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.--------Install Attentions. Highly recommend this as they give speed boost for generations. --------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)--------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-NOTE2: Needs cuda toolkit 13.0 installed as few nodes can't seek needed cuda files from embeded folder. Installer dloads the installer for you. So not so "portable" for now>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/Buy me a beer: https://ko-fi.com/1staidbag
^^^
Up
do you use negative prompts? grok always has intricate positive and negative prompts but i'm not sure they actually help
>>20197758Negative prompts have no effect when cfg is set to "1"....aka you are using speed loras. Adjusting NAG node settings will make negatives count somewhat but I haven't tested it too much
>>20197788gotcha. do you mess with the lora strength or generally keep it at one? trying to make an anal penetration vid and keep getting weird anatomy
>>20195961bitcoin miner
what is difference btween q8 and wan 2.2 remix q8 ???
>>20197914
>>20197914What do you mean?
>>20197914who the fuck embed bitcoin miner into program that takes 99% gpu power to run, do you know how miners work?!.... It's a bat file, you can open up with notepad, upload it to virustotal or whatever to see what it does and installs. Read some more to make better baits in the future. Give you 1/5 (for the bump)>>20197985One is Q8 of official build, other fan baked model. Much enjoyable method to see the difference is to try them yourself rather than asking me ;)>>20197802depends.....if the lora tends to change textures, faces etc too much, i usually lower the lownoise ones to 0.5-0.8. But mostly 1.0 is good setting to go if just one lora loaded.
>>20198633ok im not trying to be a dick but the reason i ask difference is because i dont want to install things unnecessarily which is why i am asking. u made this thing and we have a right to know and ask questions
>>20198669i didn't take it in dick way...that is an alternative model of wan2.2, you don't have to download it, OG wan works just fine, you only need one Q set. Remix claims to be bit better with movement and nsfw (without heavy lora load).....<-- is something that i cannot explain if it is or not as its kinda subjective thing.....there are dozens and dozens of models out there, sadly the only way to find you like the most is to test them by yourself. I just made the installer and added to options for newbies to easily jump in, i didn't make the models, clips, vaes, etc.
Sorry first time getting into AI gen, is it done offline or does it use internet to generate? I understand they use the GPU to gen, but I was afraid prompts etc uses the internet to gen said images or use my images to gen. there are some laws around these parts that prohibits certain gens so while iam genning my own images that are legal, i just rather be on safe side.
>>20198931Once you have all the necessary models and files, you can unplug your internet if you wish and they will gen just fine.....they won't be posted on the internet (unless you decide to do so)
which one do i run bag? i have everything isntalled
how do i switch to use QWEN or SDXL image workflow?? sorry new to this
188 seconds to render. using default settings. this good bag? looks good to me. not sure if i can make it better.
do her
>>20199295if attentions installed....then sage....but basically no matter, sage if launched via node if available>>20199343looks pretty good on good speed, i assume you have 50xx card?>>20199308on the left side there is workflow icon, click that, that show workflows i made/included, it has few for sdxl and qwen. But note that they are very basic or build for some specific purpose. You might want to looks some better image workflows from civitai
>>20199412yeah i have 5080
>>20199412i promise you i installed it but i cannot find any other workflow :(
>>20199349you can play with her
>>20199470you could easily increase the output reso to 960x example and also enable the third sampler, maybe add few step (set the slider to "1").....do another run with those settings and compare.....sure it will be slower, but since you got rig with some beef, use it
>>20199475it don't look like this?
>>20199489nope it's just empty for me
>>20199483where to add steps and third sampler? ^^
>>20199503weird....can you see add-ons folder in your comfyui root?
>>20199511
>>20199519extract the addons.rar again manually.....move the user folder from that inside your comfui folder, (replace files when asked)i dunno why yours didnt copy in during install
>>20199510
>>20199531I know what happened.. I accidentally used my older "user" folder from previous and forgot to override the new one for this. Sorry
>>20199534lol i was using your 1.4 workflow sorry about that hahahaha
>>20199534am i retarded
>>20199538well....i kinda told you guys to back them up before install, so you did what you suppose to =)>>20199541i recommend always using the latest workflow.....olders are there for nostalgia sake i guess
>>20199548post the whole workflow.....(and set lightning cfg to "1"set "total steps" -slider to 8 and "enable third sampler" -slider to 1
>>20199349when your done can you do wizard magic
>>20199554this is not a request thread
>>20199555lmao
>>20199534granted, when using blink stuff....3rd sampler effect is not much......as blink action loras has pretty per-determined actionswhat 3rd sampler suppose to do is, it should make the gen follow prompt more coherently, make movement bit better, make quality bit better.....with of cost of extra time
>>2019955311 minute gen with the settings you suggested. it was set to 6 seconds so idk but damn. looks great i think
>>20199574if you wanna bit more smooth.....set interpolation to "4" and video fps to "60".....again, takes more time
>>20199308What kind of image generation workflow do you want?>>2019957430xx card?
>>20199577lol that fucked my gpu
>>20199562here is comparison and explanation to all kiddos out there wondering wtf bag is talking about thereusing same fixed seedsame steps, reso, etcusing same prompt >Woman undress her top by pulling her top's straps aside and letting the black top fall down, revealing her nude saggy-titsAs you can see, when 3rd sampler is enabled it follows my prompts much better. This is what is called prompt coherency. Motion seems much more natural and quality seems bit better.At the end of the day....both gen did what i wanted....to see some boobies. Is those ^^ thing worth a almost double gen time...that is up to you. Lightx loras are awesome, but speed comes with costs.
i had comfy working on an old rig, it has like 16ram and a zotac gtx750ti but it doesnt even launch comfy anymore any suggestions to get it to work?
>>201997031) did you update comfy?2) did you make a change to the computer?3) when you said "I had comfy working", what were you doing with comfy4) how does it not work now?
>>20199750i did because it was recommended to updateno changesi was doing some image generationit launches and i get the error with it saying none of the boxes are checked in the help ui
>>20199851easy answer is to roll it back to the previous version you were running and see if it still works.
>>20195961anon thank you much for this. I managed to set it up on my 3070 rtx ti. I had to change the tripleKsampler for a wan moe ksampler and the results are great. Just wanted to wish you a happy new year and thanks for the new layer of masturbation i have unlocked.
>>20200528This is exactly what the purpose of this is... I wanna give you guys the tools and templates. I want you to change them to fit your needs
>>20200576yeah but your install makes it sooooooo easy to beging with. I also explored other ggufs that are merged with pron loras, and without your initial package it would have not been that smooth. ty anon and happy holiday season and next year
is there a prompt stash node ? i cant find it in 1.5
bump
>>20201902sorry forgot to put, could some wizard help out and get her giving a hand job
>>20201902>>20201910nope, wrong thread idiot
>>20201910fuck off nigger
Hey 1st.AId.bag, have you tried the newly released SVI 2.2 Pro for WAN? I heard you can make longer outputs with ithttps://github.com/vita-epfl/Stable-Video-Infinity/tree/svi_wan22?tab=readme-ov-file
>>20195961first of all, thank you for sharing all this! this is saving newbies like myself a huge amount of time. using your latest workflow with the default settings, and running the example from your screenrec, it is taking about 50 min to render. my system has a 3060 with 12GB VRAM + 128 GB RAM. the end result is looking very good, but is that around the time that it is expected to take or have I messed something up? can you suggest settings to play with to reduce render time without compromising quality too much? thanks
>>20203110Looks interesting, gonna check it out >>20203414That's way too long..... Post screenshot of whole workflow so I can see the settings
>>20195961the https://files.catbox.moe/ldk14g.rar isn't loading for me. I downloaded/installed a prior version (I think 1-2 versions ago maybe?). If i just want your latest workflow 1.5, can i grab that somewhere? or do I need to reinstall everything?
>>20203748I'll upload it later tonight when I get home and mirror the pack somewhere... Link is working fine for me
>>20203801I followed the instructions exactly and when I run the file to start comfy it finishes with "Press any key to continue..." and nothing else happens. No other window opens.
>>20203863What does the console window say before that?
What’s the best way to ensure facial consistency? I’m trying a lesbian kiss scene using French kiss and NSFW loras, but the faces get more distorted than I would like. Tried it on v1.3 and v1.4 w similar results
>>20203944Lower the Lora weights.... Or use different lora
>>20203730let us know how it goes if you decide to try out that stable video infinity
>>20203730here is the workflow, i haven't changed anything from the default settings (beside the prompt)>>20203863i had the exact same issue. installing the latest studio drivers using the nvidia app fixed it for me
>>20204226and here is the log file: https://pastebin.com/ZrA6Ngzj
>>20204226can't find a workflow in that img, or am I doing something wrong?
>>20204384metadata gets stripped from images when posted on 4chan, which is why people request catbox uploaded images, which preserve metadata and by extension the embedded workflow
>>20195961thanks for this, is it possible to store the loras on a different drive and just change the filepath in comfyui?
this is my first time using this and i'm retarded. how do i open the actual application? i downloaded everything already.
>>20204947first day for me too, find the file called run_nvidia, in the comfyui_apprentice_portable folder if you installed the attention then pick the run_nvidia_gpu_SageAttention, otherwise pick the other one
>>20203748pack mirror:https://mega.nz/file/Gpw0jCaT#PacHBlAxhthHRP4kcwjO_KxCdWfqSxyHWedruxMbiJIworkflow 1.5dhttps://mega.nz/file/uxAThLqT#j5EaInkyWoJiKAd-5wAmgN4FBxdycpyLJZOHjoRPnbk>>20204226 >>20204283All seems good, aside that the models aren't been loaded into vram. Dunno why. Have you tried other workflows?>>20204733yes.....there is tool for that made by ivo in add-ons/tools
>>20205186thanks, I got it installed today, and thanks to your archived threads, i got all the info I needed to get it running, but it put it in and empty 250Gb drive and now with all the loras the model folder alone is 177Gb, so maybe if I move that I wont have issues later
I'm having issues installing using this. Git installs when I run the script. Then I press 1 on the options screen. Eventually, it tells me git.exe is not recognized. It also gives file not found errors and makes me confirm if a file is actually a file or a folder. Eventually it gets back to the menu where you choose a number. Choosing 2 or 3 results in the error "ComfyUI" folder do not exist. Install ComfyUI first!Not sure how to get this to proceed
>>20205554Screenshot of other error
>>20205554Actually, I think I got passed this. My connection uses a VPN directly in my router. Something must have been getting blocked. I did a firmware update on my router and disconnected VPN and it has gotten farther in the process and is currently downloading files. Hopefully it is successful
>>20203414Jesus! And here I'm getting a stroke after waiting 5 min on a result that might or might not be decent:O
Been getting less face retention after using these new workflowsI have been using pisswizard workflow which worked fine beforeanything that changed in these new ones?
>>20206387I find myself returning to that workflow as well for similar reasons. Even with identical models and settings, as much as possible at least.
>>20205186If I’ve saved as new workflows based off your old 1.3/1.4 versions, can I just copy those over to the new install folder? Or is it more complicated than that? For example, I added Lora’s to your v1.3 to get a solid riding or missionary i2v set prepped, can I just copy over those json files? Or is there a way to just load your new 1.5 workload instead?
what lora are people using to make the nudify gens ? like where the girl is clothed and then suddenly her clothes vanish.
>>20206923sure you could do that......i think you are making stuff more complicated =)you don't have to save every variation of workflows....much easier way is to drag your generate d video into comfyui....it loads the used workflow, prompt, used loras, settings, etc etc
>>20206995Oh no way! Lol I had no idea I could do that… going to try that when I’m back at the computer. Don’t think it’s worthwhile downloading your latest version for the updated workflows? I have a 5070
>>20204222https://screenrec.com/share/KgI1ECbtlf
>>20207017so this is one run go run with SVI lora and setup as i understand it...have to say scene transitions looks pretty nice, hard to say where scene of 81frames ends and another starts
>>20207031Looks sick! Mind sharing the workflow?
Bump. Great series of threads OP. Ty.
>>20207031looks nice, what's the time to generate all that?
>>20206942there are so many of those, but they keep getting deleted from civitai for instance, so google for them or get them on civarchivei2v-tittydrop-e26 is a classic stripping loraNude Scanner Wan 2.2 is a fun one I think: https://civarchive.com/models/2150906?modelVersionId=2432765
>>20207031whoa this is incredible. how long did it take you to gen??
so far its been hit or miss on the tattoos for me, using the prompt no tattoo or clean skin works sometimes, does anyone have a better way to word that?
>>20205186i have tried your older workflows and their are quicker, about 10 min gen time. but when using the same settings as the latest workflow, it goes back up to 50 min. can you recommend another workflow to try out and compare?>>20206380yeah, at that speed it is not being very practicalcan anyone with a similar setup (3060/12GB VRAM) share their gen time and the settings they used? so i can have an idea of what kind of performance to expect
>>20209471well only major change to the latest workflow is the sampler.....that might the issue for you as it doesn't seem to load models into vram/ram. Dunno why that is. But if the older ones work thats fine. I have 4070 (12gb) and 64gb ram, and 5cec gen @ 720p takes about 5mins>>20208435generation speed is the same as ^^ mention above, so each scene add 5mins to generation time (on my rig) lowering the resolution makes it faster ofc
>>20209940how to use your svi workflow do u have implementation??
>>20209471I've given up on doing effective video generation, because like you, I was 3060/12gb vram, and at 10mins a gen, with a high failure rate (unless I used an established and proven prompt lora combo), I said nuts to it.And that was with like Bag's original workflow, like version 1.0I tried like the second version, and gen time went to 20 minutes, and then newer version of comfy broke the NAG nodes, which again, contributed to be abandoning video gen.And I'm not dropping 300$ + dollars on this hobby\porn addiction anymore
>>20195961Thanks for the update!The video preview seems to be gone? It still shows up in a older version
>>20210395its there....now under options.....go setting and search "live preview"
is there an example workflow for the svi vids?
>>20210397come on please dont ignore
>>20195961Hey Aldbag. First of all thank you for all your good work, i've been able to get a good grip on Comfyui !I don't know if you'll be able to answer on specific Loras, but I have an issue that I can't manage to work around.I'm using the Blink Loras from Igoon. They are great to use, but i'm trying to use them in conjonction with others Lora to generate anal sex scenes. I tried adjusting the prompts with the correct trigger words associated with the anal specific loras, tried differents anal loras for Wan2.2, prompt modification and adjusting lora strenghts but it seems that the blink Loras are more "Powerfull" (in lack of other terms) and I can't manage to get an anal scene.I've combined loras before to good results, but with blink Loras I can't instill an anal scene. might that be due to the specific training on these loras ? Do you have advice on combining loras to get the best results ?
I jumped into ComfyUI for the first time with the previous, vol 8 and since then i've had an excellent experience. Running a 3060 Ti and being able to generate 5-8 second clips with great quality in 10-20 mins (depending on extra loras) is insane. Now installing this volume separately for testing the new stuff. I don't have any questions just leaving feedback and appreciation.Thanks Bag, you're a real one!
BUMP
>>20211966Ignore? What >>20212330The thing with blinks is that the action is pretty pre-determined/pre-defined. So they kinda "overrule" any other action (high part of the Lora) you try to achieve. I think there are some insertion loras that has better "anal support"
>>20212614Would you mind sharing the workflow version, settings and LoRAs that you are using? I can't seem to get the output to adhere to the prompt too much.@Bag With the Qwen Image Edit workflow, I can't get it to generate proper cocks. It always looks like, well, 'animal dicks'. Do you know what causes this and how to fix it?
I'm able to produce this in Wan2GP using Wan 2.2 with Lightning v2 and setting the lora strength to 0.1. 8 minutes to make 5 seconds with a 2060, doesn't seem so bad. Any tips on how I can push speed and quality even further?
>>20213633cocks are pussies are as good as they are sampled into the model itself or added by some lora. QWEN edit has pretty poor nsfw support.
>>20213732>speed and quality>ANDain't going to happen......one of those moves into the opposite direction as other raises or lower. You can get more speed by lowering the quality (resolution/steps). If you wanna get more quality, that costs more time.
>>20213757Say I don't want to wait more than 8 minutes, does it look like I'm already maxing out the quality potential of a 2060?
>>20213765sadly i don't keep spreadsheet of every gfx card's max potentials and what gen wait-times acceptable for each anon out there. that is pretty good result for 2060 card thou.....i don't think you could add much more resolution or seconds in the before OOM'ing out
>>20213737Is there a better edit model for NSFW, or are LoRAs the way to go here? (Do the Qwen LoRAs work with the edit model?)
>>20213803>(Do the Qwen LoRAs work with the edit model?)if i recall, no. Edit models are kinda like inpaint models for SDXL for example. SDXL has loads loads loads of loras and different models. ZiT seems pretty popular model now.
question - does anyone else have an issue with creating any single item longer than 9 seconds? if go over 9 seconds the output is grainy, but if I leave it at 9 even with all 4 extends on, it works great
Hi, I have a question. Is there any workflow for img to img nudify? I've been looking around google for a while but the ones I found are either old or don't work. Any help is appreciated, thanks.
>>20213915have you tried changing the output file on the current workflows he provided?
>>20213998Oh now I feel dumb, gonna try that, thanks anon, much appreciated!
>>20214005im new to this too so I dont know if it will work, let me know :)
I got an error but I am pretty new and can't make heads of tails about what's wrong. Any help please? TripleKSamplerWan22LightningAdvancedStage 3: sampling failed - AcceleratorError: CUDA error: invalid argumentSearch for `cudaErrorInvalidValue' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.For debugging consider passing CUDA_LAUNCH_BLOCKING=1Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
>>20214314Did you?[] make a fresh install from the current pack version?[] did you install attention stuff like sage and triton?[] mention what workflow are you talking about?[] mention what kinda rig (mainly gfx-card and ram) you have?[] took a screenshot of your whole workflow and posted it?That error code used to output when Comfyui updated and GGUF loaders didn't....but that was in the past. That issue should be fixed now.
>>20195961Thank you for your great work.
>>20214408I was using the past version of this. I am gonna delete it and download this one instead. The old workflow was working fine, the output looked a little blurry in parts but it was ok until it gave me that error. Only thing I did different was add two more scenes (for a total of 3 scenes).Here's my specs. I am downloading the new installer rn. Which WAN GGUF model do you recommend? I was using Q5 before but it was a little unstable, should I go down to Q2?
>>20214455Q2 is absolute shit.....maybe usuable from anime stuff, not rl stuff.....You do have nvidia card? RTX one?
>>20214484Yes, I have a RTX 3060, forgot to mention it, sorry.
>>20214494This is the kind of output I get from Q5, looks good enough but there's blurriness on the hands and bikini as they move. I am willing to take the risk and stay on Q5 if Q2 sucks too much.
Call me dumb, but looking at your workflows, I cannot figure out how 1stAIdbag_SDXL_IPA.jsonI supposed to work.Looking at it, it seems to be a faceswap\img2img generator, but I put in a face picture, and get the same picture as an output.What is this for and what am I missing?
>>20214507post the screenshot of your whole workflow....maybe you have some other settings fucked up
>>20214534Dang it, I deleted everything to install the new one. Once I am done installing I'll make another test video and come back to show you, thank you for your assistance.
>>20214543>deleted everythingi really hope you backed up your models folder before that
>>20214589Never occurred to me, I may be retarded. Downloading the models again rn.I have fast internet so it should be fine.
>>20214602>Never occurred to methat why wrote the instructions......oh well =)
>>20214609can you share workflow for this>>20207031
Been out of the vid gen loop for some time.Took the old install I had, updated Comfyui, updated the nodes (specifically the GGUF ones), and pulled your latest workflow.Everything SEEMS to work, but I get a windows media player visualizer output.As stated, not a fresh install, but an updated install (3.37)Workflow used to genreate attached vid:https://files.catbox.moe/u6q42i.png3060, 12 gig vram.Sage correctly installed.No obvious errors (of note, to me) in the terminalhttps://pastebin.com/xrnC8i98
>>20214742>https://files.catbox.moe/u6q42i.pngprobably posted a wrong png..can't see the workflow, just sample of the output
>>20214777I ain't gonna lie, I don't know how you people take legible screenshots of the whole workflow, so I just posted the output with the embedded png
>>20214782Snipping Tool (Win + Shift + S)or just hit printscreen and paste it into paint for example
>>20214782it needs to be as large so that the text is visible and most of the important nodes seen there (aka nodes where you can change things)
>>20214809ignore the red outline, I loaded up my standard instance of Comfy that doesn't have all the necessary nodes for vid gen, and have a queue going I don,t want to interrupt..
What's the difference between run_nvidia_gpu_SageAttention and run_nvidia_gpu?
>>20214809why you are not sharing svi workflow and ignore??
>>20214862i will share it when its ready, svi pro has been out for over a week....if in such rush grab one from civit. I don't ignore, i don't want to put something out what i don't really understand myself.....because tomorrow you would be here asking why this doesn't work/do what i want it to do
>>20214609I am back, I ran into a little problem. I got this notification when I started the workflow. I looked for the nodes on the library to no avail.
>>20214847why you have queue going if your outputs will end up looking like winamp visualizers?
>>20214937that node should be pre-installed by installer...
>>20214953Should I rerun it? and if I do, which option do I reinstall?
>>20214944My original video workflow instance uses actually one of your oldest workflows and runs on a portable Comfyui 3.36 I never touched once it started working. (It isn't even included in your current download, so it must predate 2.1 v1.3)After that version, I tried some of your upgrades, but didn't find a meaningful difference for my needs, so never upgraded, just tested future workflows every now and then.I just wanted to see if there was a siginificant difference in gen time or gen quality with your newer workflow, since I haven't done video gens since...(checks notes) September.Also, I have an instance for image generation because I can't stand my peanut butter touching my chocolate.So, in general, I have my video instance, image instance, and testing instance.
>>20214979......what exactly did you do? did you follow the instructions? or just install the comfyui and used manager to install the missing nodes?
>>20214989>>20214944So yeah, if you want to pass on the troubleshooting, that's fine, usually I think I'm better than most people in this thread, but this one is throwing me as most unusual an error and I just didn't want to go through the trouble of wiping my testing instance and doing a complete reinstall using your installer, I was hoping I was smart enough to upgrade my testing instance, but apparently not.Maybe on the weekend I'll wipe the testing instance and try your installer proper, but every time I do that, it is more of a hassle than I think it is (looking at you, reactor and wd14 tagger)
>>20215005I installed things in order, I only skipped number 6 "Download QWEN edit model files", since the instructions said it was optional.
>>20215017well the aim of the whole installer is todo it fresh, aka for new kids.....to get everything as ready-to-go as possible. If you somehow managed to install stuff from new installer, there would be loads of mismatches due different torch and cuda setups.
>>20215025is the a ComfyUI-GIMM-VFI folder inside your comfyui/custom_nodes/ ?
>>20215040Didn't even touch the installer, just the workflow .json.Python 3.13.6cuda is 13.6torch is 2.8.0+cu129
Call me insane, but would the orientation of an image affect its gen time?All four of these gens have identical loras and similar prompts, but the vertical ones seem to take twice as long as the horizontal ones.Confirmation biais, retardation, or something else?Note to self; test square images
>>20215051Yeah, I even reinstalled the nodes just in case. Gonna try to reset my PC to see if that does something.
>>20215077PS, they're all 680x1024 and 1024x680
>>20215055new installer installs latest comfy build there is with python 3.12, using pytorch 2.9.1 and cu130. If the older workflows gen ok for you then i think the new triplesampler might be the issue for you, maybe it has dependency of better torch than you have.....but like say, not sure / dunno
>>20215084Nope, still stuck, idk what went wrong. I'll try to reinstall the ComfyUi and addons again to see if that fixes it.
>>20215117when you first time run the installer....did it run the cuda toolkit 13.0 installer?
>>20215137Ok last time I installed the last workflow from past threads I went through the process of installing this, but then got this warning. I ignored it and the workflow still worked, this time around I didn't get this installer. My guess was because I already had it installed. Thing is, since then I installed Visual Studio and I still get this warning. I ran the installer again to see what would happen and still says I don't have visual studio installed.
>>20215187I looked it up and apparently Visual Studio 2026 is not supported by Cuda yet, so I have to download the 2022 version. Is this required? It's a big download so I wanna make sure before committing.
>>20215077you would think that it should not matter if the resolutions match and pixel count is the same but:Diffusion models break images into latent grids. In portrait orientation, the latent grid has more rows than columns, which makes: Generation less cache-friendly, more memory reads per step (=slower steps) Landscape layouts usually fit memory access patterns better. Optimal and fastest gens you can get when using 1:1 square ratio inputs. Example crop the image into 1:1, or use node to make empty latent that is square or near it and resize/and crop your input-image into it
>>20215200no, its not.....you need just libs from it. Kinda odd you don't have the redist installed. Many programs (or games kinda require that)https://learn.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist?view=msvc-170#latest-supported-redistributable-version
>>20215228I think I had an old version of that or something. I deleted the custom nodes folder and I am reinstalling to see if that works. If that doesn't work I am out of ideas. I may go back to the previous workflow.
>>20215236While reinstalling I noticed this, idk if it's related, doesn't look like it but it doesn't hurt to mention it.
>>20215236you need that cuda toolkit to be installled in order that gimm node to work....sadly that can't seek the needed cuda files from your comfyui embeded python folder. I made alternative workflow that doesn't use that (rife version) so that is not that big of an issue
>>20215242there are some errors during the installation, most of them are not so important as they got resolved later during the install
>>20215248That's weird, I installed Cuda and everything. But nonetheless I selected the rife version and that one didn't give me any errors while opening it it. Some nodes are disable, but I am guessing those are optional? Partial screenshot because I can't zoom out the entire thing without loosing the text.
>>20215267well it didn't give you that error of missing node because that version of the workflow dont use it.....you can enable sage nodes if you installed it. re-select the models and other files and hit run
>>20215274>sage nodesI did install them despite having no idea what they're for, but I am definitely gonna look it up tomorrow morning. For now I gotta go to bed. I thank you so much for your help and insight, much appreciated!
>>20214910ok apologies to u bag. sorry
>>20199574>>20199343>>20199562how do you do these where it instantly transitions into a handjob?
question to the wizards, what loras would you like to exist that don't currently ?
>>20216319those instant cut are commonly called blink ones (by igoon) They were removed from civitai because their bullshit nsfw policies.... but you can find most of them at https://civarchive.com/search?base_model=Wan+Video+2.2+I2V-A14B&q=blink&rating=allnew ones igoon only shares behind paywall, since there is no safe place to upload i2v nsfw loras anymore
Any ideas for prompting a pov missionary angle like this? I've tried body "jostling, moving, bouncing, shaking" up and down but nothing works.
>>20216627what are you after? the angle of the camera or physics of the boobs? (which look pretty weird imho)if angle "camera is top view of a woman laying on bed. woman's lower body stays out of the frame" ...or something like thatif you ment boobs moving, i belive there is some lora out there for big boobs movement in civitai, that make huge juggs move more natural
>>20216704That exact angle but her body moving up and down like she's getting fucked.
>>20216715Forgot to say this is for i2v as well if it changes things.
>>20216715could be as simple trick as "crop da frame" ?
