This is a motion lora for animate diff
Added the training clip i used to the previews
2 lora sets low strength and high can mix and match if needed. Testing shows every checkpoint handled this different.
Helper lora at 0.6 is good
I recommend using the helper lora with this. low works well with high. Just the temporal lora will give 90%
320x320 8 frame image set Low - 250 steps High 350 steps
New With the new IP adaptor clipvision enhancer you can use stronger IP refs here.
Set enhance ratio to 0 and then you can up the strength of the ip adapter.
Can be looped with looped uniform or not
Non looped has better movement, looped actually loops well
used context stride of 2, overlap of 2 16 frame context length for 32 frames
Testing shows you can use context length of 8-16, seems it determines the number of for lack of a better word, pumps. 8 pumps in 8 frames, 16 is 16 frames.
If you want to use the IP adaptor at a strong strength, you can add a mask to the ip adaptor to only effect where you want, Since you know the animation being produced you can easily create a mask to tell IP adaptor where you want it applied to.
New With the new IP adaptor clipvision enhancer you can use stronger IP refs here.
Set enhance ratio to 0 and then you can up the strength of the ip adapter.
32 frames tested 15fps can low this to slow movement. Can try more frames.
Had more testing, Tested up to 128 frames and it works great.
Examples used
hijklmix v5 for base
photon for upscale
example prompt - i did not test much else.
1girl, 1boy, a (your keywords) girl is on top of a man riding cock, from behind, (keywords), vaginal sex, penis, balls
Other examples
Can be combined NTR cheating loras like 0.6 strength
https://civarchive.com/models/50060/caught-cheating-ntr-concept
Prompt
(1girl, 2boys:1.4), from behind, (3/4 view:1.2), a man is lying on a bed with a girl on top riding cock, bouncing butt, penis, vaginal sex, looking back at viewer, man under girl, cuckold, voyeurism, (angry boy in background watching:1.2), (caught cheating)
Description
FAQ
Comments (2)
woah i wish i know how to train simple but excellent scenes like this... do you have any tips for me?
How would I use this for Invoke AI?