🤠💖Hip-slammin' assertive cowgirl!💖🤠
WAN 2.2 NOTES
(WAN 2.1 NOTES DOWN BELOW!)
Update- There are much better 2.2 cowgirl loras out now. Use this High Noise lora if you want the aggressive "hip slammin" body motion, but I recommend you rely on other loras for accurate naughty bits!
Trained on 2.2 I2V A14B model. Massively increased dataset over the 2.1 version. Other camera angles work much better now, but POV was still the focus of the training data.
Weight 0.7~1.0 ... 1.0 is usually fine
Some of the start images can be seen here: https://civarchive.com/posts/21124019
Sample workflow in the 'training data' download. https://civarchive.com/api/download/training-data/2129122
HIGH model is trained on the actual 'hip slammin' motion.
LOW model is trained only on genital close-ups.
My thinking is that this gives base Wan more freedom to work out her facial expressions in the low noise steps. The low noise lora only helps where it's really needed- the genitals... Whether this assumption is actually accurate 🤷♀️ but training both on the same dataset resulted in noticeably muted facial expressions.
Base trigger:
a woman is straddling a man and eagerly having sex with him. she is squatting.Helpful (highly recommended) phrases
(trained-in to the low noise lora)
The man's penis slides in and out of view as it enters her vagina.His penis is penetrating her vagina. you can see his penis sliding in and out of her vagina.Helpful movement phrases:
she moves quickly.she moves energetically.her hips move quickly down on his penis.she vigorously slams her hips down on the man.she is quickly moving her hips up and down. I also found that increasing the high noise CFG to ~6.0 helps a lot with movement on this lora.
Still not perfect- penis often 'smooths out' losing any texture/veinyness. And sometimes it just totally breaks.
You can run just the HIGH version, but may end up with accordion cocks (penis getting smashed under her hips rather than entering her pussy).
Since the lownoise lora is only helping with genitals, you can probably substitute other NSFW/penetration lownoise loras. No guarantees.
WAN 2.1 NOTES
DISCLAIMER: Trained on 14B T2V model. But REALLY meant for I2V. Listed on Civit as I2V.
I made this for personal use and decided to share. I offer no warranty.
Input images including prompts can be seen here. https://civarchive.com/posts/16809770
I love the other Wan cowgirl's, but I wanted something a little more aggressive and in-your-face.
So this one is trained on "squatting, face focus, assertive" cowgirl, where the woman slams her hips down on the cock while her upper body remains mostly stationary.
weight 1.0 works fine for most I2V. But try range 0.7~1.0
No trained trigger word, but I took some hints from other Wan sex loras for base prompt suggestion:
a woman is straddling a man and having sex with him. she is squatting. you can see his penis sliding in and out of her vagina.Optional 🤠🏇
she slams her hips down aggressively on his penis. Landscape-orientation inputs seem to give the most consistent results.
It does also work with reverse/ass-view cowgirl, but no anus data was trained, so beware...
Training data was mostly SFM clips, plus a couple anime clips, and some MMD clips. But no photorealistic, which is probably why T2V with this lora is so hideous.
Cock'n'pussy close-ups were included to reduce likelihood of cock horror.
If you do T2V, use weight 0.7 but you can still expect some ugly-ass remi-real waifus.
Description
FAQ
Comments (24)
This job looks good.
yo this looks dope, can't wait to try
yoo any chance you can make a 1.3b/1.3b fun version?
I can try, but no guarantees. Does 1.3b have I2V? It might end up looking like shit if it has to be T2V lol. I just don't know much about 1.3b.
@icelouse wan 1.3b i2v uses i2v loras, fun 1.3b uses t2v loras
What prompt do you use to generate the initial image to get the characters in this position?
I just uploaded some of the input images so you can see full prompts:
https://civitai.com/posts/16809770
Those prompts are a hot mess lol, but the core part to get this pose is:
"squatting, (sex:1.2), (vaginal:1.2), (girl on top:1.1), (cowgirl position:1.2), (huge penis:1.2), (leaning forward:1.1), (face focus:1.1), portrait, (male pov:1.1), (face focus), close-up of face, foreshortening,"
For the 'in your face" aspect, the tags "portrait" & "face focus" really help.
"huge penis" in combination with the "sex" tags makes it generate shallow penetration (because it can't show a huge penis when it's inside a girl), which is not usually ideal, but is perfect for use with this Wan lora (input images with shallow penetration work better).
Also may help to do img2img to refine the 'face focus' aspect if your initial gens aren't close enough.
Landscape aspect ratio helps give the super-wide-spread legs.
Also note that the checkpoints listed on those images are only half-correct, because i change checkpoints for upscale. just FYI.
@icelouse WOW, thank you very much for the response!! It worked, but it took me multiple tries, the leaning forward is the key, with out it, the girl will always lean back
This training looks really promising, a few lovely samples really. Thanks for the clean explanation of prompts and even the prompt to generate the pose! Matsumoto Ichika is probably the best at this specific pose, so hot, I tried a few times but didn't know how to get it without ControlNet.
Great work. Thanks a lot.
face always change. somebody can help me???
Post an example.
I've been seeing the same thing where the face and body type will morph to the same face. https://civitai.com/posts/17042886 This example is with it at 0.92, and I included the starter image.
Same here, it's an amazing lora nonetheless but the face changes right after movement happens
Same matter here, i dont understand how others users give so consistents result
hakaan probably because they use animated / 3d style images.
I noticed those work well but when you use a realistic image.. the face always morphs into a 3D-ish character.
Could be that the creator behind this lora primarily trained this on 3D animated samples and not enough realistic ones. It's an amazing lora nonetheless.
I'm thinking about generating 10 - 20 videos using this lora. Then use Wan Vace to extract depth/pose and restyle it into realistic videos. Then use those to further finetune this lora.
Great work, I got some good results with weights 0.6-0.8, but sometimes the penis is broken, or not fully inserted?
Anyway, thank you for your contribution to the community
Maybe the images or videos you trained on all have heads?
And maybe they tend to be anime-like images?
So you trained your face as well during training
Then your face will become like anime face during I2V
can anyone share a workflow for comfyui:D much appreciated.
They broke it? it can only be added on T2V now :(
civit never stops pissing me off. They added separate listing options for I2V and T2V which is great, but broke all the models that were posted before that change.
I changed the model listing to "I2V" so you should now be able to use it.
Also already trained an improved, actual I2V version which should be posted later this week hopefully.
used it on my last couple sets and its giving me some wan flashing of background or subject at times. https://civitai.com/images/85362190
Took me a while to pin it down to your lora because i was getting some very good results otherwise.
Can't wait for the true i2v version!
Wow that's great, 1st time I've seen someone supply a link to the images they used to start with. Thanks a very helpful reference!
please do for T2V
PLEASE DO ONE FOR MISSIONARY AND DOGGY!
Details
Files
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
Mirrors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
Wan_2_1_Hip_Slammin_Assertive_Cowgirl.safetensors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors
stp-v2-1772841.safetensors
strip-v2.safetensors
Wan-Hip_Slammin_Assertive_Cowgirl.safetensors