ComfyUI Apprentice portable installer v3 (torc2.9.1/cuda13.0/python3.12)::::::::: https://files.catbox.moe/ldk14g.rar>What is it?My version of ComfyUI portable. Aimed for new apprentices who wanna jump in AI world. includes:-Installs latest ComfyUI @ python 3.12, torch 2.9.1, cuda 13.0 (ver was 0.6.0 when making this thread)-Auto-downloader for all necessary files for running my WAN, SDXL, ZiT and QWEN workflows-Install most common used custom-nodes-Install Triton, Sage, Nunchaku attention install-My Workflows for various purposes-ReActor patched for NSFW-LosslessCut program for video merging
>HowDownload the package and extract it on some fast drive that have good amount free space (100gb at least). Then----Run 1stAIdBag_install.bat--------Install ComfyUI and Manager. This is kinda bare minimum you need to install. You can install workflow's missing node via manager. Some user inputs needed during this.--------Check if your ComfyUI runs after this. If it does....proceed--------Install Custom nodes. This will install all custom nodes needed in my workflows, also other workflow's commonly used nodes (this might take some time)--------Install Attentions. Highly recommend this as they give speed boost for generations.--------Download essential WAN files and one of the WAN GGUF models (Q8, Q5 or Q2) Better your rig, bigger Q to get------------NOTE: Downloading these models take time as those are several gigs--------OPTIONAL: Download QWEN, GIT, or SDXL stuff if wanna play with image workflowsRun comfyui. On the first run it will take some time to load as some nodes like reactor will download some models needed. Once loaded, open the latest version of the workflow, re-select the all the models, vae, clip and loras. This is something you need to do every time you open up a new workflow, so the files point into locations in your computer. See the vid: https://screenrec.com/share/mkgZDHG5uIAfter that, check other settings, type in prompt and hit run.Buy me a beer: https://ko-fi.com/1staidbag
>Already got old install?----Backup (cut) models and user folders somewhere----del old ComfyUI_windows_portable folder. Then do as guided above----paste model and user folders back into your fresh ComfyUI folder after install>What do i need?-Windows, RTX card, pref at least 8GB of VRAM, 32GB of RAM, drivers up2date-You can run stuff with lower rig specs but then you would need lower models, lower resolution, lower clip duration, etc-There are some info/help to run stuff with ATI cards in prev threads-NOTE1: Major change on the pack, nothing is pre-installed, all nodes are getting installed via 1stAIdbag.bat now. Pack is now much smaller, 7megs>If/when you need help....A) take a screenshot of your whole workflow (and log console window if possible)B) state what workflow are you using (i'm not going to try fix other's workflows for you)C) what specs on your rig>Where to get loras?https://civitai.com/search/models?baseModel=Wan%20Video&baseModel=Wan%20Video%202.2%20I2V-A14B&baseModel=Wan%20Video%202.2%20T2V-A14B&baseModel=Wan%20Video%2014B%20t2v&modelType=LORA&sortBy=models_v9%3AcreatedAt%3Adesc(you need to make account and enable show adult content)https://civarchive.com/search?type=LORA&base_model=Wan+Video+14B+i2v+480p&is_nsfw=true&sort=newest&page=2(old 2.1 loras that got removed from civitai due to their nsfw bullshit policy)Old thread(s)vol1 https://archived.moe/r/thread/19910536/vol2 https://archived.moe/r/thread/19938024/vol3 https://archived.moe/r/thread/19987607/vol4 https://archived.moe/r/thread/20011173/vol5 https://archived.moe/r/thread/20035627/vol6 https://archived.moe/r/thread/20053516/vol7 https://archived.moe/r/thread/20096113/vol8 https://archived.moe/r/thread/20137696/vol9 https://archived.moe/r/thread/20195961/vol10 https://archived.moe/r/thread/20238038/
SVI workflowshttps://files.catbox.moe/hns335.json (1.0)https://files.catbox.moe/kct5jp.json (1.1)Needs svi 2.0 pro r128 high and low loras:https://huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Stable-Video-Infinity/v2.0and update kjnodes...still kinda wip thing but you gets you started
^^
Any thoughts on ltx2 so far? It looks like the generation times are great but I haven’t had a chance to test it.
>>20266308For NSFW.... Not so good atm, need to wait for some xxx fine-tune model. Otherwise looks nice, fast, 20sec gens off the bat, audio option is nice addition also
>>20264956what's in svi 1.1 version?
>>20266809do u have workflow for ltx2??
>>20267086different nodes for SVI latent, has motion amplitude control like in painter node, to fix slow motion issues>>20267456something is cooking, yes
>>20267486i can't wait to eat it!
Bump
Posted my new blink-gangbang lora to huggingface.you can find it under my profile. imb101.
Sometimes my gens end up too big for this board, what are some good ways to keep size under control besides lowering res and length?I've been re-encoding with handbrake, but that's of limited use.Not using bag workflow btw, just a simple selfmade one.
>>20269599>huggingface.thanks! will try it out later.
>>20270017increase the CRF in the output node. Try 19-22 range.
>>20270017https://www.freeconvert.com/video-compressorgo setting and set size to 4 mb
>>20270152this
The color starts to degrade quickly after a few extends, any tips?
My video comes out wonky at best, blurry mess at worst. ive tried different loras with no good results. can anyone point to what im doing wrong with suggestions on how to fix? thank youworkflow attached
>>20272690have you tried that prompt with no lora? or just the lighting lora? or a different one that has a high and a low?
>>20272690no I thought i have to have lora? I will try now. Im brand new at this
>>20272690tried to get a blowjob without loras. this is what i got
>>20272690you need to load lightx high on high noise loras, and lightx low on low noise loras....also what is that cumshot_13_1.2_covert ...is it even wan 2.2 lora?
>>20272806that cumshot lora was in by mistake, its for image gen.I see within the workflow it has links to just the lighting ones, i will try those next. When do i need to use the other loras? i downloaded a bunch of wan 2.2 loras from civitaiI appreciate the help!
currently trying the lightning hi and lo loras provided within the workflow. trying a simpler prompt? will post results when it finishes
>>20272828best result so far!
>>20272834So simple things seem to be working with the lightning loras. but anything with bj or cumshot are shaky and strange... any suggestions for better hi lo loras for these things?
>>20272858oral insertion works nicehttps://civitai.com/models/1874153/oral-insertion-wan-22
>>20273072that one says to put 24 steps? seems like its gonna take half an hour for 6 second vid? is that normal?4070 super evo265k64gb ddr5
>>20273128Use 4 step lora. high model 2 step low model 2 step
>>20273168yeah....when using lightx loras 4-8 steps is enough...DO NOT load multiple versions of lightx thou (it will fuck up your gen)
up
Any way you could add the mmaudio nodes and such in the next build for us lazy fuckers? :P
>>20267486any update on your ltx?
>>20275084Haven't got much time to build it up.... It gens, but only only horror clips
Bumps for bag
bump for bag
be careful if you guys want to be private wiz. there are many unscrupulous individuals, who pretend to be wealthy or something. they'll demand a lot of work, and disappear as soon as you ask for reward. be cautious, even if you're working for pleasure
idk if im retarded but this summer i got this to work fine. now when i run it after aldbag installer i get blurry image, blurry movement and strange (incorrect) output. anyone have a clue?
alright im getting better. took about 5 mins to gen this using just oral insertion high and low. High at 1 strength, 3.5 cfg. Low at .75 strength, 4 cfg. 12 total steps.I am going to turn down total steps and see if i can get good results.Any other tips to shortening the gen time?
>>20277725My bad here is the mp4
>>20273168when i try to use some of the lightx one i get the "lora key not loaded:" in the cmd windowwhats the problem
>>20264942a real noob question. but how is this install different from wan2 for weak GPUs?
>>20277725So you are not using lightx loras?>>20277761That is not a problem...its low lightx lora that spam those msg as it is basically Wan 2.1 Lora...... But it still works>>20277805If you mean wan2gp..... UI is diffent. Wan2gp is gradio application.... Let's just say it's made to more user friendly, this been said it also bit limited compared to comfyui.>>20277562Without screenshot of your settings, no1 has a clue
>>20278017thanks for the reply. is there a hardware minimum for your version? or can it work on lower/older GPU?
>>20278041if you use lower gguf models, lower resolutions...it should run on lower rigs
>>20278017>>20277726I did not use any lightx loras on the last vidi have added my workflow with the lightx loras added.lmk if u see any issues / changes i should make! Thanks again wizzers
>>20278387after 459 seconds i got a worse results after adding the light loras
>>20278406when using lightx loras cfg(s) must be "1"....steps 4-8, faster gens, downside is that prompt coherency gets hit....negatives has no effect when cfg is set to "1". You pick, quality or speed, you can't have both
>>20278406here she is, no light lorasseems much better. her eyes get a bit blurry.gen time: 468 seconds.any suggestions my choombas?>>20278444just saw this. will try to see if i can get quality better
>>20278452i'd say without lightx......it takes about 20-30 steps for good quality...on lightx 4 steps is enoughOn loras like oral insertion, tit insertion, blink loras....the action is so pre-defined by the lora itself, your prompt has very little weight on the out, therefor cfg (aka prompt guidance) can be set to to 1but lets say want to do stuff without loras like mentioned above. You wanna prompt something more complex, rather than some lora trigger words, then i would suggest not to use lightx and set cfg around values of 2-5
>>20278499 trying to dial it in.BTW using version Wan2.2 version 1.4 FLFV4070 super 16gb64gb ddr5265kThank you for the help!
hey bag... now that z-image base is released, will you be making a workflow for it?
>>20278552tried something a little more ambitious
>>20278559resultgen time : 178 seconds
>>20278558https://huggingface.co/Tongyi-MAI/Z-Image
totally new to thiswhen you say open latest version of the workflow, idk exactly what to doi clock workflows and this is what i see
>>20278387i told u to change your fucking sampler u fucking bastard
>>20278592dude u told me after i posted that one and i did change it lolsee>>20278559i appreciate the guidance but bro
>>20278619okay you didnt mention that. now i will try that loli see these really amazing edits on /r and would like to do them.but it seems like you gotta REALLY dial it in
>>20278647im sorry. i deleted my post because i was being very rude. im just extremely horny right now
>>20278657LMAO yes trying to figure this out and being horny is making me angryPLAASSEE HAAALP me fuck this princess
Suggestion for the Best cumshot facial lora for wan 2.2?
>>20278726Blink Facial works well for me
>>20278837Woah thats lovelyim startin to cook
>>20278837Here's another option for you guys for comparison. I use this when I don't want the jumpcut. This lora is called "f4c3spl4sh" and it's still on civitai.
>>20278969ive used this one with the suggested settings on the civitai page but it comes out THICK .which is fine sometimes.can u share your workflow or at least the strength / cfg / other loras u pair with it, if u do?
>>20278999The lora strength is at 1.0. I think i used the default prompt or a prompt from one of example videos. The workflow i used is the WAN2.2_(v1.5d) that came with the1st Aid.bag zip file. I don't think I changed any of the workflow settings at all. I just load image, load lora, and hit run.
>>20278999Cum loras are some of the ones I have specifically noticed playing with lora strengtj can have a desirably effect.There was one that at 1.0 it was like the subject was having white paint dumped on them. Cutting it down to 0.7ade it much more reasonable.
hello bump
if a Lora has only highnoise, should i just have nothing for low noise Lora or?
>>20279744yes thank you. .8 here for strength and it seems about perfect
>>20280614not sure whats up with the wiggly hands tho lol
>>20266809tell us more about these 20 second gens
>>20278969much prefer this to the cut loras. they have trouble preserving the face, and the essence of the photo imo
>>20278969i tried f4c3spl4sh using the prompt on civitai and got this, thanks im gonna be using it more often
>>20280858what did you prompt for this to get 2 men?i tried a few things it came out janky.
>>20280905i just used the one thats in the lora on civitai, nothing mentions a second guy, not sure why it happened but it works
>>20280905>nude man enter the frame from the left side. girl sucks his penis. Other nude man enters the frame from the right side and cum on girl's face while she is sucking the penis (need some oral insert / bj lora and cumshot lora)Or cumshot lora only...cum aesthetic is what i use mostly, or mix with f4c3splash 0.5/0.5>Multiple nude men appear/walk in frame. Cum shoot out their penises, cum lands on girl's face. Men's upper bodies stay out of the frame.
bag are you going to update workflow with z-image base now that it's out?
>>20281051i dunno.......to be honest only images i made with z-image are those i made while making the z-turbo-workflowbut is it just like model update? it won't work on old z-img workflow?
>>20281060i really dont know. im not smart like you to understand. but this is the base that was released todayhttps://huggingface.co/Tongyi-MAI/Z-Image
Bumps
>>20280858The cum always disappears when i use this one?? how do i stop it from vanishing??
>>20282090vanish!
>>20282090im not sure, again on that one I just used the prompt that came with it and that was the result
>>20280991>>20282128do you add a light lora? strengths? cfg? prompt?care to share those numbers please
>>20282233no changes to the workflow as it came, i just started a few weeks ago so im only using the workflows that came with the install
i have many problems with my workflow, can u tell me were i can get the right basic workflow back? im noob 8(
what workflow from the pack are you using and what changes did u made? pls help me
>>20282090there are half dozen cumshot loras out there for wan 2.2, some work better than others. And it's kinda what you prefer...f4cesplash for more realistic transparent cum, cum aesthetics for bit overkill thick paste like cum...or like i said before, mix them togetheri prefer the later so i mainly use cumshot aesthetics plus old wan 2.1 thick-cum lora together.
>>20283046I think i've dialed it in.using the facesplash plus lighting and sensual bj with varying strengths., the issue was i was adding in the sensual bj at too high a strength
>>20283046golden oldies from wan 2.1 (cumshot_i2v, cum on face, thic cum) work ok as well on wan 2.2
>>20283162not a request thread...
>>20283162this is where u learn to do it yourself.teach a man to wiz...
i try and im getting better every try.. first was drinking water than milk now the is playing with her tongue... but i am still to stupid.. but im having fun :-) ty guys
Up
does it work on mac?
>>20272756I hate when that happens. She does, too.
>>20282363Wtf is the appeal of triple samplers? More precision? I have yet to see a video that looks any better because of it.
>>20278837They work sometimes. If you don't care that the girl makes weird faces or takes a load to the face before the transition then yeah, its great.
>>20285126>>20285128no idea I dont even know what they do, but this one >>20280858 and this one >>20278837came from that same workflow and its fine for me, like I said im new to this and still learning, almost 2 weeks now so havent learned how it all really works just going off the premade workflows
>>20285126>More precision?Basically yes....TripleSampler adds one extra high noise sampling stage (high noise controls the movement mostly) at the beginning with few steps without lightx lora so you could use higher cfg (guidance) = it make prompt coherency better + allows you to use negative prompts (when cfg is "1", negatives have no effect) Quality-wise, there isn't that much effect90% of times i keep it off as most nsfw loras have pre-defined action.....its enough to type just some keywords into the prompt and you will get what it promises.
I'm completely new at this, it took 1.5hrs for a 6s clip... is that expected? Is there anything in particular I should do to iterate? Does it go quicker for iterations, or is it starting from scratch again?3070 8Gb, which is obviously underpowered for this. Should I make changes to this workflow to account for that?
>>202860501 hour? You have 8GB of ram? Also get a simpler workflow.
>>202862078gb vram, 64gb ramare there any workflows in this pack that you suggest? if not, can you suggest one for me to go find?
>>20278837where can I still find those blink loras? I remember there where several (blink misionary / blink anal) but I can't seem to find them anymore on civit or archive. Only a doggy slider which unfortunately only creates blurred results for me
>>20286050are you sure it recognize your card?.....speeds like that comes with when you gen with just cpu. It should note the card on the console window when you start upbut anyway lower the rig....lower the resolution, lower, steps to 4, use lower q models (like 5), don't use 3rd sampler, enable sage attention if you installed it
>>20286381most of them still on archivehttps://civarchive.com/search?q=blink&rating=explicitnew ones are released weekly, but behind paywall
Would an AMD card work? ive got 32gb ram, 7800xt and 7800x3d
>>20286563they should.....but you need to seek comfyui zluda. https://github.com/patientx/ComfyUI-ZludaMight to take some effort and time to install, but instructions seems good on the site. It seems like there is no portable version of zluda out there, so you need to install python, git, vc dist libs etc in before (might sound hard, but it isn't). I'm not a best person to help you out, but maybe some other anon is
>>20286393yeah pretty sure it does since it maxes out the vram on and card usage while running>>20286397cool I'll take a look at this, thanks!I'm just getting errors now, no idea what I broke... I'm going to diff this workflow against the one in the archive ig
>>20286563very easy on linux, i'm running 7900xtx 7950x but i'm the type to install manually lol
>>20287365The errors were because I had tried to enable previews and changed the mode. Setting it back to default resolved themDamn, I've got this working now, at least for simple "take off her top" kinds of prompts, I'm blown away by how well this works...
On civtai, should I be filtering by only these models for image to video?
>>20286381Awful lot of stuff gets put onto hugging face for back up. If you specifically search for adapters for WAN I2V it all comes up easily:https://huggingface.co/models?other=base_model:adapter:Wan-AI/Wan2.2-I2V-A14Bfilter by name blink and it should all be there. Lots of other interesting things on there that aren't on other sites.
I'm struggling to get their shirts to come off more naturally. I've tried a few different prompts now but it seems like the shirts keep coming off without their hands actually touching them, and then they just sorta vanish over their heads... Any tips?
>>20288007https://civarchive.com/models/2233193?modelVersionId=2514015try that...only high lora needed
bumb
>>20288602thanks! this turned out great
bump. ENF lora released on HF.
>>20291622ENF lora is...?
why are the older loras removed from civtai for nsfw policies but there are still a million new ones there?
>>20292506it seems ban hammers hits mainly on i2v loras....yet many t2v loras work ok on i2v setups, so i dunno, and seems like they don't know either
>>20291965ENF=embarrassed nude female.>>20292506>>20292511I spoke to the civ mods after they pulled down one my loras this is what they told me they look at:>if its a real person? no>if its a model that switches scenes to a sexual act or does undressing? no>if it is a model for some adult character thats simply trained to create said character? sure its probably fine>if its a minor character? it would depend on the model's intent>but everything is case by case id need more information on what the model is and what its meant to do
>>20292511>>20292681Ok that vaguely makes sense, if they ignore the fact that many if not all of the nsfw models work on real people lol It's funny that there are no undressing models there but you can make someone deepthroat a huge dick with an android number of models there Also I didn't know some t2v can work for i2v, guess I'll have to try that out
>>20292837Idk what I was trying to type where it says android lol but you get what I was saying
What HW are you guys running? I'd love to upgrade from my 3070 to something better suited for this, but idk what to optimize for. Is it just vram? Presumably core clock also plays a part, but should I favor vram over it or do I really need a balance?
>>20264942I will hopefully have a 3090 soon
>>20287855Some t2v lora work for i2v.
is there a way to delete gguf models from comfyui? i have a 5090, so i use q8, but i added q2 out of curiosity through your installer but obviously i have no use for it now
>>20293957no you're stuck with them forever now.
ty all for your help.. i got my little wiz shop working. now im stuck with my prompt problems. i have to wait about 2 hours per try. can you give me some helping hand again and post some of your good promts so i can see how they are written and learn faster? if u have the time promt and lora combination would be a dream! ty for the good help im having so much fun :-)
>>20294020i just don't see any file to delete them
>>202940572hours? What your rig specs? For prompting.... I tend to use examples from civitai Lora galleries. Try not to overload the loras.... One Lora, one prompt for it gets best results.When you are using lightx loras with cfg 1, prompt coherency is pretty poor...even if you write very detailed prompts, they get "dumbed" down on the gen.
>>20294729They are in comfyui/models/unet folder
>>20294749thanks! for some reason, windows explorer wouldn't show anything when i searched for gguf or q2
back to top with you
>>20294747does more loras really lead to issues for you? what are your specs?
>>20294747i found my problem now.. my first job is about 30 minutes, than its 1hout than 2 hours the forth in que was the whole night.. i have to restart all evert time i think. must me some kind of overflow somehow. i only have a rtx 5070 but its okay for playing around.. ty for the help
anyone got any thoughts on the various lightx/lightning loras and weights? after lots of experimenting without tidying stuff up i've ended up with a shitload of different ones
i just realized comfyui is causing a LOT of writes to my ssd. i am using q8 with a 5090 and 64 gb ram. would more RAM alleviate this?
Which is the newest version of the workflow? v1.5d or v1.12?1.5d produces good results for me, whereas 1.12 and 1.11 (I stopped testing there, dunno about the other ones between 1.5d and 1.11) produce weird artifacts like the background disappearing or the subject being entirely blurred out...
>>202980611.5d is the latest or the svi version
how safe is it to run this thing on runpod?
>>20292681any examples of the ENF one?
I was using the old "run launcher" version, I've now updated to the newest. The vfi 1.1 isn't working but the 1.0 is. The error is saying some nodes have are using a newer comfy version but it's all new and updated. Is there still a problem with gimm? It did install but what about the included addon .rar pack? I didn't see info about this.
>>20299475Nevermind both 1.1 and 1.0 are working now, maybe replacing gimm in custom nodes folder from the addon rar pack helped I'm not sure
For the svi wf, what if I'm using built in lightning lora in the gguf? I've tried just bypassing the lora nodes.
>>20299573That should do.... Bypass or delete the lightx lora nodes
>>20299446kinda like blink ones
Does anybody have any suggestions for how to make it respond better to the prompt? Is this a bad prompt? What could I do better?For a while, it was doing a lot better. I changed the weights and loras to try to fix some other issues (like the shirt randomly disappearing even though I had the shirtlift lora), and it resulted in the prompt being followed less strictly as well as a choppier output for some reason...Any tips would be appreciated!
>>20300558Yeah. Don't use default Wan 2.2 models and use variants like Smoothmix or NSFW FAST.
>>20300890like one of these?https://civitai.com/models/1995784/smooth-mix-wan-22-i2vt2v-14bhttps://civitai.com/models/2299142/smooth-mix-wan-22-i2v-v20-gguf
>>20301015to add to this person's question with a possibly stupid question: what folder do those "checkpoint merge" go? the checkpoint folder? or the lora folder, since i assume it shows up in comfyui in the lora section
>>20301477If checkpoint.... Then checkpoint folder. But I think those mentioned above should go to diffusion_models. Checkpoints usually have clip and vae build in, so you need to use checkpoint loader node
>>20299446>>20300158the model is trained mostly on 5 second blink cuts from forcedfantasies videos. So it does 1s of clothed and 4 seconds of ENF. There is a fair chunk of boob sizes in the training material but for whatever reason it seems to always make them bigger so adding the flat chest lora into the mix can help with that.
>>20301630ok, i am a total noob so i ask bc they're labeled as checkpoint or checkpoint merged on civtai so i am confused lol
>>20301677is flat only on the archive?
>>20301719no its a t2v which works with i2v.https://civitai.com/models/1983079?modelVersionId=2244856
imma bump this thread and let everyone know that i am very horny right now
random question for you wizzes on etiquette ... what's the go-to for posting in a thread another wizz has already blessed? do other wizzes get annoyed?
>>20303669i would say; no, doesn't matter. If the image is good, there might be few wiz who put that in the oven at the same time, so why not post them out? Even when using same tools, outputs look different, and thats the kinda cool thing about all this. There are no points given, not a competitiononly etiquettes i have; a) if the request maker acts like ass-hat when getting stuff done for him for free, sage/hide/ignore forever. Most of the wiz follow this ruleb) don't do same images that are posted every fucking day, in every fucking thread, those have been done multiple times already. No1 likes greedy ppl.
>>20303793thanks bag, great guidance
>>20301015I had the most success with these.https://huggingface.co/Bedovyy/smoothMixWan22-I2V-GGUF/tree/mainDon't use Smoothmix v2.0. They are inferior in everyway if you are trying to do smut.
>>20303852cool I'll give it a shot ty! what does the M or S suffix mean?
>>20303866medium and small.....small is bit smaller (duh) and less ram demanding (loads faster) , also less precise/quality
sorry for being a noobso this would go into lora folder and would be used either by itself or maybe one other Lora to enhance it?https://civitai.com/models/1307155/wan-22-experimental-wan-general-nsfw-model?modelVersionId=2073605and then this one says it is a checkpoint, but what folder would it be in? would it be replacing the gguf stuff?https://civitai.com/models/2053259/wan-22-enhanced-nsfw-or-svi-or-camera-prompt-adherence-lightning-edition-i2v-and-t2v-fp8-gguf?modelVersionId=2647324or, someone mentioned needing checkpoint node. if i do not planning on actually adding stuff to a workflow, do i just avoid anything that isn't labeled Lora?
>>20303963Dont use the NSFW lora. It sucks, obsolete and overrated. Search for Smoothmix XXX lora instead.
Does this program work on the newest MacBook Pro?
Does this program work in a Mac? I have the newest MacBook Pro
>>20264953>>Where to get loras?>https://civitai.com/search/models?baseModel=Wan%20Video&baseModel=Wan%20Video%202.2%20I2V-A14B&baseModel=Wan%20Video%202.2%20T2V-A14B&baseModel=Wan%20Video%2014B%20t2v&modelType=LORA&sortBy=models_v9%3AcreatedAt%3Adesc>(you need to make account and enable show adult content)I always get "1 models have been hidden due to your settings." when I enter that link. But I am logged in with an account and have set all the settings to allow mature content.I did try to hit the "refresh session" button in the settings, and I also tried logging out and back in.Can you help me getting the lora?
>>20264953 Two questions after following all the setup steps:1. Does this only work in a browser? I thought it was it's own program. But when I open it (run_nvidia_gpu.bat), I get a command prompt window that shows me all the booting up process, and then it opens in a tab in my Firefox browser.I had the official comfy.ui downloaded and installed in the past, and that was it's own program window.2. I don't know how to add the recommended loras that I downloaded. I put them all in a folder inside your compfyui apprentice folder. But when I boot it up (and it launches in a browser tab), I have that " Launch lora manager" button at the top. When I click it, I have several options like "fetch" and "download", but nothing simply lets me select the folder/path to my loras.Do I have to put them in a specific folder manually?Thanks for all you do!
>>20304707This issue has been solved.I had to switch the search result filters to "sort by relevancy"
Best model for image edit and undress ? I have 12gb vraiment
Does everyone else get videos that are always 2 seconds longer than what you set it for?
Anyone know of a blink lora that does doggy style, from behind, with the woman looking back so you can see her face, instead of just ass?
>>20305934Check your workflows interpolation node. If the movement in your videos look like they are fast forwarded, the workflows settings may be off.
>>20306010You can find the deleted Blink loras at Civ Archive. You don't really need the lora for it but simply use "jumpcuts to" whatever you want in your prompt.
>>20306059Yeah I've seen some there but didn't see a doggy one like I described. They kinda seem to only do 1 angle. I haven't needed with them too much yet but the fov slider was the easiest to useI tried doing it anyways with a prompt with some blink with doggy but it just gave an abomination lol
>>20303852Do I need to change anything else other than switching to these models? I tried just switching and keeping the rest of the settings the same and the result came out awful...
>>20306351When you are using lightx loras cfg (on the sampler) must be set to "1.0"
>>20306831base_cfg? lighting_cfg is already 1.0. The lightx loras were disabled here though. I'll try anyway
>>20306867oh yeah...i mislooked, then all ok on that part. I dunno about that smoothmix model. judging by the filename it doesn't have anything baked in. Some custom wan2.2 models have lightx loras baked in.....in this case you need to disable any lightx lora loaders.When running stuff without lightx....you need to increase the steps into 20-30 area, also set the cfg to 3-5 area
>>20306096something like this?
>>20308176yeah. did you use the blink lora or eas it just prompting? Also how do you get the camera to move in on her face like that? I always get a fixed camera angle
>>20308181its the blink back doggystyle one, and i did tell the camera to move in to her face and have her mouthing some words. I got it from the link posted here.
>>20308176>>20308192what was the prompt?
>>20308228the lora page has example prompts just adjust it as needed
Any idea why, with the same exact settings, most of the time it takes around bl6-8 minutes to make a video, but then sometimes it takes 20+ minutes?64gb ram with a 5090 and 9800x3d if that matters It very rarely happens, just curious as to what causes it
>>20308410depends of the orientation of the pic.....best case scenario is to do 1:1 images, then landscape s.....last portraitsthis is mainly due vertical lines...try it, crop your low speed image to 1:1 ratio, and try it again
>>20308668Ok thanks for replying, good to know. Though a lot of my pics that don't take forever are portrait, so I was wondering if maybe it was page file stuff, because my ram does max out, but idk if that would make it slower
>>20308668And yeah next time it happens I'll test it with 1:1 to see, or also just try it again with its original ratio to see if it takes forever every time
>>20308410What resolution are you running these at? I run at 640p with a 5060ti 16gb and the majority of my 5 second vids take around 180 seconds. I figured a 5090 could just pump them out.
>>20308668Ok so the most recent one I did, the original image was 480 x 485, and the output video was 832 x 848. I'm guessing that wouldn't account for the insane time for make the video, since most of the videos with the same setting are between 4 to 10 and many are way less of a square than that one which
>>20308410>5090>6-8 minutes to make a video,something is very wrong here. 5 second video at 480p on a 5090 is 70 second to gen.make sure you use lightx and sage attn 3.
>>20308744I have the length at 7 seconds (which outputs as 10 second videos for whatever reason) I just left all of his other setting default which was 12 steps and max output length as 848. But it's not the time I'm concerned with,nits why randomly some videos take like 5 times longer
>>20308754I'd post a screenshot but I'm not home. I do now the sage things were set to auto and I don't even know what lightx means lol I just stayed messing with this a week ago
>>20308746848 is pretty hard call unless your got 32vram card and nice amount of ramI address to you all again....."how can i make stuff faster and look good?" ---> ain't going to happen, you can only choose one
>>20308772Well that's just what the 1.5 version defaulted to, and I have a 5090 and 64 gb ram. Again, I am not complaining about the length of the videos, I'm just wondering why 1% of the time, the video takes 500% longer time than all other videos It's not the length or quality that is as issue, it's the very rare, wildly inconsistent time that happens sometimes
>>20308756i don't recom making clips longer than 5sec....why they are longer is your interpolation settings (if interpolation set to 3 @ 32fps, it will look tad slow-mo)
>>20308780this could be because ram is fully offload after a gen.......this might also result you getting colorful noise outputsreboot on the comfyui is the only option there
>>20308785Roger that on the interpolation. So should I change it to 2? I literally have no idea what I am doing obviously
>>20308797>ram is fully offload after a genram is NOT fully offload after a gen
>>20308798yes......wan makes 16fps by default......interpolation 2 means it doubles the frames. if fps 32, it looks smooth
>>20308804Ok yeah I figured you meant not lol>>20308815Ok. Is it default at 3 interpolation, 848 max output and 32 fps or did I accidentally change something lol? For the 1.5d workflow? Is that the one I should be using? Sorry I know this isn't supposed to be the comfyui faq
>>203088323 is default because its my workflow and i like kinda slow-mo action>Sorry I know this isn't supposed to be the comfyui faqthis is kinda what these threads are for...learn, play, ask, play more
>>20308842Yeah I hadnt't noticed the slow mo(or not in a bad way) maybe bc I've only been doing the oral insertion Lora which I guess looks fine slower haha
>>20264942Any worthwhile loras coming out lately? or did they kill them for good
>>20308842Since you made the workflow amd obviously know what you're doing, is this good advice about sage attn?And yeah IDK what lightx means I'm not doing anything crazy yet, just using basically verbatim prompts from the Loras creators to get the desired effect >>20308754
>>20308863>you made the workflowyes>obviously know what you're doingnoi just play around, these are my workflows, how i gen stuff,,,,i just wanna give you guys the engine. There are dozen (better) workflows out there, get them, see how they work, make your own
>>20308780>>20308797my guess would be he's got a non-cuda version of torch somehow and that 1% is torch doing a compile.
>>20308780You will never get this
>>20308923How to optimize the SVI 1.1 wf? There's no option to change the steps below 8 on the slider (I'm using built in lightning .lora in the .gguf models).
>>20309884right click the slider node, select properties panel, they you can set up the min and max values
Currently looking at loras while this installs. Noticing they have high and low options. Do I download both or is one better to use than the other?
>>20310036you need both....load high ones on high loader, low ones on low loader
>>20264942Testing out the 1.1 SVI Workflow and I am having trouble getting consistent results. The first extend usually is somewhat decent, but the 2ond and 3rd ones break with unnatural movement, wierd artifacts etc. Any tips to stabilize it a bit? I kept the default settings, turned on TripleSampler and am trying to keep my prompts short
When using Sage Attention, is it best to leave it at auto or select one of the provided ones?
>>20309990Thank you Bag. Every time I load a workflow the seed in static by default not random. Also on every other workflow usually I can just click the input node then ctrl + v to paste the input, however in the SVI wf it just duplicates the input node instead of adding the input.
In SVI new workflow loads with seed changed to recycle instead of random*
>>20310047Hey bad, how close are you on an SVI workflow??
>>20311237bag*
>>20308176any advice to a noob on how to keep the face accurate? for me, the faces look nothing like the original once the angle switches
what is the difference between the Sage Attention and the normal comfy? im new and im still trying to understand.. also i cant que anythink.. if i do so it will stop working and is stock for hours with tripleKSampler.. i use a 5070 and a i9 14900 ty for the tips
>>20311309blinks are really good for retaining identity... if you don't use one you need to let wan try and capture the identity within the first 1 second or so of run-time. If you are still getting face drift then its probably a lora messing up the identity... a lot of cumshot, or even pussy loras can effect the identity. I don't use Dasiw because it changes the face, sometimes smoothmix will change the face slightly too.
>>20311915idk, i was only using the blink back doggy lora and nothing else
>>20312311that identity seemed pretty good to me, wan isn't perfect at it. unless you can add some controlnets/code or tweak reference latents or conditioning identity will always drift somewhat.
>>20312402idk if this is what you meant, but i am not the person that posted that video clip. i agree it looked fine in that clip, but none of the videos i generate do
best lora/prompt for undressing fast? also showing natural tits and pussy? good example here of it >>20309739don't think the boob flash or titty drop loras i have are good with pussy
hmm all of this is really fun, but is it worth it?is there anything better i mean?
>>20313117is what worth what?
what do people mean when they say using/not using lightx lora?all i do is download a lora from civtai and put them in the high/low noise lora section. is there something else to do? should i be using the default 4step that comes in that section along with downloaded loras or no?
>>20313387brother listen, just do a bit of research and don't try to rush things. I haven't used bag's flow in a while but it has a lot of helpers and good things in there. If you keep coming up short do some research on what each thing does ... this is an entirely new tech just like anything else. You got to learn to walk before you run. It took me 4 months to finally get good results but I spent a long damn time genning garbage and making mistakes. Use AI to help with AI as well chatgpt is free and it can answer most questions about what shit does and where/how to do things.
>>20313467that is what i am doing, asking people who might know the answer...bag's workflow worked for me, first time, and i have had zero issues with it, so as far as my needs go, i am already running with the help of bag's stuff. i have never generated garbage using the stuff he has given us in this threadi am simply asking what the terms mean, since there are people in this thread who know the answer
>>20312561Thats lazy undressing. You can make a bitch undress quickly just by making the video short.
>>20311616Sage attention is some sort of python witchcraft that supposedly shortens generation time.
>>20264942thanks aidbagis there any reason why this one uses RIFE and not GIMM? cause didn't your 1.5 or 1.6 workflow (cant remember) move to GIMM?
>>20313978latest workflow has two version...rife and gimm. There was some issues with some people getting gimm to run (gave some weird cuda errors)it's easy to switch rife --> gimm add gimm node, hold shift, pull the lines from rife into it's corresponding slots, delete rife node(and you need to add gimm model load node and plug it in)
>>20314816oh sweet thanks anon, works. have no cuda errors on my systemgod bless ye
about half a year ago i got 250-300sec gens on my 5080 with pisswizard his flow and aidbag installnow i did a fresh install with sageattention and aidbag flow and my gens take up to 1000seconds.anyone know what im doing wrong?
>>20315245lower output longest side, set enable third sampler to 0
>>20315245Lower output longest side to 720, enable Patch Sage Attention (high) and (low) on the bottom right, lower interpolation from 3 to 2. Since you are running a 5080 you can run third sampler... if you keep it enabled: increase steps from 8 to 12, and increase "base_quality_threshold" from 20 to 30
>>20315527I meant bottom left*
Going through Igoons huggingface page and I see "WAN-2.2-I2V-Combo-HIGH-v1" Either I'm a retard that cant see any specific details or its not listed. What is combo? and I heard recently that he had a gangbang lora, cant seem to find it. any help appreciated.
>>20316421what am i missing about huggingface? like, not having pic/video examples of that a model does, like civtai does, seems bad yeah?
>>20316445I'd be happy with a discriptor of what the lora actually does with some triggers for the lora.
I'm struggling to get this bikini to come off nicely. It keeps getting lifted completely over the head in a weird way. I just want it to get pulled to the side or something more natural like that.Any suggestions? Any loras that might help? The shirtlift lora mentioned above doesn't seem to help...
>>20316655"the subject" lol, use loose terms , "the girl takes her top off" "the girl pulls her top off" etc. there is too much minute detail, start simple then add extras.
which is the newest/better workflow? the 1.5d or the rife version?
>>20304006is that this one? https://civitai.com/models/1995784/smooth-mix-wan-22-14b-i2vt2v
>>20316789they are the same....only different interpolation node
I'm having some trouble before I even start lol. Comfy UI won't install because Cuda 13.0 won't seem to install no matter what drivers I update etc. Anyone else had this issue?
>>20264942I'm really struggling to download the package, is there a mirror maybe? Ty for your work
>>20313387It was a big jump from the original workflows to using the painter motion multiplier node and triple sampler in thread before this.>>20317524I think the minimum card series is Nvidia 20xx as my 1050 doesn't have compute capability over 5 (necessary for torch 2.7.1 and numpy).Try looking back 1 or 2 threads for the 5GB download package and workflow 1.5 or even easier is to go back to the 1st thread and use the original.
>>20317995The thing is I have a 3080, so it shouldn't be an issue
>>20317713When I tried using the custom install from FAId, it would install everything but autpclose on startup, for what I assumed was a torch issue with regards to my 3060.I was able to install comfyui and everything else without using the installer.
for those having problems with installerhttps://docs.comfy.org/installation/comfyui_portable_windowsthat installs the basic version of comfyui. after that you need to manually install the manager and all the missing custom-nodes as you go and open new workflows. installing sage and triton is not required, but recommended. It need some knowledge of basic console/python commands, but nothing that there is not info aroundthe purpose of my installer is the automate most of these tasks. sadly for some the installer shows the middle finger, for reason i don't know. If you have rtx card with cuda and drivers up2date, all should be ok. Comfyui portable is basicly closed system, everything it needs are inside the python_embeded. Only reason installer wants to install cuda 13 toolkit is because GIMM interpolation node seeks cuda libs from systempath rather that portable/python_embeded/libs.....This might be fixed now, i haven't checked what custom nodes got updated since i've made the installer. But interpolation is important step to make smooth gens, i couldn't left it out
>>20317713>>20318148>>20318300It is not the installer itself for me, the catbox.moe link just won't download (either doesn't download at all or fails at some point). If you could share just the workflow somewhere else, I can probably scrounge together the rest
>>20320596there was mega mirrorlink on past few older threads
>>20264942I get this error message when trying to run it.I THINK I followed everything from the video. My guess what's going on: I'm supposed to enter a file path to where to export the video in the "CLIP" window? If yes, I don't know how to change that.
>>20322350you have to re-select everything everytime you open a new workflow. If you've done so, make sure your files are completelow q8 wan model is (15,406,608,896 bytes)
>>20322350>I'm supposed to enter a file path to where to export the video in the "CLIP" window? If yes, I don't know how to change that.video are autosaved in comfui/output/%date%/ folder.....once the gen is done, you can right click and save it in your own different location as well....there is no need to set paths anywhere
im getting great gens in good time. but how do i actually get longer videos without using the slider? i have to combine multiple gens in some way? i know its in the workflow but im too retarded to figure it out after a lot of testing
>>20322577use svi workflow......or make a gen + save its last frame, use that as new input image, gen again......once done join the the clips together with some video tool
>>20322583does that come with the install or you mean the one on civitai?
>>20322649>>20264956
>>20322656sorry, shouldn't have brushed over the thread that quickly
Glad I tried the SVI flow, this is great!
>>20323224what model did you use just for taking her top off?
>>20323531Search shiratlift at civitai archive
>>20264942Noob question: Which value do I have to change in your workflow ( 1stAIdbag_WAN2.2_(v1.5d) (Rife-Version)) ) to make the created video longer? They are about 8 seconds long when they finish. But it seems like just what I want to happen really starts, the video is over. What would I have to do for a 15 or 20 second video?
>>20322495Thank you! I thought I had all, but missed one!
>>20324724General advice is NOT to make gens longer than 5 seconds. After this Wan doesn't really know what to do, things start going in reverse or other weird stuffYour gens are 8secs because interpolation... It adds frames between original frames, therefor making the video longerCurrently using svi loras and workflow is the best option to make long videos with Wan
>>20323531see >>20288602
>>20324846ok so you just put nothing into the low lora? i thought i read someone saying to put it into both if there is only one
>>20264942two questions:1. I have issues making the woman in a picture give a blowjob, if there isn't a dick in the original picture yet.The dicks that come into frame in my projects always end up pure nightmare fuel. They look like monster meat sticks without a dickhead. One time, a man entered the frame, with such a moster slug dick attached to his chin, and she sucked that.What can I do about this, for pics where there is only the woman in the frame?I tried several blowjob/facial loras.Is it about my prompts? Should I actually say something like "an erect peinis enters the frame and she starts sucking it"?Or should I just write "she performs a blowjob on the penis" and then it will just do the right thing?2.Can I combine different loras?Lets say, I have a lora for blowjobs, a lora for facials and a lora for deepthroat.Can I load each of them into a lora slot, and then use a prompt like "she sucks gives the penis a blowjob. He pulls the penis out of her mouth and shoots cum on her face. Then she takes it back in her mouth and gives a deepthroat blowjob."?Or will things get all messy , and I should only ever load and prompt one of these loras at a time?Curious about your tips!
>>20324856Not in this case......if you use old wan 2.1 loras, that are just one single file, then you could load it on both loader....thou i think loading wan 2.1 just in low-lora loader is enoughin wan 2.2; consider high lora as the action part....it has all the movement. Low lora has has all the details and textures. In that shirtlift lora, it only has the movement of subject lifting their upper clothes up (basicly it animates/directs the arm movement, it leaves rest to base wan model and other loras.>>203249781st question: usually there is info on the lora (on civitai) if the lora has been trained with only penis in the initial frame or not....most of the cases they are not. what i use most of the times is the oral insertion lora as it can add the man (and penis) into the frame.prompt: "Nude man enters in frame from the left side. Man has erect penis, his upper body stay out of the frame. Woman sucks the penis"There are other loras as well....oral insertion lacks bbc. so i then company it with biggestdicks lora or some bbc_blowjob lora2nd question: for basic wan gen, in 5 sec, all that......yeah, don't.....You can combine multiple loras but i don't recommend using multiple loras that do similar actions, or using two loras that do opposite actions....like your example, if you have blowjob lora and lora for man fapping next to her face and cumming.....both of those actions can't happen. it picks the lora that has more weight. -Do you first gen with just blowjob......save the last frame of it (use oral insertion lora for example)-Use that last frame as your next input image. then load your cumshot loras...gen and save the last frame-Use that cum filled frame as next input, load blowjob loras again and gen....(note that now you need to prompt that woman has cum on her face, and load some cum lora too, otherwise wan will try to render normal woman skin)then join the clips together........ORuse SVI workflow
i had this running fine on my 4080 super, just upgraded to 4090 now i keep comfyui keeps crashing on stage 2? any ideas?
>>20325987apologies i guess i had a stroke typing this but aye crashed since upgrading to 4090, anyone else running a 4090 without any problems?
>>20325469Thanks for the reply:About your answer to the second question:If I do this step by step and xreate multiple clips from the end points:Is there an easy eay to stictch them all together into one clip?Like alittle program? (or easy workflow)Maybe I already have it? I installed everything from your amazing installer!It would feel a bit de-motivating to always boot up my shitty video editing software and create a video project in there.
>>20325469Me again:about the SVI workflow you recommended for the second question:I did download it from your reply to the top post.But that thing os way over my head right now. I don't know what to do in that maze of a workflow. Half the boxes are covered purple, the other half get maked on red for error, when I try to run it.Maybe I can wrap my head around it at a later point, when I have learned more. :-)But it's good to finally have an idea what that workflow is used for specifically!
Getting a GIMMVFI_interpolate list index out of range error pop up at the wanvae stage after stage 3 on the 1.5d workflow. Preview also seems to just be a black square so I think there's an issue with interpolation since switching gpu.
>>20264942using your workflow and the "oral insertion" lora, I'm trying to make her suck his dick. (First frame of this video is the starting frame.)I already tried "She kneels down as he opens his pants. She starts sucking his penis".But that only make HIM open HER pants and then the video is over.Now I tried simply "she sucks his penis" and got the video I posted here as a result.What should I do?
>>20326898If the preview shows then the problem is with the sampling step, not the interpolation stageHave you tried other Wan workflows, do they work
>>20326675Purple mean the node is bypassed. Red means the node is missing or there is problem with itUsing it like ain't that hard or different than other workflow.... Reselect the models and other files......if you want to use it as basic workflow.... Disable (bypass) all extend groups using the radio button switches. Go to initial scene group, select your action loras and type prompt, hit run
>>20326987Well the issue is that there is no penis in the starting image. And Wan is stupid enough to realize that man might have one in his pantsI would do it. Get assume the position Lora from civitai..... Use that on your first Gen. Save the last frame of it if using non svi workflow On second Gen use oral ins lora and prompt something "lady kneeling in front of man. Lady unzip his pants and pulls out erect penis from his crotch area. Lady sucks the penis
>>20327905Oh okay. I assumed that the Oral insertion" lora was the answer because of this reply:>>20325469Now I will try "assume the position". I'm a bit worried that none of the example clips show a guy/dick, but I will give it a try! I also tried the lora "Cheek fuck insertion", because the example clips on that lora's page had a dick coming out of nowhere and looking real. But All I got with that one, was salami slugs again. (made this one with that lora)question: With what you have written (that she unzips his pants in the second gen), that would mean that his dick still wasn't visible in the last frame from the 1 gen that I made with "assume the position". Doesn't that create the same problem? Well, I will go try it anyway later.
>>20328224note that examples on civitai lora page's gallery are not made by ONLY using that specific lora. if you click the sample video, some of them list all loras in use
>>20280905could you please share your workflow for this one? Would be highly appreciated!
>>20326987this would be my approach to using svi workflow
>>20328551
>>20328551Thank you for your patient support!I will try that tomorrow amd report back.
I just want the dick inside her already to creampie her and pull out. wtf o.O!!Help with prompt> A woman's vagina. A man's anatomically correct penis and balls appears on the screen and starts cumming. The size of a man's penis is realistic, similar to half the size of a woman's face. A man quickly moves his hand back and forth along the shaft of his penis. Massive amounts of cum shoot off the cock and land on her vagina and pubic area, covering her vagina in viscous, slimy-translucid, messy semen. Her vagina become coated in thick, sticky, clinging cum. The semen continues to drip down her vagina. The view is from the frontal. Bukkake, cumshot, fujifilm, masterpiece, male_pov, stable camera movement, enhance realism, photorealistic, soaked, stable face, consistent face, no face morph, she keeps facial features. HD. Best quality. The lighting matches the original image, the colors match the source file, and the scene remains centered in the frame.
>>20328635Pls sir >>20329396
>>20329380What loras in use and where did you copy that prompt? And what are you trying to achieve? There are loads of lines in that prompt that contradict the other
>>20328551>https://files.catbox.moe/kct5jp.jsonI'm currently trying to copy all your settings from the screenshot. On your screenshot, "control after generate" is set to "fixed". But for me, it is greyed out, with no option to select anything. I don't know how important that one is, but I liked your result so much, I want to copy everything for now. What can I do about this? (or does it really not matter?)(using the SVI 1.1 workflow that you posted for download under the original post)
>>20328551I searched both civitai and civarchive, but I can't find your lora "wan2.2_23High noise-Cumshot Aesthetics.safetensors".I guess there is no shortage of cumshot facials for me, but I would love to try it out, especially if you seem to like it. Can you please point me in the right direction? :)
>>20330090I forgot: The same goes for the low noise version of it, "wan2.2_Low noise-Cumshot Aesthetics.safetensors"But I guess that, if you can point me in the right direction, I probably find both in the same place :P
>>20330094UPDATE:I THINK that I found them with online searches and results on Huggingface.They are not named exactly the same, but close enough.I found 23High noise-Cumshot Aesthetics.safetensorsand 56Low noise-Cumshot Aesthetics.safetensorsI'll try them out and cross my fingers :)
>>20328635Okay! Here is what I got! (trying to copy your settings from the screenshot as good as I can)The dick still looks a bit weird.any tips for improvement?(I will post a screenshot of my entire workflow in a reply to this post.) It surprisingly only took 11:30 minutes on my RTX 4070 Super. That's not really longer that it took me to make 8 (5) second single extend videos with the "1stAIdbag_WAN2.2_(v1.5d) (Rife-Version))" workflow.I would have no problem, if it took 20-30 minutes, if that meant getting the best possible quality.I have my PC do this on the side while I play console games, so it's not a matter of saving a few minutes.
>>20330247workflow
>>20330247trying to help troubleshoot this:One theory that I have about the slight quality issues with the dick:When I used your installer to install everything, there were several install options where I had to make a choice based on the power of my PC.I don't exactly remember what they were or what I choose, but I remember, that I didn't choose the highest options, since I only have 12GB VRAM and don't have a RTX 5090.Could that be the reason?(I am still a total noob at this, so this theory could be totally wrong. But if it was the reason, I wouldn't be 100% sure how to change it now, without un- and re-installing everything, and which specific options I would have to choose.)Other than that, there is one difference between your workflow screenshot and my workflow re-creation: the greyed-out "control after generate" that I mentioned above. (as well as a few differences below that, where I have things like "lightning_sampler" and "lightning_scheduler" that are empty/greyed in your screenshot, but available/active in the workflow template that I downloaded.)
Sorry, me again :-OI have run the workflow from my screenshot 4 times now. (The other times MINUS the cigarette part in the first prompt, obviously.)In two of those 4 creations, the dick comes out of the guy's butt, like this.I assume the key to avoiding this lies within the prompt, but my brain is not yet wired to speak "compfyui prompt speech". What would you do? (I know you wrote that the negative prompt field doesn't do anything in this.)
>>20330298Yes my version looks bit different as I'm constantly changing thing..... Options that are greyed out means that setting is coming from some other node
I cannot for the life of me generate a decent looking pussy during missionary or hardly at all. I've tried to train my own to varying degrees of success but the pussy always seems to get fucked up. I've tried remix, dasiw and smoothmix but these tend to change the faces of the people slightly so identity slips. Any ideas/help here? What do people use?
>>20327865the preview does show but its just an entirely black square then i get the error after stage 3
I have experimented a bit with the prompts, to make her put her hands behind her back at some point. Also to avoid her standing up again.I feel like I'm really getting somewhere!Issues are still mostly around the penis:Sometimes it's blurry and weird, sometimes sharp and realistic. And the penis almost always jiggles like jelly in the later section. Any recommendations to improve these things? :)
>>20330379well if you used the same exact prompt for that image......there is no man in it. (and the prompt state there is)the difficulty of prompting svi is that you need to kinda imagine how the scene would end and take that into account when prompting next extend
Another one I made with this workflow.One thing that I also noticed, is that the face/head of the woman often looks quite different that how she looked in the starting frame.But I guess that there is nothing I can do about that, with the changing position of her head at the different end points. (Or is there?)Kinda just wanted to some results that I've made with your help. :) Feels great to make progress.(Same penis issues)
>>20330795 I will check the prompt and look at that, once my current workflow has finished rendering! Thanks for the tip.
>>20330795Okay, I have not managed to see where the cause the blurry penis is in my prompts.I understand, that it's difficult if there is no penis in the original image. You managed to create a "good" penis in your clip from >>20328635. My goal is to get results like that.The video in this post is my latest output.I will post it's exact workflow with prompts for this clip in just a moment. If you are so kind to keep taking time to help me (highly appreciated!), please let me know what to change to have a penis that looks normal the entire time.
>>20330847 workflow and prompts
>>20330850zoomed in on the prompts, in case they are hard to read if you are on a phone or something.
>>20330850this is why you always post a screenshot.....much easier to find the issueISSUE IS:you have high loras loaded in low-lora sectionshigh ones go in to high lora loader, low one into low lora loaderhigh ones contains action, movement etc.....low ones details and textures.
>>203308731. Lesson learned! Will always provide a full screenshot with all further issues!2. OH. MY. GOD.I feel so stupid.I just figured out that there are "HIGH" and "LOW" buttons on top of the civitai page.I always assumed that whatever I get from the download button on the right, is the only version provided by that page.Thank you so much!
>>20330873Works great now!!
>>20264942 Can you recommend any good alternatives to "assume the position"? I don't want them to put their hair behind their back. But it's so strong in that lora, it often comes through even if you forbid it in the prompts.
