A LoRA that generates in/out motion of a dildo into the vagina of a woman laying on her back with legs spread.
Please consider supporting me on RiotModels to help me in continuing to make new LoRAs. Your reward is exclusive content of the Seven Sisters in the most naughty of animations.
MAJOR NEW UPDATE (14-Nov-25): VERSION 2 NEW WAN 2.2 HI/LO NOISE FOR CLOSE-UP & FULL BODY I2V! HALLELUJA!
New training videos, and splitting training sets into full body and close-up shots have finally yielded the results that we've all been waiting for. Fantastic, smooth motion can now be consistently generated along with a much greater tolerance of body position.
This Wan 2.2 i2v LoRA consists of FOUR PARTS:
High Noise for close-up framing
Low Noise for close-up framing
High Noise for full body framing
Low Noise for full body framing
FOR BEST RESULTS USE BOTH CLOSE-UP AND FULL BODY LORAS IN YOUR GENERATIONS, EACH AT STRENGTH 1.0 (4 x LoRAs TOTAL)
Notes for use:
Keyword: piston_dildo_style
Use my new Dildo Posing Pony LoRA to generate your base images.
Training data has a bias to right hand moving the dildo. So if your generated image has the girl using her left hand, horizontally flip the image. This will guarantee better motion.
This full prompt produces fantastic motion with a wide variety of poses: "piston_dildo_style, solo woman, relaxing on bed, dildo inserted into vagina, slow speed, deep thrusts, she pushes the object into and out of her vagina repeatedly with rhythmic strokes, the dildo disappears inside her vagina and then comes out again with the same shape, sliding in and out, in and out, face at top of frame, anus at bottom of frame, dildo is of the same color throughout its length, her vagina closes tight when the dildo is removed completely,looking at viewer throughout,expression shows pleasure and sexual excitement"
Trained locally on my RTX5090 using Wan2.2i2vA14B high and low noise models with separate training sets for full body and close-up camera framing. Musubi Tuner with 1e-04 LR (high noise) and 5e-05 (low noise) 720x720 resolution, network 32, 20 frames at a time with 10 frame slide over the 90 frame source video.
Can be used well with Wan Lightning LoRAs, with CFG=1, Steps=4.
You should get good fluid in out motion 90%+ of the time in a wide variety of positions using all four LoRAs.
No plans for updated T2V version at this time, sorry!
My Wan motion LoRA full training process is covered here in my new tutorial.
I have also written a series of articles here on Civitai on the subject of Video Motion Training.
Description
Much better motion adherance with this MoE High/Lo Noise combo.
FAQ
Comments (20)
I'm having a hard time stacking up images for this lora. what t2i lora specifically do you recommend for sdxl or pony? it could be my prompting but i can't get any good left side angles and good dildo inserts.
It can be a challenge, believe me.
I am using this one: https://civarchive.com/models/740498?modelVersionId=828097
But it appears to have been removed unfortunately.
If you're getting left handed images just horizontal flip the image in your image editor (I use XnView for this purpose, but most editors have this ability).
@Generation3dX i found a good dildo lora, however the angle is still not correct, but i will try to use xnview or any editor for flipping angles/images. thanks for the tip and your loras are awesome!
@Soogo101 which one did you find? I'm interested in finding better ones for t2i.
@Generation3dX https://civitai.com/models/218726?modelVersionId=450905
All issues should now be resolved with the new version (2.0) and my new Pony t2i LoRA (see main notes)
@Generation3dX ahh nice! i will download and check it out soon!
T2V for WAN2.2 would be nice.
I will book a slot to attempt this at some point, but I'm not optimistic as I've found T2V even harder to train than I2V and dildos are a real nightmare for AIs to understand.
i2v is difficult to implement
The subject itself is very difficult. It's not like sex positions which are relatively easy. The AI models really struggle with small objects that have no understanding of disappearing into orifices. The model was published with these strong caveats included. But it was better to publish than do nothing and see what the community could do with it.
@Generation3dX Yes, it is indeed quite difficult! Thank you for your sharing!
its easy jusy add a translation promt - the scenes starts with the camera zooming out revealing a cinematic scene with
All issues should now be resolved with the new version. The difference between v2.0 and v1.0 is like night and day. Don't forget to use my new Pony t2i LoRA also for setting up your base images. See notes in main description.
Start with an image where the dildo is NOT inserted. This way the AI has a better idea of what the portion of the object inserted looks like.
@javbukkake it does indeed seem to work even with the dildo initially not visible. So the AI really learned the concept of a dildo and its insertion. Winning!
Hi. Can you tell me how to generate a good video in i2v? What models should I use? or what workflow should I use? I always get blurry frames or accelerated videos
This is a more touchy lora in terms of getting it to work due to the known problems of wan recognition of objects being inserted instead of body parts. i recommend you start with my fingering lora first by downloading one of my showcase videos then dragging it onto your comfyui workspace - it'll then bring up the workflow and you can then install any missing nodes and models.
I'm a newbie. Could you please tell me where I can download and install it in comfyui? Please provide screenshots if you don't mind.
@ofitserovalexey728 He told you. Drag one of his showcase video from the lora page into your comfy, it will give you his WF. Then you install the nodes you're missing from this WF