Excellent model, can you please quantify GGUF versions? It would be a great help for people with low VRAM like me.

Hey, its great job buddy, did you intent to make GGUF version?
i had problems that when on Ksampler. Comfy does randomly disconnected with notice "reconnected" on cmd has no problem.. just stopped running or show new info..
any solution? im using your workflow too.

its 32GB ram and RTX 4060 8GB.. (usually had no problem but its happend that i decide use your safetensors)

Masters always achieve results in a simple and efficient way.

Can you tell us what you trained it on? What was the purpose of the training? How many videos/images? Realistic or stylized? Did training teach it some new concepts? Ah, so it is a base model with some loras merged in. What did you merge in then? Does it understand NSFW stuff (and which)? What lightning lora did you merge in?

Stupid question: what's the difference to the plain WAN models?

Please make a GGUF version. Cannot run this with limited VRAM.

I wonder, what loras have you merged here? i hope you readed every licensing per each lora.. coz i dont allow merges

Hi Digital-Pastel,

I just wanted to tell you how impressed I am with your work. You are undoubtedly one of the best artists out there right now, and your simple and fast workflow is truly inspiring. Your creations are always of exceptional quality and showcase incredible talent.

Thank you for all the time and effort you put into your art. Keep it up, you are a real inspiration to all of us!

See you soon,

lighthorsexajz830

Can we get a GGUF?

you should list the loras used in this merge... even more so if you use the early access program.

"No Loras were used to make the videos on the showcase. Try to make a video without using Loras first.

When using Loras, start by setting their weight between 0.3~0.5 and increase it if necessary."

-- here you missing the following:

LIST OF LORAS MERGED IN THIS CHECKPOINT:
1.
2.
3.
4.
... and so on

Be fair, don't monetize other people's work.

I've been toying with different wan models, loras, and workflows for like 2 weeks now, all with mixed success. it's way harder to iterate on these video models because they're huge and even a 480p test takes at least a little bit of time.

from the ~10 videos i've created so far, this model should be considered the baseline for a consistent, highly performant wan 2.2 i2v 14b model. it's EXCELLENT. thank you for creating this! i was able to strip out most of my loras and the motion issue is pretty non-existent.

Oh, my God. I'm absolutely thrilled. This model (and, of course, the author's workflow) is flawless. I was afraid that there would be difficulties... but it's so Mid PC friendly! The results are excellent. Thank you for that!

Anyone else having issues with blowjobs in particular? Namely, the woman's face morphs, especially if the view isn't a closeup. Lips change color and she gets a "pac man" face with a lot of up and down jaw movement. I'm using the I2V models, 3/3, UniPC/Simple, Shift 7.

Edit: I'm a dumbass, apparently. I was using one lora for bounce effect and it was causing way more issues than I thought.

This model is fantastic. Please keep refining and tuning it. It needs more NSFW ^_^

I cannot consistently make a video of a naked woman standing still.

Most of the time when i try, she is still bouncing up and down slightly.

im a bit confused, i cant find smoothMixWan22I2VT2V_i2vLow.safetensors anywhere. i only have the file smoothMixWan22I2VT2V_i2vHigh.safetensors. what low file should i use instead? wan2.2_t2v_low_noise doesn't work together with this one, it just creates tv static

for 113 (7s) length video , the last second video always has camera reverse effect , reverse but not the same render of the earlier time

I am hoping this is worth the two dollar hooker charge I just paid

ModuleNotFoundError: No module named 'triton'

why?

Is it possible to get camera movement out of this? I'm having trouble prompting the camera to move in i2v worflow.

Pros: It's great. High quality.
Cons: Face consistency is fairly weak, characters tend to be constantly bouncing and jiggling around.

The fine movements are impressively accurate. However, when it comes to facial consistency or when the character is clothed, there’s a high probability that nipples or similar details end up being generated.

Can you get someone to make a GGUF of this please?

I absolutely love using this Checkpoint but one thing really bothers me. When I do cumshot videos (which tends to be quite often) the male always goes full on hip thrusting while nutting. This happens with every LoRA I've tried in the ranges of 0.30 - 0.50. Using just the checkpoint without a cumshot LoRA is not an option since the texture is quite bad. Does anyone else have made this observation? Maybe it's because of a baked in Missionary LoRA or something, I don't know. I would love to know, which LoRAs already are in the checkpoint. That would make "bugfixing" a lot easier.

does this model need to have NAG added to it for negative prompt adherence?

Incredible model, I like it.

Hello, do you have any idea to make this model into a lora separately?

Offers superior prompt adherence and dynamic capabilities compared to other mixed models.

good model！but just high noise

NVM, we were already told what to do if it was not coming out right, increase the CFG and try again!!

I love the model, it does quite a great job, but I am having the same issue that I see a lot of others noting, the bounciness is infuriating, I can't seem to craft a prompt that will get rid of the constant bounciness. even with just florence and a 2-3 words added in the positive prompt, it's still bouncy

Would it be possible to give more information about this diffusion model? I would like to know with what kind of data it has been trained with. sfw or nsfw? And what exactly the difference between this and the normal wan 2.2 i2v diffusion model is. I would love to know.

Absolutely the BEST model out there, I don't understand does this have lightning inside the model? I can generate amazing 5 second clips with great nsfw prompt adherence in 4 steps without using the lightning loras at all. I also use the k3nk workflow so I have been generating 20 second clips composed of connecting 5 second clips in the time it used to take me to make a single 5 second clip and also the movement, quality etc are on the level of the regular 14b without the lightning loras. Superb work!

The model lacks data on Asian women and overall leans towards Europe and America

Please release a .gguf version.

After using one of your workflows, I've come to the conclusion that this checkpoint is simply amazing! I don't have to use any loras for the most part. It can do pretty much everything! The key is having the right workflow for this model!

I applaud you, this is the best I've seen here. The model's understanding of what I need from it is 10/10. On a 4060 ti, a 10-second video takes 200 seconds on average. Bravo. The only problem I noticed was the penises being too long. I tried to combat this by pointing it out in negative comments, but it didn't seem to help. Maybe the creator could offer some advice.

Unfortunately the first test is of poor quality compared to the model I2V I post my rendering to see the result :(

Without GGUF of all this models, it's totally unusable.

非常棒的模型，无需lora就能很好地遵循指令，我的视频生成时间节省了40% ！！ 期待后面产出 S2V 模型 ！甚至愿意donate ！

Незнаю как ты сделал. Но у меня NSFW делает без всяких Lora. Даже лучше чем оригинальный wan 2.2

Sorry but this is the reality IT MUST BE SAID THAT THE TEXT TO VIDEO WILL NEVER BE AT THE HEIGHT OF THE IMAGE TO VIDEO

gguf q8, please,

Does it incorporate accelerated LoRA? The test shows it's even faster than when I use the Q8 model with accelerated LoRA.

Great model for NSFW but the female anatomy in that area is impossible to look at during sex scenes. A total turn off to me. It's totally butchered. I had to mix this with the remix model to get better anatomy down there, although now the prompt adhesion isn't as good. Love the model but just wanted to point this out so it can be improved upon. Let me know if you think I might be doing something wrong.

Amazing model and workflow. Thank you!!

EDIT: I saw that other users have already reported this, my bad. How sad, it's a really annoying flaw in such a well made model.

Well, the model is really well made, my compliments, but I want to bring to attention a "problem" that often occurs in generations, bouncing breasts.

Probably given the NSFW nature of the model, in most cases (at least in my generations) the breasts bounce even when standing still, even when they shouldn't.

I don't know if it's just my problem, if it can be solved, and if you'll want to fix it in the next releases (if there will be any), but in any case, great work, really.

Are there gguf versions of this available?

I use the new checkpoint with the WAN2.2 Smooth start-end frame workflow.

The generation speed is amazing, but I found some minor flaws:
1. The characters seem to be too "dynamic", which is great in some situations, but not in other situations where the characters need to be more static.

In my experience, I can modify the weight of lighting LORA to control the degree of the dynamics with the original WAN2.2 checkpoint.

Don't know if there is any possibility to do something similar?

2. I am not sure if anything I done wrong, it seems that the last frame of the video will become brighter?

Hi! it seems that the file contains only the text to image workflow... Am I missing something? I'm looking for the image to video workflow...

For multiple reasons, it would be nice to have the list and applied weights of the merged LoRas.