Thread for AI Generated Videos made on open source/open weight models.Thread Rules:>Generations from models that are either available to run locally or have upcoming local releases planned only>No prominent glitches>No Spamming (No more than 3 gens with the same starting frame)>No political shitGenning initial images on non-local models and then animating locally is fine.see>>>/g/ldg for guides and technical supportWan 2.1/2.2 Guides:https://rentry.org/wan21kjguide (May run better on lower spec machines)https://rentry.org/wan22ldgguide (Current local model of choice)
>>29773899
>>29773899>>29773951I cannot imagine what sequence of life events would lead to someone watching Princess Mononoke and think: "Yes, it would be so hot if she farted into a diaper and also that diaper was a person."
>>29774002What part of NO ONLINE BOT SHIT don't you get?Here is the thread for people without functional braincells: >>29771235
>>29774043It's clearly an automated spammer. Just like the ones advertising models' instagrams.
>>29773871Nice OP. Anyone got more titfucks they can share?
>>29774073
>>29774073Also take this one
>>29774080Hotfreckles are nice
>>29774092Body freckles are criminally underrated
been using this...https://huggingface.co/Phr00t/WAN2.2-14B-Rapid-AllInOne/tree/main/Mega-v12
>>29774441Have you tried any others? You don't seem to be getting much motion.I use this one and get good results: https://civitai.com/models/1995784?modelVersionId=2260110I've also heard good things about the Dasiwa checkpoints.
>>29774567yeah, most of them feel the same because I barely give it a prompt, Im still learning stuff but I think my prompts are getting ignored. I fairly new I dont know if the problem is that checkpoint Im using or my workflow settings. I will try that checkpoint later, thanks.
>>29774600Smoothmix has decent prompt adherence, though Wan in general can be tough to get to do specific actions without loras sometimes.Detailed prompts definitely help, but it can be tough to find the sweet spot between under and over describing. My suggestion would be to look for vids you like on civit and see what they prompted. Start from there, and adjust.There's a workflow specifically for the smoothmix checkpoint if you suspect something might be wrong with your workflow. It's what I use, too.
>>29774763Looks like your diffusion isn't resolving based on all the fuzzy parts. You might need to increase your steps.
>>2977478828 steps wan 2.2, its just I'm asking it to do crazy shit prompts it doesnt wanna do because.. logic reasons.
>>29774815Plus some of my loras are cranked up to ridiculous settings to try and get the outcome i wanted
>>29774826Yeah, that'll cause issues. Are you running with any speedup loras, or a checkpoint with them built in? The high steps may be hurting more than helping. You might actually be able to get it to resolve better by using a speedup lora ONLY on the low noise side, and using low steps there. You can set your ksampler to have different numbers of steps without issue. (Eg: Set sampler 1 to 28 steps, end at step 14, then sampler 2 to 6 steps, start at step 3.) It's helped me in the past with removing fuzziness from 2d styled gens. Might help here, too.
>>29774977Ah to be able to lick all the way up the underside of that cock, to the balls, up the taint and dive one's tongue into that perfect asshole.
>>29775280dam bro
>>29774441Do you or anyone else here have the issue of phr00ts models constantly offloading? When I want to do batch gens, every 3 or 4 gens it offloads everythiing and has to take like 15 minutes to load. Its a great model but the model loading speed is abysmal.
>>29774075That poor fishie in the background… Lol.
>>29775368That's Flounder, you uncultured swine.
Request please can anyone do miss rachel thanks
>>29773871Good thread opI'm genning local with comfyui using chroma for t2i then wan2.2 for i2v, I really prefer more realistic looking stuff. Have any of you tried the longcat model? Seems really interesting for making longer vids as it samples a bunch of frames from the previous vid to extend the next meaning smooth transitions, workflow I've got tough you can't add loras which sucks.
>>29775547The longcat models apparently only work with kj nodes. I tried and kept getting errors. Just waiting for longcat and svi to be properly implemented into comfy (apparently there's going to be some custom nodes for them or something).For now, its either supervacetools or cranking the frames to 350 and praying not to oom (svi loras kinda work but for wan 2.1).
>>29775613The workflow I had it running on was using wanvideo nodes, not sure if they are kj or not
Is there a guide on how to generate img2video with comfyui using real voices and with lipsync?
>>29775986Not sure about a guide, but I'd look up infinitetalk. AFAIK, that's the best way to integrate audio into WAN with lipsync.
>>29775986Look for pixaroma on yt, ep.60 of his tutorial is on infinite talk, voice driven ai vid with sync
>>29776075Cutest RE girl
>>29775547I'm doing the same, also looking for a proper way to concatenate multiple videos without the jank. Oh well, I guess we're closer, but not there yet.
>>29775337I had the same problem using a workdlow I found on a sticky guide,what workflow are you using? I'm currently using something based onhttps://huggingface.co/Phr00t/WAN2.2-14B-Rapid-AllInOne/blob/main/Mega-v3/Rapid-AIO-Mega.jsonand it works just fine.
>>29776704I think I've fallen in love with fake pixels...
>>29776075got more RE girls like this
>>29776220>>29776223>>29776419>>29776469Can you make these locally? I use Wan 2.2 and can never got the blowjob motion or head grabbing to be as good
>>29776857Should've mentioned using ComfyUI for I2V
>>29776857yes just depends on what blowjob lora you use. there's like 5 different onesthis guy uses wan: >>29776075
>>29776857Lora strength, prompt, and fps can affect speed and motion. If I want fast/aggressive acts I will usually put "fucks her mouth like a sex toy" or something similar
>>29775188.. too much epilepsy
>>29776857I like the Ultimate Deepthroat Lora. Does a lot of the heavy lifting for me.>>29776819Only Rebecca for now. I'm still running this set of gens.
>>29776704More of this please
higgsfield the only one i get working
>>297774511, 3, 2, 4
>>29777457is this order of whos hottest?
>>29777465Yep, now spin them around, I want to rate their asses too :D
>>29777501ran out of credits...
>>29777520No worries =)
>>29777451What a waste of a gen
>>29777605You should have been aborted.
>>29777627is this your gen anon?what do you use to upscale?
>>29777666Not mine double trips, merely a reposter.
>>29777627 you can find hotter girls giving head in a gas station bathroom
>>29777694im still very early on in my comfyui journey, looking for ways to improve fidelity.
>>29777451What’s the work flow of making this? Program, model, ect
>>29777723Go to one of the ai local generals on /g and they'll steer you in the right direction. Very helpful bunch of chaps there.
>>29773967Probably being japanese would help in that
>>29777732ive been through the trenches of trying to use chatgpt to fix my comfyui issues, its been hell, now i have a decent install with triton, cuda, pytorch all playing together nicely though. Finally feel like i can spread out from fairly basic flows.
>>29777589>>29777627This is the local ai video general. There's a grok thread here: >>29775378
>>29773871I love these tittyfucking videos too but it's a shame that AI can't seem to grasp the cum part of it yet; majority of ones interpret it as the woman suddenly drooling heavily.
>>29778300some are better than others
Where does one go to browse locally generated no limits content?
>>29778785>no limits content?You're not fooling anyone you fucking nonce.
>>29778785civitai?after that i dunno
>>29776704>>29773897I need a name, a link, something I love this
kek
Can someone plz post an updated link to this folder :- https://gofile.io/d/vLl9HLEspecially the ghost grope videos(and more if any)
>>29779313That's walkanon's gofile, I think. I don't know that he posts his stuff anywhere else. I think it just gets updated when it gets updated.
>>29774600>>29774629this is very nice, no prompt at all. I just had to disable some loras that the workflow uses to supposedly speed up the process, but I found they just messed up the result and sisabling them din't really made the process slower, weird...
>>29780102Huh. Speedup loras built in? Are you using the "WAN 2.2 S. Workflow v2.0" workflow or the "Adv. WAN 2.2 S. Workflow" option? It sounds like you might be using the latter. If you're using the smoothmix checkpoints as well, I definitely recommend the former. It's probably better for the checkpoints you posted as well. The v2.0 workflow has been updated more recently. Most Wan 2.2 checkpoints have the speedup loras built in, so using them alongside those checkpoints will 100% fuck up your results. It's like using the same lora twice at full strength.
>>29780102Don't forget to set your CFG and stuff back up if you take out the speed up lora
>>29780246Im using https://civitai.com/models/1995784?modelVersionId=2260110and this workflow:https://civitai.com/models/1847730?modelVersionId=2091039
>>29780263Try switching to this workflow, it's the one I use and it's built specifically with that checkpoint in mind: https://civitai.com/models/1847730?modelVersionId=2264611>>29780257In this case you wouldn't want to, because the checkpoint has the lora built in. It's still designed for about 1.0 CFG.
Interesting, The provided workflow is very similar to the one I use, a bit streamlined. I loaded some NSFW Loras from here, I dont remember where I read it was reccomended If I wanted extra spicy stuffhttps://civitai.com/models/1307155?modelVersionId=2073605I didn't change anything else and this came out.
...and this is the result without using anything else, just with the workflow as is
bros i wish i never discovered runpod. spending too much money training 2.2 waifus.sidenote: shouldnt my 4090 be enough to do them locally? cant run it without getting oom errors.
>using a lora for an adult video game character>all the results end up being cunny>noticed it was taken down on civitai a while agohmm
>>297804074090 is fine, you probably aren't using optimized workflows, quant models, or tweaks like blockswapping
>>29774740What's with those things flying around
>>29773892yo thats the bees knees! what model did you use?
>>2978041i should've specified, I've only tried with AI toolkit. I tried getting musubi to run locally but cant get it set up. Whats the best for training 2.2 locally?
>>29780407Shit, you can do this with a 3070 and enough patience. The 4090 is the second best consumer card to use. You may just not have enough system RAM. I recall I'd occasionally get OOM errors when I tried to make especially large resolution, longer clips before I added more RAM to my pc.>>29780381The smoothmix checkpoint also has some nsfw loras baked in. You shouldn't need a generic NSFW lora, though you probably still want loras for specific sex acts. Also make sure to use lower strengths with this checkpoint. Start your loras at .3 strength before adjusting them, and be especially cautious with low noise loras when you're trying to get a 2D look, as they can give your gen fuzzy motion, or even shift it to a more 3d look. To oversimplify it: The high noise sampler determines the motion of your video, while the low noise sampler determines the style and details. Don't be afraid to only use a high noise lora, or to lower the strength of the low noise loras, if you're trying for a 2d style.
>>29780468I have 64 gigs of system ram but still nothing.
>>29780416Smoothmix checkpoint with the Golden Boy Anime style lora, as well as deepthroat/cumshot loras. Prompt: A goldenboystyle dbz style 2d animation of an angry blonde woman with large breasts giving a blowjob to a man with a girthy erect penis. She tilts her head as she bobs her head up and down on the penis. She pulls back and the penis trembles before releasing a spurt of white semen onto her ace a moment later. 90s anime aesthetic. Low framerate, vhs artifacting.
>>29774441this one's perfect
>>29780483Yeah, no clue then. I'd just try a different workflow.
this is happines to me
>>29776530Just his default workflow. I went back to the regular v versions as there's too many loras baked in that change the face in the mega version. Then again, he's released a ton more since I used it so might try it again.
>>29780617do yourself a favor and use >>29780282>>29774567as said in these comments
I kinda messed this one up with the prompt but you get the idea. portia doubleday
Is there somewhere you can dl celeb loras?
>>29781450This one would be nice as missionary.
>>29773871ok i got a 3080 earlier today, and was curious on if i can get decent enough renders with only 10gb of vram. if so, what models should i be looking for, and what settings are needed.
>>29781591You definitely can, though it might be slow. Not sure how many guides there are specific to 10gb vram, but search for Wan 2.2 8gb VRAM. There should be quantized models that will fit in your vram and you can find workflows designed for mid-to-lower end cards on civitAI.
Are there any phone apps that will produce uncensored stuff?
>>29781611I did download a wan2.2 model that sorta works on my GPU and I have swarmui, but it's sometimes janky and fails at the end whenever I try to render anything larger than 80 or so frames
>>29781636How much larger are you trying? 81 frames is the standard. You can kick it up to 113, but you'll start to run into issues with prompt adherence and it'll try to loop back to the start. Works for basic sex, but not anything more complex. You may also just need to use a lower resolution and upscale.
>>29777451Very nice, has a mind control vibe to it.
>>29781648i can sort of get renders with 81, but only a couple before my linux installation crashes. and they're quite jank, though i attribute that to user error/inexperience.
>>29781791Wan isn't really made to generate clips longer than 5 seconds.
>>29781803I'll probably see if there's any other models people usually use.
>>29779169Damn I wish this was the first post in an AI Deepthroat thread
>>29774043Any chance of more like this of cute girls waving their fingers no? I love these.
>>29775280couldve just said>to lack a spine
>>29773871
Needs more white pp in cute girlbutts
>>29780407There are exceptions, but you typically need to load entire models into your GPU's VRAM. 24GB seems to be the entry level for image / video gen and the only consumer cards to have that are the 3090 and the 5090.
I guess the spamming blacked pajeet is back
>>29782664OP removed thread rules
>>29778841NTA but shut the fuck up moral crusader. This isn't real you're not helping any of the vulnerable.
>>29782672There's site rules against spamming anon
>>29782664I hate that shit. Meanwhile the other board threads are basically dead. I miss when there was more creative types making crazy shit. CivitAI still has some heavy hitters though.
>>29782715People will gen what gets their rocks off, I don't think the batshit crazy stuff will have a huge audience kek
>>29782715You'd have thought with this guys obvious passion for blacked ai he would actually find a model that can animate his low tier gens, but alas the IQ just isn't there
>>29782740He literally is just doing it to troll
>>29782745I half-think that it's a deranged /pol/tard doing a false flag to get the "no blacked" rule put into the op, as insane as that sounds. btw they're not local gens
>>29782759Yeah but the no blacked rule does fuck all in vdg people can still post it there without repurcussions as long as it isn't shit tier flooding
>>29782769His stuff has been removed before from reports
>>29782740ranked 1 porn pose model creator on citivai, 4800 million views, 230m downloads
>>29782824Not a good generator evidently
>>29782824Then you don't need to flood threads on 4chan, streetshitter level iq
>>297828244800 million
>>29782824Don't see him on the leaderboards
>>29782824You sound like Chris Chan when he said Sonichu has 5 billion fans.
Uh oh, Stinky is back.
SAR I AM TO BE HAVING 4800 MILLION VIEWS
WE ARE TOP CIVITAI CEO
>>29782891Kek
>>29782871I gots this many viewers
Another instant classic from the top civitai creator!
>>29782937ok timmy
>>29782220No silly, they are not the one who would be doing the licking, I am
>>29782937I guess this is not your first dance with this guy?
>>29783075He spams the ai threads to the point that they had to add "No blacked" to the OP. Which in turn invited the racists in who start screeching whenever they see a black person, who in turn scare the actual contributors away. It's a circle of shit.It's a shame because there are some solid interracial gens I've seen in recent threads that aren't just cuckoldry fetish shit.
>>29783107The jannies don't act on reports of him flooding threads?
>>29783208It's inconsistent, and slow. Just report for flooding and carry on as normal.
Hope one dedicated spammer doesn't kill this thread. I liked the overall helpfulness and lack of fighting compared to the main vdg.
>>29783285Yep was chugging along nicely
>>29773871The perfect tits are ai tits
>>29780898source?
New thread >>29784274
>>29784278We're not a bump limit yet and you forgot to add back the No Blacked rule.
>>29784284we are not adding a no blacked rule
>>29784290Ok, enjoy 90% of the thread being this guys shit>>29783347>>29783416>>29783420
>>29784299theres already vdg for gatekeepers and people jumping up and down because they cant skip a thumbnail, chill
>>29784299He doesn't respect the no blacked rule in the threads that have it, and he's clearly not respecting the no spamming rule. So there's no point in adding the rule. The only thing it'll do is draw in the no lifers from pol.
>>29784306>skip a thumbnailBro, the last THIRD of this thread just got toasted by that guy in a couple of hours. It's not about skipping a thumbnail. And if anything this thread is the gatekeeping one, it was made to keep all the retarded grokfags from spamming their shit.
>>29784338and slapping no blacked in this thread is going to solve that? hes already breaking site rules by flooding threads, piss off back to /vdg/
>>29784299it already is?
>>29784321>global no racism outside of /pol/>ban blackedIn systems of online governance, conflicts often arise between localized regulatory authority and overarching normative principles. When an individual moderator or thread author attempts to impose a rule that excludes content or participation on the basis of race or ethnicity, such a rule directly engages the superior anti-discrimination framework that governs the wider platform. Within a hierarchical legal model, the global rule functions analogously to a constitutional or jus cogens norm: it establishes a universal prohibition on racial discrimination that subordinates all derivative or delegated norms. Consequently, any locally imposed prohibition founded upon racial criteria is rendered void ab initio, as its discriminatory character places it in direct contravention of the higher-order mandate of equality. This illustrates the principle that delegated regulatory power must operate intra vires—that is, within the bounds of the superior legal order—and that any exercise of authority ultra vires, or beyond those bounds, is legally ineffective. Hence, while sub-communities may legitimately determine thematic or procedural rules, they cannot derogate from the platform’s fundamental anti-racism commitment. The logic of this hierarchy can be expressed formally: let R represent a rule grounded in racial exclusion, and N the overarching norm of racial neutrality; if R ¬N, and N is hierarchically superior, then R is invalid and the conduct prohibited by R remains permitted—¬(¬C) C.
>>29784425>Wahhh you're rayciss if you don't want blackshit spammed in every thread
>>29784447reading comprehension 0/10
>>29784284my bad on the bump limit
>>29784459You're asserting that prohibiting blacked content in the OP contradicts the "No Racism" global rule and is therefore rendered invalid by it.It doesn't, got cry about it.
>>29773998You're gonna burn in hell you bastard, kek
>>29784481i see you're here because no one is posting blacked in /vdg/ need new grounds for your argumentative side?
>>29784508I'm here because /vdg/ got taken over by grokslop.
>>29784518right and the start of this thread was promising pajeetBBC is gonna spam, no need to nuke the spirit of the thread cos of the retard
i really wonder if forge/reforge can work well with sdv, i've been contemplating downloading them since i'm most familiar with their layouts. though i don't remember seeing any sort of prompt support which worries me.
>>29784359>and slapping no blacked in this thread is going to solve that?it quite literally works for the other thread so yes.>hes already breaking site rules by flooding threads,reporting him didn't work for the months of spam he did prior to the rule being added so clearly mods didn't think his stuff was spam beforehand.
>>29784364he's very clearly talking about future threads.
The problem isn't that his gens are blacked, it's that they're garbage. I made the thread, and I didn't want to rule out gens like this one just because one troll's blurry slop hits the same tags.
>>29785475amen anon
>>29785208I know Forge Neo has some Wan functionality, but I honestly feel like video is moving and improving too quickly to be locked down well into something rigid like Forge.
>>29782678this is a great shot. I'm also trying to get this but i keep getting messed up vids. what prompt did you use?
>>29782912lmfao this shit had me fucking dying. idk how anyone masturbates to this stuff when its comedy gold
>>29775547>wan2.2 for i2vis there something like this i can run with a 16gb card? i2v specifically
>>29785622I've been doing Wan 2.2 IV2 with a 16gb 4080. You should be fine as long as it's an nvidia card.
I don't know who needs to hear this, butYou don't have to restart ComfyUI to load new loras. You can just press R and it will re-import all models.
>>29782715Gym stuff is so niche it’s a good example of what people should use AI for, needs more sweat maybe.
>>29786147>nichegyms are real you know
>>29786171NTA, but they probably meant like, muscular women just pumping iron at the gym without doing anything explicit, which is pretty niche.
Some of these turned into dildos on walls and headstones. Happy accident, in my opinion.Order for characters after hex maniac is probably Smug Wendy, then Nami from One Piece in her Alabasta Arc outfit.After that, I'll need to pick some new characters and gen some new base images.
>>29785635>I've been doing Wan 2.2 IV2 with a 16gb 4080. You should be fine as long as it's an nvidia card.nice, i could only find the 14B model for i2v, is that correct or is there are 5B?
>>29786296the 5B sucks and there are less loras made for it. you should just use a quantized 14Bhttps://huggingface.co/QuantStack/Wan2.2-I2V-A14B-GGUF/tree/main
>>29786296There's also quantized versions of the smoothmix checkpoints that have been recommended here: https://civitai.com/models/2083588?modelVersionId=2358087When using a quantized model, you generally want to use the largest one that will fit entirely within your VRAM. If you're unfamiliar, quantized models are basically simplified to fit in a smaller space. You should be able to use a Q8_0 model (the biggest, most complete one) with your card and see little to no difference when compared to the unquantized model.Be aware that quantized models, usually in the gguf format, will require different nodes than non-quantized models. Some workflows will have the ability to switch between either, though some will expect one or the other. Non-quantized models usually have the ".safetensors" extension instead of the ".gguf" extension.
>>29774581Worth a bump. Kinda weird motion.
>>29785533check the metadata its embedded
Finally trying the dasiwa checkpoints. They feel like a tradeoff. I'm getting much less sameface warping compared to smoothmix, but the motion doesn't seem nearly as fast or willing to move things as far.
>>29787264Im trying to nail down decent movement in my i2v currently relying on multi loras, makes such a huge difference to get fast smooth movements.
