🚨🚨🚨MISSION CRITICAL: PLEASE READ🚨🚨🚨

@CubeyAI will try it. thx !

@CubeyAI I'm trying with i2v. It took me about 200 tries before I found something that worked as requested, but only about 3 times was worth it. Are there any scenes that are rubbed out rather than inserted into the learning materials? The penis comes out as if it is being rubbed out. I'm sad that it didn't work out...

14B I think, he said it's compatible with T2V and I2V, I2V is 14B. Also version number is v014

Did you use the "WanVideo Lora Select" node to load the LoRA (if using Kijai nodes)? If that's not it, maybe try updating comfyui and the nodes

Ughh there was an old version of WanVideo Model Loader saved as -Main without the lora. Had to delete that folder because it was blocking the new node out.

@CubeyAI thanks for the response. I was using a photorealistic image, do you believe the lora will only respond well to animation?

Same with me. Doesn't seem to have any effect on any of the realistic input images I gave it
*big sad face

@CubeyAI don't sweat it, i've uploaded worse. I'm pretty addicted to video gen, so I'll be testing the update shortly.

@gurilagardnr If you get good results please let us know the settings! Ive only got 1 sorta coherent generation the rest are just a mess

@basedbase I'll have some examples to post tomorrow. I've had good results so far with reverse cowgirl. It's about 1 in 3 generations, but that's livable. My biggest hint is translate your prompt into simplified Chinese, including the trigger phrase "penis goes in and out of vagina" 阴茎进出阴道 Missionary is ...ok....sometimes she rips the dudes dick off or stretches it out, which is horrifying and hilarious, but the LORA definitely provides penetration where none was possible before for realistic images.

@gurilagardnr lmao also noticed that or the guy going at light speed. Also had one where a guy just battering ramed her with a penis. Like he held it like a log and rammed her lmao.

Setting it to around 0.6 works for photrealism.

@CubeyAI If I dont mention the penis and vagina prompt they just have facial animations and basic body movement and if I do mention it its very glitchy. This is with and without the lora way worse without the lora.

@CubeyAI Honestly if wan wasn't as glitchy as it is I would be tempted to pick up a Rtx 8000 48gb for training LoRa's for it

@basedbase give it time, new shit always goes through it's growing pains. In 30--60 days, unless something really amazing comes along, this is the one. It's the only one fully trainable. Shit's about to get real.

@gurilagardnr Hopefully would be nice to finally get good AI NSFW animations

What other loras? I can't even see a category here for WAN let alone lots of loras to try?

@CubeyAI Are you sure this Lora is the correct one? It is very small compared to other WAN loras...

@azeli There are other loras... Just search wan lora

@Lady_Valeria There isn't even a category here for WAN, if I search "wan lora" there are 3 loras currently including this one.

@azeli There you go, there are three :)

@CubeyAI So there is an underscore? is the doggystyle tag also doggystyle_sex?

@Fleshcrafter I have still yet to get it to work even using their workflows from dragging the video to comfy UI and used suggested prompts. Always results in either pure hallucinations or the legs and penis flailing around no Idea why nothing works. About to test with just a very generic AI hentai missionary photo and if that doesnt do anything there is something seriously wrong with Wan 2.1's ability to do NSFW.

@CubeyAI Ive tried 14b 720p and 14b 480p with the exact same issues. Genuinely confused tried every combo of things I can think of.

@basedbase I've got a decent pearl I posted. Should have all the meta's intact to drag into comfy. It is possible! There be gold in those steps!!!

One of the beliefs I've arrived at after maaannnyyyy nights being up till 3am cooking my GPU is that each image has its own soul of sorts. Its own sort of implied physical action that the Ai picks up on. Some images never work, and some work very well even on a variety of seeds. There must be some subtle implied movement within the resting state of the photo's anatomy that is either anatomically correct and leads to a good video or the implied movement doesn't actually fit what is possible within the anatomy and thus the video comes out twisted. So my main suggestion is to try a couple a seeds per photo on a lot of different photos. These are just my autistic musings and not professional opinion.

@CubeyAI Personally, i've only had success with the 480p model. Get a lot more bizarre stuff and flashing lights with the 720p.

@Fleshcrafter I have come to the conclusion that its most likely not possible for it to do anything furry that is not photorealistic. As I am 99% sure there is nothing like that that it was trained on. I spent countless hours and I am pretty sure im right.

Please tell me how many videos were used to teach this LoRa?

@CubeyAI Damn 700? Did you train the 14B model to get the LoRa or the 1.3B? DIffusion pipes docs say for training 14B you need alot more than 24gb of vram. Currently testing training LoRa's on the 1.3B to be used with i2v 14B.

@CubeyAI and how long did the training take and at what GPU and VRAM did they train?

@CubeyAI Thanks for the update! It now somewhat works I uploaded the i2v generation I got to the gallery but it did not fully follow the prompt. Any suggestions to get something like the animation you got in the first video you posted?

@CubeyAI Have you tried changing denoise from anything but 1? In SDXL denoise changing can help alot but im not sure how it behaves on this model

convert your prompt to simplified chinese. I definitely get better coherence and motion doing that

@gurilagardnr Ill try that been SLOWLY been getting better results. I don't think its an issue with the LoRa as ive tried without it and its horrific. Been getting better results changing seeds. If the seed is randomized you get only glitches but something like 1233 as the seed is alright just not good. Its really an issue with Wan I think the only good video gens ive seen with it are generic scenery shots and things like that.

@CubeyAI I think im going to train an i2v lora in diffusion pipe and see how it turns out. What are the VRAM requirements? I have 2 12gb GPU's so a total of 24gb of VRAM. I saw that you trained on 24gb of VRAM, id like to train on a large dataset of videos but im not sure if VRAM requirements scale based on the dataset size or just based on training a lora from the 1.3B i2v model vs 14b model.

@CubeyAI Currently training a 1.3b t2v lora hope it works for 14b i2v, initial results were still veery glitchy but alot better than using one not tuned for furry. Playing with Epochs and repeats 14 epochs with 10 repeats per epoch is the initial working one with a rank = 64 and 1, 15 frames cant go above 15 frames strangely because I have over 10gb of vram to spare if I up to 33 frames its allocation error. We will see this is all very new so got to experiment!

@CubeyAI Testing the workflow I got from dragging the first video into comfyUI completely maxed out all 12gb of VRAM and 64gb of system memory 45 minutes for one generation used to be 3 minutes wan is so unpredictable lol thats at 89 frames as well

The safetensors file is corrupt or invalid. Make sure this is actually a safetensors file and not a ckpt or pt or other filetype.

I mean to the workflow

Error while deserializing header: HeaderTooLarge File path: /workspace/ComfyUI/models/loras/wan-nsfw-e14-fixed.safetensors - This is the title of the comfyui bug report.

same error when starting the workflow with the 16GB model

Is there no mistake in the name of LoRa?

I apologize. Everything works fine on both GUFF and full-size models. LoRa uploaded with an error to the server from Comfyui

@ratamahaeva757 How were you able to get this LoRA working with the block swap and GGUF? The "WanVideo Lora Select" node outputs a "lora" type, but I'm not sure where that could connect to

https://civitai.com/models/1301129/wan-video-21-native-i2vandt2v-fastest-workflow-ggufteacache been having luck with this one

I too have idle 4090 that could help if I only knew how.

@NoneofYourBusiness2 Your not even using it? C;mon bro I'm dying over here and you're just idling it? At full throttle??? You have one of them there 48gb dual width China models? Just swimming in the vram???

@ToxicBot THB I left it running all night and came to 200+ WAN videos, maybe 5 of them looked perfect

Adding virtual vram fixed that problem for me.

@CubeyAI Hi, thanks for replying! I do have example images/videos with workflow metadata if you wanted to diagnose, but you're right that it's usually the first or last couple of frames with the issues.
If you're curious, though, the workflow I'm using (with slight additions/modifications) is this one:
https://civitai.com/models/1278171/optimised-skyreelswan-21-gguf-i2v-upscale-hunyuan-lora-compatible-3060-12gbvram-32gbram-kijai-wf

@CubeyAI Hey, I figured out the issue. It was a custom node in the workflow I was using that was causing the issue. LoRA is flawless as everyone else has said already. I edited my original post. Thanks, dude! 🎉

ty I thought my sampler setup was wrong with greyed stuff. gonna try without power lora rightree

Download the mp4 file and input it directly into comfyui.

@CubeyAI I think by preview he means clicking on it and it showing up in the left pane in-page.

@maaade prompts are not in the preview, they are embedded into the file itself, download/drag/drop into comfyui, and by file I mean the gif.

just need to install

I have a template here cloud that works well
https://civitai.com/models/1317373/runpod-wan-21-img2video-template-comfyui

Still need fast nodes with rtx 5k?🫤

@mikebobby681369 5000 series is only a roughly 4%ish increase in actual performance (like rasterization), however, they've doubled the AI cores so it's still technically a leap forward. In AI only... but they didn't significantly increase the size of the vram so the upgrade is minimal for us. It's mostly for making AI frames to boost all those poorly optimized games up to 60fps.

same here with RTX 4090.

sageattention problem could be avoided by switching atatention_mode from sageattn to sdpa in WanVideo Model Loader node, but Triton seems to only support Linux.

triton error message:

I don't know where to insert the python code given at the end of the error message.

ah, resolved. Bypassing WanVideo Torch Compile Setting node do the trick (although this may affect generation speed...)

doing it with "switching atatention_mode from sageattn to sdpa in WanVideo Model Loader node" in my comment above works for me. hope this helps.

@CubeyAI Thanks, finally I managed to make these work.

running commands below after installing C++ tools worked for my environment (comfy desktop).

.venv\Scripts\activate

.venv\Scripts\python.exe -m pip install -U triton-windows

.venv\Scripts\python.exe -m pip install sageattention

@CubeyAI does this workflow require triton and sage attention to generate?

@CubeyAI thnxths

download one of the gifs, drag and drop into an empty workflow, they should be embedded.

@zmikez you just blew my mind man!!! much appreciated!

@kantorid120 not every gif will have them embedded so it's kind of a crapshoot if it works or not, but most people who create loras/checkpoints will typically have a couple images that have their workflow embedded. Finding an amazing image only to find out it doesn't have the workflow included is peak sadness lol.

@CubeyAI bro please help🙏🙏🙏. I tried almost every video from here and comfyui didn't see workflow in any of them. I downloaded them as mp4 because that's the only way google chrome lets you download them

@crovelmf547 same problem here, files has workflow in metadata I can see that but when dragging into Comfy it shows "Unable to find workflow in xyz"

@CubeyAI I just managed to get it working totally randomly by installing some nodes from a workflow found on yt. I don't know which of these allowed my Comfy build to read mp4 metadata but it is one of these 3: “rgthree-comfy", "comfyui_videohelpersuite", "comfyui_essentials". Now I can get the workflow from your videos, I will play with your lora for the next few hours! :)

@CubeyAI Yes. As @seriousduck11 pointed out, the problem was that one of the extensions "rgthree-comfy", "comfyui_videohelpersuite", "comfyui_essentials" was missing. I installed them and everything worked for me too. I am a complete noob in this matter and I did not have them installed before. Please, could you list these extensions somewhere in the guide, so that other newbies could understand it easier?

@CubeyAI Thank you. I think i going to put this on a hold as the whole version and what not little overwhelming to me atm. I think i even have a non working sageattn installed. I'm going to re install my pc (want to anyway not just for this) and do a decent research about the whole thing. Thanks for the reply.

https://github.com/kijai/ComfyUI-WanVideoWrapper

You need to download them, they are not just magically there when you run comfyui.

Then you need to restart it and refresh your browser window.

hi, did you fix this problem and got workflow?

@crovelmf547 Yes it works. Just download a sample video posted here. Drag and drop it on ComfyUI. Don't know why it did not work for me two days ago. I redownloaded the videos and tried today: it works.

I had to install comfyui-manager (explained in its repo), then find the custom node manager and add "VideoHelperSuite"

@CubeyAI No just the safe tensors after switching every single gen is great whether using my lora or yours wan is not very picky anymore with the prompting as well, fp8 is the lowest quality and fp16 is the highest quality. It even works great on art styles it wouldnt work on at all before after swapping to the fp16

gguf version exist

@Chiefkor Link? Are you implying that the GGUF version will perform the same way as the FP16 without loss in quality?

Can you explain what you are talking about? I don't even see fp16 versions of WAN 2.1 i2v:
https://huggingface.co/Kijai/WanVideo_comfy/tree/main

@Clocksmith @Miramir sorry im so late, you guys find anything u want ?

@Chiefkor My question was to the OP but maybe your gguf response answers it?

@Clocksmith gguf version are quantized from the main repo not from kijai repo
https://huggingface.co/Wan-AI
If you need any gguf version just go on city96 repo
https://huggingface.co/city96
for wan I2V
https://huggingface.co/city96/Wan2.1-I2V-14B-480P-gguf/tree/main

@CubeyAI  All the videos I generated using your Lora are getting like in the example I attached, when you generate them they turn out fine, I download the workflow with your settings, changing only the resolution to generate faster, the example I attached in the link, and here are all the videos I made almost the same. i don't understand what I'm doing wrong?

@CubeyAI If you don't know how to fix something like this, I'll try it myself, anyway, thanks for the lora.

@CubeyAI thank you and good luck

@CubeyAI I figured it out note for any future users don't use comfyui portable

@goonempe Did you also follow the instructions in the article? Or did the workflow just work out of the box when you switched to the Manual Install version of ComfyUI?

use SDXL for NSFW.

I haven't gotten good NSFW on FLUX here, but I did on Tensor.art: https://tensor.art/images/796454041249935402?post_id=796454041245741101&source_id=nzy-o17ikkG0p_ArbXfy-xIn

(You need an account, and to turn NSFW on, or you get a 404 error.)

It's insanely LoRA dependent, and you need to know what they wrote in the captions. If they wrote book reports, you have to write a book report.

Some that work well on Civit are the non-face altering titjobs one, and bra on titjobs, and Blacked (can't say enough good things about Blacked, you get 4 poses, put a Bottom Queen or a Thicc Goth Gf on it, it freaking rules) I think, as they work nearly every time on TA.

With normal lora like the Standing Anal lora I linked you to, or the footjobs lora, it will give curse after curse, and then there will be a nearly flawless one. It's like one of 8 on TA. Here, it's one in a million.

Does it work on human anatomy/I mean on human physiology, if you know what I mean, on women, does it enhance the authenticity of the organs

speed seems to be heavily affected by a combination of the number of total frames and the frame rate. It also seems pretty consistent. The main issue seems to be if you combine this lora with other loras and they have different frame rates during training you'll get slowly bouncing boobs with super fast thrusts, or slowmo boobs with regular thrusts.

@CubeyAI So, you create your videos with Comfyui?

@CubeyAI Thanks to you, I now understand how this works.

I have one question. It seems to me that this LoRA model has been trained on motion. Is there any workflow or educational material available regarding the training process? Or any references I can refer to?

I haven't been working with AI for long, so I want to learn and experiment with various aspects. If there are any resources I can refer to, I would greatly appreciate it.

From what I see, FLUX has a corresponding LoRA for FLUX, and WAN has a corresponding LoRA for WAN.

So, I would like to train and create my own models for specific characters, motions, and NSFW settings. If you have any guidance or resources, Thank U!

@CubeyAI 

Thanks for responding.

So how did you pass the moderation filters? I'm trying to get some images I made animated with civitai, but it keeps getting rejected on what I'm guessing is them being pornographic. Which is crazy because I'm using their tools to create those images in the first place.

@CubeyAI ohhhhh. That makes sense. So civitai will allow the production and publishing of nsfw images, but not videos.

So then, it is possible to animate the same pictures that I made with with civitai? As in, I want to use the loras that I found, with Kling, because they produce results that I like.

(I'm watching the tutorials BTW, so I'm trying to ask questions that haven't been asked by a FAQ or video yet.)

@CubeyAI i see! Alright, I think I got it from here. But before I jead out, I was hoping if you could look at the images and videos I posted and give me your thoughts. I'm trying to animate those particular images and in that specific way -Kling or something as good or better than Kling.

It's a generalized nsfw lora for sex positions to make WAN img2video generations. It seems to be focused on missionary but can also handle other positions (with limited success).

Does not seem to work all that well with fellatio though.

@zmikez not only for missionary, it's for everything, it even makes boobs sway better than standard WAN. I think the dude started it as a missionary lora only, but in time it has grown into something bigger (thus the name change into 'General NSFW Lora'). I use this lora with all my gens

this is the best lora i have used thanks to @evilstormhot109 for sharing it with me and thanks to the creator @CubeyAI, it is indeed has better jiggle effect than other lora sex positions, go try it and enjoy

@evilstormhot109 awesome sounds great. Can it do realistic? what would be some good key words?

@CubeyAI https://civitai.com/images/67509193 Its from 720p result :) not perfect couse there are some artifacts like low bitrate or something but pretty good i think.

In your prompt, describe the person as "mute" and it will cause them to stop talking.

Thanks, I cant seem to find those nodes in comfy do you have the exact name I should input in the custom nodes search?

so i found those nodes but how is that lora being tied into your workflow? every lora loader ive seen has a model and clip input and output, where does the single lora output node fit into an img2vid wan workflow?

did you ever figure it out? it would help if you post the error message from your cmd terminal too

@PookieNumnums Not yet. It works fine without loading this LoRA, but I'm stuck when trying to integrate it.

whats your image res? and what model Wan2.1 you are using?

https://civitai.com/models/1496200/txt2image2video-wan21 use it with this works for both

@TheCollector2 this uses i2v 14b model

i believe it means somewhere along the way you have a mismatch between models vaes clips etc. did you ever figure it out

hey i figured it out trying to use the lora and build a workflow from scratch with no assist. basically, youre using the wrong model type. You are probably using either a) a vace model setup b) a text to video 1.3b model, or c) the wrong 14b image to vid model. you need to use the CORRECT 14 Billion parameter Image to Video model. At least thats what was causing the error for me. The model i swapped to that didnt get that error was the Wan2_1-i2v-14B-480p_fp8_e4m3fn model. I was trying the text to video models on an image to video setup because with the Vace and Fun model workflows you can actually get away with using the 1.3b text to video models in a 'faux' image to vid workflow by using a single starting frame. I had one working a few days ago but the quality wasnt the best and i didnt feel like trying to dial it in. Anyhow, use the model i mentioned and see if that helps!

https://civitai.com/models/1496200/txt2image2video-wan21

What example videos are you talking about? I don't see a workflow anywhere.

i missed something.. whats the difference? is there more versions of WAN?

mines better check it out

@CubeyAI 

on the page of the model
1 step- click in the three dots at the top right, next to the bell, click and choose edit model

2 step- roll the page until find commercial use and select use on other generation services

and save

@CubeyAI thanks =3

@CubeyAI That's a good question. 🤔 When I was training loras for SDXL the UI used allowed me to perform operations such as adjusting weights of blocks, merging loras and saving the new lora, etc. but the one you are using doesn't offer that function as far as I know. Hmmm... Perhaps linking the comfyui wan blocks node to a save lora node would let you save the lora without those blocks? I'm just not sure if save lora blocks would accept wan loras as an input.

is this true of all wan loras or just this one specifically?

@blip I tried turning the blocks off due to abovementioned reasons so I can't really say if other lora's that I've found wacky can be fixed with the same method. But I've used loras that work perfectly fine by themselves, without any block adjustments.

@CubeyAI Thank you for the heads-up. 👍 How long were the videos on average? What framerate?

@CubeyAI Thanks 👍 I've heard that using higher framerates than wan's default 16fps in the training videos causes slow-motion to sometimes appear in the generated videos (or maybe it is because the videos in the sam training batch need to have the same fps?). I have experienced the slow-motion effect too and some loras make everything slow motion, so I figure it would be best to change everything to 16fps if you intend to have more runs in the future (to be on the safe side).

@CubeyAI Sorry. When I attempt to create with this model it says “something went wrong:( try refreshing your browser” however when I do that I doesn’t do anything, I’m just curious if you’ve heard or seen this before. Thank you.

@CubeyAI I was just hitting the create button on the post, I’m new to this system. What ish the best way to use it. How would you recommend I use this if I’m on a Microsoft surface 9 laptop. I’m sorry for bothering you.

@CubeyAI I usually use OpenArtAi but I just discovered this so I’m trying to figure it out. Again I’m so sorry for the trouble.

@CubeyAI do you have an example workflow for this lora?

install rgthree custom node and add "Power Lora" node in the workflow

@RealHUGE thank U

根據我的研究 體位搭配適合(除了傳教士體位以外 其他體位建議再搭配其他體位LORA) 液體噴臉不太行會散射 乳搖效果很好 不用額外搭配額外乳搖

did you try unplugging and plugging in your pc?

I made an nsfw video and its blocked due to mature content wtf??? help me

Long story short, we click a button and a video comes out. That's all :3 xD

Same Issue but with image-to-video. It says the image is "blocked for whatever reason then change prompt" -_- ...

I got the same issue to and contacted support. Didnt want to waste more buzz, let me know if it gets resolved for you.

@AiGen1234 The issue has been resolved. Now, when generation fails, it no longer just shows a blank box—it displays the specific reason. I received a refund for the failed generation from a few days ago, so I'm not sure how long it takes for a refund to be issued now when a generation fails.

@ss15801538070269 , thanks

@CubeyAI Yes, it was rejected by the image-to-video service provider. However, Civitai doesn't show it as a generation failure. I don't understand why the restrictions for text-to-image and image-to-video services have become more numerous, stricter, and easier to trigger blocks recently.

i use gpt to translated

There's a new blocking feature on top of the regular ones now, too. They stealthily removed the option to use I2V for NSFW unless the image has metadata from a1111 or Civitai. If they detect the result or the prompt is NSFW and the image doesn't have that metadata they'll block the video and not give you a refund.

Yeah this lora doesn't work on site anymore, very annoying...

@CubeyAI  it continue with the "Blocked Video Mature content restriction. Please adjust your prompt and try again." message and now i can't even load every wan I2v lora like "wan POV missionary" or "wan cumshot", its tagged as unavailable. And ofc i lost buzz without refund again.

@CubeyAI I don't know! For a simple twerk video, they now consider it mature content and we can't even get a refund. I wrote an email to Civitai yesterday to explain the situation, no one replied, not even to confirm that my complaint had been received.

This happens if you have upscaled the image or using a custom / resaved image. The metadata in the image must be from CivitAI. You can read about this in the Status page.

They changed the rules a while ago. You're not allowed to make NSFW using i2v unless the image has been generated with a1111 or Civitai and still has the metadata.

It also triggers the 'blocked video mature' when using images made with Lora/Embeddings using fictional characters, made on-site. If they have a name that triggers the bot. Even though the celeb/real-people have been removed, an AI-person embedding I used triggered it. I think the name Imogen was the problem.

Check you don't have embeddings/loras in the final pic that you use, either as model or prompt. I'm guessing artist names might even trigger it now too, if my hunch about the 'imogen' thing was right.

I've never used an off-site picture for video gen, and never used any celeb loras so this is really fucking annoying. But once I worked it out I blew 20k on vids hah.

I have the same problem, although it seems people keep posting videos (on site generated)...

@QueenIS I'm seeing that too, could be a bug or it's only partially available to some, which would be strange.

@CubeyAI For some people! Let's wait and see...

@CubeyAI it's not on the tab of video generator options as of today, idk if it's a bug, maintenance or error.

I'm very sad!! Yesterday i'd very god NFSW gay videos...

the cost of video generated raised to 400+. it's crazy

@Muscle20 I'm sorry Mr. Muscle hopefully it's up now

@superwai777974 they can't be serious

why can still someone post video with wan ? are they paid member?

Try TensorArt, it's not as good as Civ, but fair.

@ToxicBot Will do thanks

@ToxicBot tensorart is not allow more than one lora

It’s disabled for me as well, I’ve just posted old generations before this happened to get some buzz lol. But yeah Wan is totally disabled from

What I can see

@RingGuy fkn sucks bro, it's the only one that does decent nsfw, lightricks is trash

we strongly demand civitai that wan be included in I2V again

It's been a couple of days, and we can't use the WAN generator. What's going on?

Sooo, what do we think? New censorship tools (my guess) or new charges?

@superwai777974 agreed

@ToxicBot I hope it ain't your first guess, that'll be the death of nsfw video generation.

UPDATE: It's back up but alot of loras have been removed including this one, they ended up moderating tf out of wan.

Wan generation is back, but I can't find NSFW wan. they maybe updating it or something

@God_Of_Thicc I hope thats not the case and they just updating the list

good news, WAN is back. I think they need to update the base model of lora so they can show up

@aiprompter2 I hope so

@superwai777974 let us hope 

@CubeyAI I hope you will update this LoRA for updated WAN, it’s goated 🔥❤️🙏

@QueenIS Many keep videos in drafts and post as we go (me included).

@CubeyAI  please update Base Model of your lora from Wan Video to Wan Video 14B i2v 480p so your lora can appear in new wan generator

@CubeyAI thanks bruh

its gone for me again. Anyone else having issues

@aiprompter2 hmm let me check real quick

@aiprompter2 it still shows up for me both the 480p and 720p but the Wan nsfw base model of cubey only appears on the 480p option.

@God_Of_Thicc It's working now. Sorry for the alarm. I am trying to get the deepthroat lora to show up in resources but it is not showing up, message the guy about it waiting for reply back

@aiprompter2 same it seems to have disappeared.

@CubeyAI thanks

@CubeyAI is there any good cloud comfyUI that you know? i dont have a powerful computer at home

You don't need this to run on your machine locally. This looks like it was made to get around restrictions on running models through civitai. WAN works just fine in comfy and the appropriate LoRa. No a1111, I love it and still use it for images but I couldn't get anything other than nightmare fuel with animatediff. Just download every LoRa you can fit on your drive before they disappear and try them out one by one. Rename them or you'll pull your hair out trying to find them.

@CubeyAI Thanks a lot. It took me quite a while to understand I don't need those letters A, R, U etc.

1. Dont use Multi GPU 2X 24GB, try using A or ADA single GPU 48GB or more

2. [ warmup_steps = 10 ] [ blocks_to_swap = 32 ] [ type = 'AdamW8bitKahan' ] [ lr=2e-5 ] also try longer training on 80-100 epoch save every 10

I dont know this lora author setting, it just share my experience training Wan2.1 14B T2V I2V and get stunning result.

you dont even need a lora for that. try something like this in the prompt:
"A woman takes off her shirt, revealing her naked body and breasts"
"A woman rips her jacket off, exposing her ..."
"A woman pulls her shirt up, exposing her belly button"

The trick is to also mention what is revealed/exposed, Breasts might not get nipples, unless you ask for it, or use a lora

same problem with you too. It did generate the action of removing clothes but somehow the clothes just stay on the body

Undressing LoRAs are banned on this site.

are you using i2v? this helps create motion vs generating limbs that are not there. however, there are other LoRAs that help with the ideas listed.

Noob_ee yep only img2vid, well since it says general nsfw I thought it would work on that lol, but even in motion are very still, barely moving

Noob_ee like what? need me a LoRa to generate a man/penis

xff343 these loras are mostly motion loras, the do not generate limbs. the limbs come from the image beforehand.

Noob_ee i tried on here again, and the head was moving at speedlight lol,usually a common problem is they either bite the penis or when moving back, its just half of it, any reccomendation?

alternative_Universe set the LoRA strength to around 0.9 and lower the output (save video FPS) FPS to 16. Also randomize the seed and try a few more times. also, i will be than welcome to run a trial on my end, ill need all the metadata and LoRAs to troubleshoot the generation.

I'm also using comfyui

CubeyAI open your example video doesnt do anything. nothing opens. ideas? thanks.

muelleimeradresse12629 you need to add a power LoRA to the workflow, plenty of WFs out there already built. I use the Power Loader node. the power lora loader goes right after the model loader node. let me know if you need more details. ✌️

follow this for 8gb vram and less and just remove Lora (pusa) which is given in the video and add the nsfw one

CubeyAI It does have some effect, been experiementing with it a fair bit. But it seems alot "weaker" on wan2.2 in comparison.
Some other loras that worked with wan2.1 are just straight up throwing errors on wan 2.2 and dont get applied, which I assume tends to indicate that the model weights for wan2.2 have actually changed?
I dunno how it works tbh though or if other loras just werent trained properly tho *shrugs

CubeyAI Sorry to disturb you. Could you put some materials for female masturbation in it, such as touching the vagina or inserting fingers