ltx23 - i2v 1.1
Slower but better audio and hip movement , use at 0.8-1.0
ltx23 - i2v 1.0
Same as previous ones , use at 0.9 or less to have speech
Audio doesn't always sync up
ltx2 - i2v 1.1
Higher rank and resolution , less jittering , with audio.
Use : fast slow from below from behind held in arms missionary pov riding on top
lying down side view
ltx2 - i2v
Use at 1.0 str , or 0.8-0.9 if you want video to follow the text prompt better
Add scene description and script "like this"
Wan2.2 - I2V bulge HIGH
Use at 1.0 str , or 0.6-0.8 if you want video to follow the text prompt better
If it's too slow sample more frames in High , speed up to lower count , sample Low.
Example 81 high > 65 low
1.1 - 720p version , I2V bulge , with no mixing
Best at 0.7-0.8 str , add other loras for better effect
Around 3100 steps
Same prompt below
1.0 - 480p version , I2V bulge , mix test
https://civarchive.com/models/1331682/wan-pov-missionary?modelVersionId=1623136
Mixed at different proportions to add motion that was lacking
Also tried merging at higher proportion so this lora is too strong at 1.0 in most cases
Doesn't always sync , try different prompts/settings
Trained on 15 videos 48 frames each , 512x512 , around 3800 steps
Settings used:
lora str - 0.4 - 0.7frames - 49
steps - 10
cfg - 6.0
shift - 5.0
Promt:st0mach ,she is getting fucked , missionary position, pov , stomach is bulging with each thrust,with her legs spread having sex with a man.A man is thrusting his penis back and forth
Description
FAQ
Comments (21)
Thank you! 😘 I've been waiting for a Wan2.2 version for a long time. Are you still using those same 15 videos or did you expand the dataset?
Ty ,
Partially , some didn't fit the style I think . Was too slow/fast etc , this is a bit experimental and has different style to it . Tried also speeding it up but that made clips shorter but it should be more responsive. (maybe)
tldr 50/50 synthetic to get similar dataset.
@blo01 So how large of a dataset did you use in the end? What type of clips are in the dataset, style wise, 3d renders? Anime? I am asking because I noticed the anime faces of my girls improving when your lora is added to the mix and I am wondering what could have caused it, so that I can include this magic fix in my own lora.
@Remiel 3d/2.5d mostly , 33 clips , it gets less blurry if you get to 0.04-0.05 loss , otherwise idk
@blo01 Thanks for the info. 👍
I don't know if its just me, but in doggystyle images, instead of the kneeling girl getting a stomach bulge, I am getting a pair of balls to appear under the girl out of nowhere, and the guy with a stomach bulge 😂 Maybe the dataset is lacking doggystyle? Or maybe captions are not clear enough in who is supposed to have their belly bulge? 🤔
dataset does lack doggy pose yeah...
so it might add unexpected parts if desired parts are not visible
Now I'll go catch some fairies
I really wanted to like this, but the 2.2 LoRA exaggerates the bulge way too much and also seems to affect the original image (ie changing face, losing details). The 2.1 LoRA is still the GOAT from my testing.
2.1 didn't work that well sometimes , I might have overdone it
@blo01 The only downside with the 2.1 version is that it slows down the movement too much.
Basically, your 2.2 version has excellent motion but overdoes the bulge and changes facial expression in my experience (the dreaded blank anime stare), while your 2.1 version maintains consistency from the original image and has great motion with the downside of overall slower motion.
wan2.2 version is working really well, thank you for this!
I'm getting great results in a FF/LF workflow, both high/low loras set to 0.7 strength, and with a stomach bulge already present in both images
ever thought about a throat bulge lora?
(wan2.2) Great lora,bulge is really good, works most of the time..
but it will make penis and pussy too soft and wobbling sometimes, will mess up penetration.
idk why the wan2.2 lora always B-ulge the PenS not stomach.
it is almost perfect. couple of things: if the p--is or the b-lge is too big, the b-lge ends up just doing a slight woggle so maybe for v2, find some videos for those extra large sizes? (videos from animators like de-pstr-ke and Big green come to mind) also wish we could increase the b-lge size with some sort of setting/prompt
I've finished testing this with several animation (posts coming today and tomorrow) and my thoughts are:
it needs a good prompt
it needs to be so much weighted than the other loras on the mix to work and that makes:
The non-bulge movement result is similar to what you get with just a base model (sex position/movement loras are almost doing nothing).
for a V2 version I would try to make it work with lower weights and isolate the bulge movement to make the rest of the scene to be affected by other sex position loras.
Nice job anyway :)
v2 is good.
Details
Files
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
Mirrors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0machLOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
Stomach_Bulge_v2-2_i2v_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors
WAN-2.2-I2V_st0mach_bulge_LOW.safetensors