestly I haven't even looked at all of these clips. I just interpolated the whole folder and slapped them together.
>>
File: Kim_Possible_02342_Mega_Loop_CRF.mp4 (3.5 MB)
3.5 MB MP4
>>
File: Aqua_02349_Mega_Loop_CRF.mp4 (3.7 MB)
3.7 MB MP4
Alright, that's gotta be it from me for today. Going to slap something easy into SVI again and walk away.
>>
>>
File: 211317 00001.mp4 (3.9 MB)
3.9 MB MP4
>>30216035
I have yet to come up with a decent idea, that actually makes use of SVI. More outlandish stuff doesn't seem to work, and 28 seconds of exactly the same thing doesn't really justify the increased generation time for me.
Really wanted to give redoing this gen a try, but SVI fails at even generating the clip of her showing her teeth.
I'll probably settle for some ruined orgasm with POT stuff, as that requires some more length to work.
>>
>>30216208
I think randomizing some of the little details like facial expressions may be about as far as I take it. I see they're working on flf for SVI so I probably won't bother setting up a loop myself, but yeah, I think 20 ~ 30 seconds of a simple scene is probably it.
If they do give us some decent flf with it, though, now there's a game changer for me.
>>
>>
>>
File: 213506 00001.mp4 (3.9 MB)
3.9 MB MP4
>>30216093
The voice generation part is really easy to do. Used to use chatterbox for it, now with Qwen3-TTS it's even easier and better.
There's no decent solution to lip syncing though. Used Infinitetalk, but that really degrades the quality and changes details a lot, even when doing v2v. Here's the original gen. Movement is a lot better, and it's detailed enough that you can make out her bush peeking out of her pants.
>>
File: 215307_UPIN_00001.mp4 (3.8 MB)
3.8 MB MP4
>>
>>
File: 1750871765901567.mp4 (1.3 MB)
1.3 MB MP4
>>
>>
File: nurse_joy2_Interp.mp4 (3.5 MB)
3.5 MB MP4
>>30215149
Begging your pardon, sir. I think you missed a few frames.
>>
>>
File: i2v-RIFE49_00008.webm (3.8 MB)
3.8 MB WEBM
>>
>>
>>
>>
>>
File: muscular-women-low-samples_00027.mp4 (1.4 MB)
1.4 MB MP4
I'm in heaven
>>
File: 119885050.mp4 (1.6 MB)
1.6 MB MP4
>>
>>
File: 119852808.mp4 (3.8 MB)
3.8 MB MP4
>>
File: 119864825.mp4 (2.7 MB)
2.7 MB MP4
>>
>>
>>
File: 001856 00001.mp4 (3.5 MB)
3.5 MB MP4
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: vibe-video-cml8coge.webm (1.3 MB)
1.3 MB WEBM
>>
File: 0241_00001.mp4 (2.8 MB)
2.8 MB MP4
>>
File: HelpMe.mp4 (2.2 MB)
2.2 MB MP4
>>
>>
File: DontTalk.mp4 (2.4 MB)
2.4 MB MP4
>>
File: ebony112.webm (2.9 MB)
2.9 MB WEBM
>>30212364
>Real futa with pussy and dick
ew
>>
>>
File: 170856_00001.mp4 (2.5 MB)
2.5 MB MP4
>>
File: reimu2.webm (3.9 MB)
3.9 MB WEBM
>>
File: 1769974031834000.mp4 (383 KB)
383 KB MP4
i am again begging. PLEADING. for gens of the Folgers incest commercial. this was a peak moment for my kind of freak and it deserves to be remembered not as it was, but by what it inspired.
no requests. no position. just anything at all. the best part of waking up is just a single cup
>>
File: video4e48809a-68ad-4e4d-a1a9-05dedefde09a.mp4 (474.3 KB)
474.3 KB MP4
>>30218835
Yeah, last one so I can then refill credits hopefully, I'll refer to some other internet board though which seems allowed.
>>
>>
>>
>>
>>
File: Data Transfer.mp4 (3.4 MB)
3.4 MB MP4
Sound
>>
File: MirandaButtfuckingProtocol.mp4 (912.4 KB)
912.4 KB MP4
Sound.
>>
File: NewSkins1.mp4 (3.3 MB)
3.3 MB MP4
Sound .
>>
File: NewSkins2.mp4 (3.8 MB)
3.8 MB MP4
Sound..
>>
>>
File: NewOverwatchCharacter.mp4 (3.1 MB)
3.1 MB MP4
Sound
>>
>>
>>
>>
>>
>>
>>
>>
>>30219188
AFAIK you should be able to run WAN, as there's even a ComfyUI version for AMD. Fairly likely that it won't run decently at all. You'll probably have to make due with lower resolutions and longer generation times.
>>
>>
>>
>>
File: i2v-RIFE49_00006.webm (3.8 MB)
3.8 MB WEBM
>>
File: refueling2.webm (3.9 MB)
3.9 MB WEBM
>>
>>
File: 6_00001.webm (3.8 MB)
3.8 MB WEBM
>>
>>
>>
>>
File: 01-FIRST-hi14_00046.mp4 (1.4 MB)
1.4 MB MP4
jesus fucking christ
>>
>>
>>
File: aaea87f3f3fe4cad8a3b88ec807c6dea.mp4 (3.3 MB)
3.3 MB MP4
>>
File: e5a03731cfca4a789c95521b00055b9b.mp4 (3.3 MB)
3.3 MB MP4
any idea to get better hip thrusts?
they always move mostly backwards and it makes the penis come off half the time
>>
File: 2435a383a6bf40cbac4ede229753812c.mp4 (3.4 MB)
3.4 MB MP4
>>
>>
File: 7cead8980bde4b0e905a2555817e2550.mp4 (3.6 MB)
3.6 MB MP4
>>
>>
File: Pt1-ComfyUI_00551__x264_004.mp4 (3.7 MB)
3.7 MB MP4
>>30220528
First 10 seconds is ManandLivingCockSleeve lora, last 5 seconds is cocksleeve lora with FastMove Model. Cocksleeve and DrmLay lora does hip thrusts very well.
>>
>>30220528
Whenever it starts removing the penis I found you need to describe the women better like "both women are futa with penis and testicles attached to their crotch / woman on left side of frame has penis and testicles attached to her crotch, woman on right side of frame is female" or something similar
>>
>>
>>30220893
There's a guy who used to spam these threads for months with literal hundreds of bizarre low quality blacked gens each time and insult anyone who told him to pipe down. So the rule was made, something he ignored, which caused him to get banned like 3 or 4 times in a row (reporting him for spam only started working after the rule was made), which finally made him stop. The rule is still there just to prevent him from ruining the threads again.
>>
>>
>>
>>
File: 20_39_47_raw.webm (3.6 MB)
3.6 MB WEBM
>>
File: vid_00149.webm (3.8 MB)
3.8 MB WEBM
>>
>>
File: missionary.mp4 (1.1 MB)
1.1 MB MP4
>>
File: missionaryordoggy.mp4 (1.2 MB)
1.2 MB MP4