Bite that lip!
Wan 2.2 I2V v1
Definitely happier with this run than the last one. Still recommend highest resolution you can manage. Also, I had really terrible results using SmoothMix checkpoint. Not sure why. Best results I have seen are using the base wan. Test videos use Q6_K.gguf models. Also, I recommend messing with the lightning lora weight. I got better stability using 1-1.5 weight on lightning lora, or 2-2.5 if using other loras for sex acts for more speed/movement.
Wan 2.2 I2V v0.1
I'm seeing better results at higher resolution. So 720x1280 seems to be the best. Lower resolutions have sometimes worked, but only if the face is large and the primary focus of the image. My guess is that in the compressed latent space, the lips are just too small at lower resolution for the model to focus on, much less for the Lora to affect. So if you need small resolution, make sure the lips are big in the image reference.
I have a version of the core workflow I use here:
https://civarchive.com/models/2076123/wils-style-workflow
Description
EXPERIMENTAL
This is my first attempt at WAN LORA training. I think it came out good enough to post, but it's... also sorta busted. Get's super flickery/color shifted flying solo. Seems to do okay stacked with other LORAs.
FAQ
Comments (11)
cant download the i2v wan 2.2
;; I tried re-uploading as individual safetensor files instead of the zip, let's see if that works. Otherwise it just shows them as having requested a scan, so maybe they aren't verified yet? Never had that problem before.
This time it worked, Edit: both versions Hi & Low (L & H) are normally downloaded by now, I could download through direct download & from within LoRA Manager
@darkwil never ever upload zips for model files, literally a worst practice
@darkwil no worry was able to download now legend
@ultimo_intento Yeah, I just wish it was set up better to deal with the high and low. Guess I just need to get over it.
@darkwil Yeah it is OK, btw you have done nice job on those expressions, makes my ladies more hot & seductive ;)
wan 2.2 version is like a dream come true... been waiting for something like this that looks actually good! i do have a slight problem with the quality when i use it. when i run it at 480x720, the quality is blurry and the lip biting looks weird... when i do 608x1024 or 720x1280, i have better luck with that. but even at higher resolutions, it seems to look slightly weird for me still. how did you get the results in the examples? they look almost perfect! (and yes i am stacking the twerking lora from: f1985) i am also using the smooth mix workflow but with a different checkpoint (not smooth mix checkpoint)
EDIT: i ended up figuring out what to do. regardless, best lip biting lora out there!!!
@TypicalLoraLover Great! Yeah it seems to be a bit dependent on the size of the input image. Like effectively how big the mouth is pixel-wise. I got terrible results on anything less than 480x720. 720x1280 seems to work best unless the face is more prominent/large.
@darkwil yep, what I did though to figure it out was generate an image that is 704x1088 or 720x1280 and set the resolution of the video to the same resolution of the input image!