If anyone could please generate her being fucked, hair pulled and doggie, or, anything with cum, anything sluty, or BBC. I will be forever grateful.Any workflows for me doing it myself will help too
bump
Still bumping =(
>>20415765I hear this place is kinda expensive.
>>20416356lol kind of a wild generation.. lol
>>20415797
>>20416416Thank you!I would to have your workflow so i could generate my own
>>204164233 gens all looked like this. pussy balls lmao
>>20416495I am fighting for my life to generate my own. So I will take it! lol My focus is her and being fucked
>>20416517I mean what do you want to see exactly? Cowgirl/missionary/doggy? Each one is its own workflow and requires tweaking so if you say all ...she gets a dick and sucks herself off
>>20416551You know I would say all lol. If I had to pick I would say cowgirl and Missionary. I like rough sex so I have been trying to get a face down ass up, a doggie with hair pulling, and gangbang bj LORAs to work but it just creates a clurster of blurry body parts
>>20416568Some loras do not play well with others or with some models. If SmoothMix does not like the lora I will go with the default Wan 2.2 model. It happens sometimes.
>>20416568Think I saw a roughsex lora on Civi (with hair pulling) but didn't grab it so will have to try and brute force prompt it.
>>20416572Yeah, i got two loras to work now.Doggie Style: https://files.catbox.moe/2m3fr3.jsonFacial:https://files.catbox.moe/qhb2vw.jsonI want to explore other models and more complex workflows but I dont think I am remotely close to grasping that without using other peoples workflows and then just then downloading the necessary VAE, Unet, ETC.I also want to have a pic generation workflow so I can better setup my IMG to VID but couldnt get that to work either. That is why I came to 4chan
>>20416575Yeah they do have onewhich I got the workflow from the img on the lora page and it wouldnt generate anything close to what examples it was showing.Like I am frustrated so I am thinking does everyone deal with this or am I just retarded.
>>20416595I have never used anybody else's workflows. 9/10 they have some tricked out crap that breaks comfy or you need to install uber special nodes for. I just use the default template and add whatever loras i need. Might not be as "polished" as some workflows.. but works most of the time lol
>>20416637Could I get that workflow just play around with? I use grok for recommendations but everything it recommends makes it worst
>>20416423Here's a couple doggy. usually I extend my gens by got a buncha requests on the burner.
>>20416663Crap that was the wrong one..my bad.
>>20416664
>>20416667
>>20416664This is perfect! Could I get this one pretty please!
>>20416667and the cowgirl. I go a little over the top but thats my style
>>20416672These are perfect....
>>20416675You mean the workflow? Lemme see but I thought 4chan doesnt allow metadata
>>20416682I appreciate it. its hot but I think I am more in the realistic and slutty stuff. More dicks, big dicks, rough
>>20416689>https://files.catbox.moeI use this website. its worked on the workflow I got from someone else on 4chan. I think you can group them up.God bless you. I am forever greatful
>>20416687The workflow is barebone super basic. Wan 2.2 i2v template with below nodes/loras:SmoothMix - base modellorasNSFW 0.5igoon Blink-Front Doggy 0.4KSamplercfg 1.5 hicfg 1.1 lo
>>20416722Prompt was:The video begins with a close-up of a woman. The video then jumpcuts to the same woman now having sex in doggystyle position. She is lying on her stomach, facing forward, with her head turned slightly to the side looking at the camera. the woman has clenched fists. the man is passionately thrusting his penis into the womans anus doggy style. he fucks her like a cocksleeve. she cries out in pain.
>>20416730Honestly not even sure where to turn on the Metadata. used to be in VideohelperSutie but now with Ksampler thats all gone.
>>20416722Is Wan 2.2 i2v the one with the white in the templet section?and I am trying to not annoy you so you keep working with me.SmoothMix - base model what node is that go into?I am sure you saw this but this is how I post .json on 4chan. I think it multiple .jsons at once.
>>20416738>Yeah, I am retarded, so I dont understand. I think I am just wanting the nodes so I dont get any of that wrong.I will ask Grok about metadata and stuff
>>20416568>I like rough sex so I have been trying to get a face down ass up, a doggie with hair pulling, and gangbang bj LORAs to work but it just creates a clurster of blurry body partsThis was done with no loras. Just generate an end image in Qwen image edit using its NSFW model, you can find it on huggingface. Start frame - endframe templates generally work better than loras, just prompt in how the gen got there. It already knows what the endframe is so you don't need to be super detailed as long as the movement makes sense.
>>20416738>Is Wan 2.2 i2v the one with the white knight in the templet section?
>>20416752Just go on comfy, click browse templates and look for the Wan2.2 i2v.. think there is only one. Not the SVI or whatever else just i2v. if you don't have it yet you will need to download the SmoothMix model as well.. thats on Civi. the Igoon lora you gotta get on the CiviArchive page as Civi banned it.
>>20416759Any chance you can try one with her?
>>20416765Yah the workflow wouldn't help you anyways since the Igoon lora (which pretty much is what makes the vid) is probably not gonna be found. Gimme a sec to dig. Here is all the Igoon stuffs, you will have to find the Hi and Lo. For this vid I used .....igoon Blink-Front Doggyhttps://huggingface.co/june19925/wan/tree/main
>>20416765Yeah, I think its the white knight one. Grok says it makes me download a compact verison for some reason. The only nodes are:
>>20416775My gen puter is offline so can't walk ya through it but if you go to browse templates, sort by video, look for Wan 2.2 14b i2v... unless they nerfed the shit outta it cause never heard of Wan 2.2-Lite
>>20416775https://gofile.io/d/bVswFQthat's what you're looking for, just drag and drop it into a new comfy tab.
>>20416799>https://gofile.io/d/bVswFQAwesome! What lora do I change if any? Does this allow me to put igoon files somewhere or do I need to add nodes?
>>20416722This is where I am now. I have everything downloaded
>>20416799Here I am with this
>>20416868I would replace diffusion model with SmoothMix but thats optional. Just add Iblink lora and NSFW lora after the lightning and you should be up and runnin.
>>20416925Does this look right?
>>20416925And I fixed the prompt
>>20416949Do I need to change the Chinese negative prompt?
>>20416925Or should I be on the other workflow?
>>20416944Not quite. For each lora you need the HI and LOW version. LOW gets plugged into SmoothMix -LOW, HI gets plugged into SmoothMixLOW. Don't mind all my orphaned loras and spaghetti.. gotta a lotta requests that use different loras but here is the basic concept.
>>20416963I just delete the chinese stuff and add my own in. usually not necessary unless it starts giving you lot of crap you dont want but typically if you put it in the negative it will still show up lol
>>20416925This is how I have the other workflow setup
>>20416768Bump
>>20416963Still working on stuff. where does the connections go after stomach buldge?
>>20417005Okay dude I cant explain it any better so you get the John Madden. Hi goes to Hi, low goes to low. You HAVE to have both. You can move the nodes they don't have to stay in the little boxes.The Stomach bulge lora, ignore.. you don't need it and honestly it doesn't work with the Igoon stuff, you would need that for a different gen.
>>20417019Honestly at this point I would swear you are trolling me. This stuff is all over reddit and youtube bruhhttps://www.reddit.com/r/comfyui/comments/1mpc71h/where_do_i_place_a_lora_in_wan_22_i2v_workflow/https://www.reddit.com/r/comfyui/comments/1mbmscq/wan22_workflows_demos_guide_and_tips/https://www.youtube.com/watch?v=CfdyO2ikv88
>>20417019This is where I am at. I meant the top right connections with stomach buldge. does that need to connect somewhere
>>20417034Sorry, I was creating the loras. I couldnt get LoadLoraOnlyModel to work so I went IC
>>20417034This is where I am at currently
>>20417049
>>20417043Oops connected the lighting to nsfw. Just dont know what the right of stomach buldge goes to
>>20417034Now here
>>20417034It came out a blurry mess... ='(Please sir, I am not trolling...
>>20417034
Maybe scroll up and actually read what was typed dude.The Stomach bulge lora, ignore.. you don't need it and honestly it doesn't work with the Igoon stuff, you would need that for a different gen.
>>20417085Should I just bypass all the loras when I am specifically doing a doggy prompt?I didnt download the stomach buldge, I just though maybe what its connected to is something I need.Like I was saying I am not trying to frustrate you. Thank you for helping. I am retarded and trying to make this work
>>20417088I dont know why its not working. this is normally the results I get.
>>20417085This is my .jsonhttps://files.catbox.moe/a7t3fr.json
>>20417091Drop your CFG down to 1.5 or lower on HI no more than 1.1 on low. Also there are different versions of SmoothMix if you got the one with LightX baked in, bypass the LightX lora and go connect straight into the NSFW loras. Lastly, sometimes it takes a few runs. If it doesnt work with smoothmix, go to Wan 2.2 base model (you will need LightX for that model). Keep all other settings the same. Also, don't forget to change the seed until you get a gen that looks close to right.
>>20417107and yes, delete the Stomach Bulge lora.. its for stomach bulge so you wont be using it or need it.. thats for my crazy ass CockSleeve gens
>>20417107Okay. and thank you again for working with me. Do I need to bypass the missionary, cowgirl, and back doggie if I am doing doggie front or is those okay to keep active alwaysI bypassed the light2v and got all blue but there is shapes now.. lol
>>20417107Yeah, ran 4 times and its all just blurry messes.I will paypal you money if you get me through this. I can hope on a discord call or whatever works for you
>>20417126ALL the Igoon Loras are very very specific and basically generate an AI from scratch with her face in a certain position. I would ONLY use NSFW and ONE igoon lora for what you want (doggie/missionary, etc). Just ONE. the connect that to the model sampler and hit run. Thats it.
>>20417138Idk man, I dont want to waste your time anymore... I see i had different load clip that yours so I am going to try that.Again I appreciate the effort.I am pretty lost why this isnt working for me
>>20417171Sorry couldn't get it running. Im off to eat and chill for rest of the nite. May be back lurkin later if you get stuck.
>>20417176Yeah, load clip still gives me a blurry mess. idk what else to do. If you have anymore suggestions when you come back than that would be awesome. Attached is a screenshot where I am leaving offEnjoy your night
>>20417183First word in your prompt should be Smoothmixreallism. Also try Wan 2.2 base model instead of Smoothmix to see if it works with that.
>>20417183Not who you have been talking to. For real just use Remix 3, it has everything you need baked in! Worry about gguf and Smooth later. Much later.https://civitai.com/models/2003153/wan22-remix-t2vandi2v
>>20417191>I'll try tomorrow. I will be active on here. I need to go to bed
>>20417195What would remix replace? Smoothmixreallism? I can try that. thank you very much for the suggestion. I have been trying very hard to get good at this stuff and been lurking on 4chan. I will be on throughout the day tomorrow since I work from home
>>20417212Yes, it would replace smoothmix, The problem with smooth and the gguf models is they are heavily lora dependant for nsfw stuff, this can be a pain in the ass when using more than one as you have to juggle the strengths to avoid them conflicting. Remix has all the basic nsfw stuff built in, You will only need loras for very specific things with it. In other words it's beginner friendly.
>>20417212Should I use the same workflow or is there one you can give me or suggest?If I use the same workflow, should I change anything to get remix to work?Ill try both tomorrow.
>>20417285same workflow at least for gens made from scratch, is using both a start frame and end frame you would use FLFV but don\t worry about that for now. As for loras disable all of them unless their highly specific ones like blinks, otherwise Remix has no use for them.
>>20417315> This worked!Do you have any tips on keeping her face accurate? It doesnt look like her in the scene
>>20417212Tried the new prompt and still blurry.
>>20418636The diffusion models are back on smoothmi which needs the lightning lora, you need to reselect remix.>>20418613Facial drift is a thing, it varies from gen to gen. the best source images are blank neutral expression facing the camera like a passport image.
>>20418782Yeah, I was trying to try both suggestions. lightening works with remix.I like the smooth video where her butts in the frame and you can see it ripple. Do you know any loras that have that or can remix do that without a lora?
>>20418782Some what I was able to create.
>>20418901
>>20418613Glad ya got it runnin bro. Like I said if smoothmix wasn't working you shoulda tried Base Wan 2.2 model but at least Remix works. Took me about 4 months of fiddling with various settings CFGs/prompts/Loras/base models to get comfortable with WAN. Another big thing to check is that your Photo resolution matches or is divisible by your Video resolution (ie. Don't try to gen a landscape video from a portrait photo). Most Loras are trained at either square or "standard" resoluitions 480/720/1080 so best to crop or resize photo to that resolution and match the video resolution or the videos may come out wonky.
>>20418896Missionary creates a cluster of bodys
>>20418915could I just resize the photo?
>>20418915I greatly appreciate it. You and that other guy who was helping me. I learned alot just working through this problem and wish I could repay you guys.I couldnt have got this far with both of your help.
>>20418915You can also "stitch" together 3-4 5 second clips if you want a longer scene. Some programs do it automatically (SVI/LTX, etc). Basically just save photo of last frame and start another gen with that then combine with some video editor program. With some Loras like Igoon you can't really add any other effects with the Igoon because it is so specific so if you want a cumshot in the same vid that starts with an Igoon you would need to grab last frame and make a new gen with Cumshot prompt/lora.
>>20418924I want to learn this, but idk if I am there yet
>>20418916Some loras are pretty specific with their prompts. Maybe try one with the exact wording for the prompt see if that works then tweak it. Also some Loras do not play well with some models. If it does not work with Remix try Wan 2.2 base Pretty much 100% of Loras are trained on Wan 2.2 base so they should work 100% of the time with that..
>>20418924Do you think you can help me get missionary and the cowgirl to work? or a lora that does those that will work with remix?
>>20418934Okay, sorry, posted before I saw this
>>20418919Yah thats what I do 90% of the time. Plus you don't want to try to gen a vid at like 2500x3500, you will get out of memory unless you have an RTX 6000pro. For requests I resize photo to 720 max (if orig photo is 600x1900 i will just resize the 1900 to 720 and let windows do the maths so it comes out at like 250x720 or so.For vids I gen for myself i heard best resolution for Loras is 1216x832 or 1248x832
>>20418950This is what I got with what i believe is standard wan 2.2
>>20418960Yah thats weird. Sometimes I have to restart Comfy through the manager when it starts giving me weird shit (or restart the entire system). Its happened to me before when doing a bunch of different gens.. clears out the vid card/ram when it starts spitting out stuff that don't look right.
>>20418974Also make sure all your nodes are connected correctly. I have used workflows/previous gens that for whatever reason Comfy disconnected the nodes or connected ones I never used for some reason.
>>20418978Make sure you don't have a buncha workflow tabs open in the top of the browser either.. sometimes that produces weird stuff too.
>>20418993Once you get your comfy working.. I would unplug it from online. Comfy is NOTORIOUS for pushing updates that will break it. I keep mine offline and only go online maybe twice a year to update stuffs. I have read soooo many stories about people having to resinstall Comfy or troubleshoot after some updates.
>>204Thank for the tips. I have auto updates off now.Still Missionary to work. I wish this was easier or faster.. lolI was thinking of trying maybe a different diffusion model
>>20419041Make sure you use the LightX lora with the Wan 2.2 base model. Should only be like 100 seconds. It's a process for sures but once you get it figured out it goes pretty smoothly unless making a new type of gen from scratch. its always an experiment to get the right Lora strength/model/prompt combination. Even with some when I do requests will pick a workflow that worked perfect before it will come out wrong. AI is gonna AI.
>>20419050I have a rotation of 3 base models I use regularly so if a Lora doesn't work well with one base, will try the other 2. if it doesn't work with any of the base, time to restart the computer and that usually helps.Base models i use are SmoothMix, Wan 2.2 and the NSFWFastMove. The Fastmove is good for fast/lots of motion but usually loses a lot of detail because of all the motion.
>>20419056by base model I mean diffusion model.
>>20419041I don't think its just the "autoupdates" that break Comfy. When you launch the BAT file you will see it also checks Pytorch and a buncha other stuff regardless if you have updates on or not.
>>20419056Dude, you are spitting out fire tips. I tried missionary after restarting my computer and it was close.How do I know if its a lightx lora?
>>20419056Also I think for me, I thought the loras themselves were always the issue so I constantly only troubleshooted them.
>>20419072Think the LightX is the one that came with the template. Looks something like this in top left corner. Use that with Wan 2.2 base diffusion model.
>>20419124DO NOT use the LightX with FastMove or SmoothMix (unless it is the latest SmoothMix). Read through the documentation with the download you have. One SmoothMix has the LightX baked in so if you add the LightX Lora to that it will produce the blurry crap you got. The new SmoothMix model does not have Light X so you need to use it. Again, read through the explanation of where you downloaded the model and it will tell you if you need LightX also or not.The Smoothmix I use has it baked in that is why i bypass the LightXLora
>>20419132Okay, I am tracking. Thank you!
>>20419132I got missionary to work but it doesnt look like her lol so working through that now.
>>20419160Yah thats mostly an AI thing. Usually higher resolution helps but some depends on the Lora too. The Igoon ones all basically produce the video from scratch with AI so there is bound to be some facial/body changes. If you want to 100% keep the face (most of the time) just do a zoom out to sex (camera pans out to show a man thrusting his penis into the womans vagina) or similar. You will need to play around with Loras, lora strengths and prompts. Welcome to the rabbit hole of AI vid gen lol
>>20419171I feel like I am a wizard in training. God bless you.
>>20419124you have it on 8 steps. you need to change start at step and end at step to match or it'll come out as gibberish.for example if u wanted 3 steps on the high pass - start at 0 end at 3, then the next 5 on the low pass - start at 3 end at 8. if those numbers don't line up with the step count it will look like a mad abstract art painting.the step count means total btw not each, at least on that workflow.
>>20419212This is wrong?
>>20419212I fiddled around with it and whats in the workflow seems to work changing to other settings came out crap. May tinker with it some more at some point but for now works okay... appreciate the tip though :)
>>20419234is this guy acting like me or did I read this wrong? lol
>>20419229that one is right because the start and end steps match the total step number. the one with 8 steps was wrong. think of steps as one setting btw just always put the same number in both boxes, it's start at step and end at step that split the number of steps up for each passso for simplicity on 8 steps split evenly the left box would be 0 and 4, the right box would 4 and 8.also learn to use start frame to end frame eventually as it can pull things off no lora can do since you force the AI to go to a pre-prepared image. that will need an image editor, namely Qwen with a NSFW model.I'd leave that for now though, as it's kind of information overload.
>>20419234Yah this is getting confusing. I am one of the Wiz tutors helping dude getting his Igoon lora / Wan working.
>>20416356oh i forgot i have an ltx 2.3 gen of this with audio, not fully set up for nsfw yet though as it's new. so only silly stuff for now until better loras come along.
>>20419269I would do good with just clapping and moaning. lolAre you the guy that posted the orginal cum/food gif? lol
>>20419269I've been ratholing a buncha LTX 2.3 loras/models just worried its gonna break my working Comfy with all the dependencies and whatnot. Plus don't really wanna learn a whole new program but is what it is. May try and get it up and running next weekend on a separate install/SSD..
>>20419320I change the steps and it made it blurry and cartoonish
