Wan 2.2 update. The preview images were all created using the lightning loras. The workflow can be downloaded in the "training images" download.
The main part of the prompt is:
A woman is straddling a man and having sex with him. She bounces up and down on top of the manDepending on the angle of the camera, you can either directly see the genitals or not. You can add this to your prompt to try to get them to show up:
You can see the woman bouncing up and down on the man's erect penis.You can prompt for different angles:
{The view is from the front and at a 45 degree angle|The view is from the side|The view is from behind at a 45 degree angle|The view is POV}By default, the woman's knees will probably be pointed down, so if you want more of a squatting position with her knees up, you can prompt for that this way:
She is squatting with her knees raisedDescription
Stopped blurring the faces in the dataset.
FAQ
Comments (29)
Just finished spending all day with your BJ model, supposed to be going to sleep, now I am seeing this and I am fighting Temptations... Okay, just a few quick 30 minute generations and ill be in bed before I know it...
I queued like 8 and tripped the circuit breaker
Wish I had the compute to use these... keep up the great work!
Future request : PremiumBukkake Lora for hunyuan (will tip alot of buzz for this)
Thanks for making this! The movements are great but as you said the resolution is low. I had to use a powerful upscaler and face upscaler.
Yeah, the one difference this has over my other loras is I didn't include any HD images to go along with the video. I'm thinking that may be a critical step with Wan. I will try to update this at some point.
@dtwr434 This must be a key to many of these loras, because the resolution feels low by default much of the time. It would be a shame to have to include the entire video, but maybe including the first and last frames in 2K would suffice.
Thanks for creating this!
What workflow are you using for generation? Tried downloading the attached videos but looks like it's not embedded there.
I'm using a pretty basic one just based off the examples here https://comfyanonymous.github.io/ComfyUI_examples/wan/. I just add in the lora node.
@dtwr434 That makes sense to use the vanilla lora loader for t2v.
I've been delving into wan i2v and confused myself because I'm trying to figure out how to add a lora to i2v to help guide the video. Still searching for how/whether it's possible to do this!
Ah ok. It's definitely possible as a bunch of people are doing it, but I can't say I've looked into it myself yet. People have uploaded their workflows to civit though, so maybe check some of those out.
@dtwr434 Thanks. I found that one and realized that I could just add the lora too. I should've posted back here. I'm still need to experiment more before reporting on how well it works.
"Stopped blurring the faces in the dataset."
I assume it was causing an issue and not creating random faces as intended? Is this a Wan vs Hunyuan difference?
I suspect it might be. I've had at least one person mention running into issues with blurred faces in the POV missionary one at high resolution, and I definitely had some issues with very weird looking faces with this one. But it's unclear to me if the blurring is at fault, or if just the overall low res look of this lora is causing part of the problem. I'm guessing it's a little of both.
Currently I'm thinking for Wan, you need to provide a set of HD images along with video, to prevent this low res look, and it doesn't handle blurred faces well. The actually ideal way to deal with faces is to use the masking feature, but I don't know of any good way to create masks, and I'm sure not going to do it by hand for each frame of video.
@dtwr434 What kind of mask do you need? Maybe I can make a workflow to help you with this, if it will make Your life easier to create those amazing Loras...
@Sam_A Hmm, this is what I was thinking of, but now that I'm reading it, I wonder if it only works with images and not videos: https://github.com/tdrussell/diffusion-pipe/blob/main/examples/dataset.toml#L46
@dtwr434 I see... If you want to test with a video, I can provide a workflow for you that will generate the video in same resolution of original with the faces masked (inverted), so you can test it. Better than do it manually.
I get weird result (flailing etc) when trying to use this i2v. Is this primarily meant to be a t2v model? I'm using the pinokio version of WAN 2.1.
I've only tested it t2v, but based on comments from my other loras, I'm surprised if it wouldn't work well with I2V. Maybe try lowering the lora strength? I have yet to try I2V with any of this stuff, as I'm just trying to work through my Hunyuan loras to bring them over to Wan. So, unfortunately, I don't have any advice or a feel for how well it is meant to work.
I feel like overall, this is the worst quality lora I've made, so I will need to revisit it either way at some point.
Actually, someone just left a comment on my blowjob lora saying they're having trouble with pinokio with that one too, and I've gotten good feedback for the I2V capabilities of that one. Maybe there's something about pinokio that is making it not work well with these loras.
@dtwr434 Thanks, this helps. I'll test and investigate further. If I find out anything, I'll post here. thanks for your work and reply!
@dtwr434 following up. It might have been the version of Wan2GP on pinokio. As of the March 13 update which specifically addresses loading LoRAs, I'm getting great results now. Using CFG 4, shift scale 6, 20 steps. On my 4080, can get up to 150 frames. Fast and coherent. Thanks!
@dtwr434 Hello, the problem with Pinokio is that you can't control the weight of the LORAs; by default, they all have a weight of 1.
@cuyjin it defaults to 1, but there is a field to enter weights of loras. For the pinokio (actually, the Wan2GP) install, I'm pretty sure it had to do with broken LoRA support before the March 13 release. Everything seems to be working now.
I think T2V lora will work on I2V, but not gonna be as good as lora specifically trained on I2V.
@ltcdrbroccoli can you give me a tutorial dude please i am trying in pinokio to but cant seem to get the image to work they just undress then nothing happen i load lora in advance mode do i have to do anything else
Very good LoRA. Thx.
whats the key token to speed up their hip bounce? their movements kinda sloppy in my output. "bounces up and down fast" didn't work.
Hmm, I don't have any specific caption for that or anything. The lora should just be taking care of it for you. What strength are you running it at?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.