Wan - Foot Focus / Feet on

Most of the preview videos are I2V and use input images posted to my Pressed lora. See the gallery there for credit to original artists!

The previews with the photorealistic mature kimono woman are T2V.

Trained on 14B T2V model.

Works for T2V, but it's really meant for I2V.

I specifically made this to accompany my Pressed image lora. In fact it was trained on a lot of the same data.

But you can input any POV image. Best results when the feet are already in view.

Typical prompt base:

a woman is lying on her back with her knees to her  and her knees bent. she is  with a man. the man is completely out of frame except his torso and . the woman is being vaginally . the man's  is sliding in and out of her . the woman's feet are the focus of the image. her feet are on the man's . her feet are pressing against the man's torso as he moves forward and backward. she is in the pressed .

ℹ️OTHER NOTES

• If the input image doesn't already have the feet contacting the torso, I recommend doing >5 seconds animation to give her time to move them into place and still have time to do footsie .

• "ankle grab" images tend to work really well, and even if the input image doesn't have ankle grab, there's a good chance he will grab her ankles/calves in the animation.

• Happy accident: Input images where her legs are straight-up with soles visible can give interesting results. For such images, she may not put her feet on the torso, but rather smother the viewer/camera with her soles. No guarantees.

⚠️KNOWN ISSUES

• Extra toes! Wan is surprisingly good with feet, but it must really like toes because it often does 6+. Input images where 5 toes are clearly visible seem to work best (don't give it a chance to infer extra toes)

• I2V works with all styles cartoon/anime/semireal/photoreal. But seems best for photoreal or 3d styles. It may deform anime faces. ymmv.

• It only does POV or "slightly from side" ... for now. You can try other angles, and they sometimes work, but the training data is not there in this version.

• It's not really "pressed" ... it's not very aggressive, more 'cute' than '.' But I am trying to improve this and make it a little more with feet pressing harder on the dude.

• It does okay with the naughty bits, typical Wan . If your input image contains a huge , it may appear to be violently sucked into the and become an average pecker😨 Also, is hit-or-miss, often appearing to tear right through the and connect to the 😱

• T2V may not work so hot. The girl's head is often out of frame, and it can really scrunch up her torso. Most truly pressed (training data) has the girl in this scrunched-torso position and the lora seems to exaggerate it with T2V.