1.0 - i2v with audio.
Attempt to add audio , might be over trained set to 0.8
0.9 - i2v no audio.
i2v trained on mix of ai mostly for bouncy look , ty to Allardice
Set 0.5-0.8 strength with keywords or with descriptive prompt for smoother results.
1.0 will also work but has repetitive AI look to it , could be better with other loras.
Start with Riding backshot animation. / Riding frontshot animation. + text.
Description
FAQ
Comments (16)
I'm finding this Lora at very low strength (0.1-0.2) can help bring a little jiggle physics into any video, not just sex ones.
nice find , it's just adding vertical motion ,need to try it
you can also rotate image and rotate video back to get funny results
@blo01 Are you prompting any specific for the penis going all the way in? Im struggling there
@redlittlerabbit Not really , if you use other loras they all average so stuff won't go all the way in.
If it's outside use other lora with riding one.
Also it's seed chance and input image angle bound from what I saw.
May want to correct the base model to 2.3
You're right ,thank you
At the moment of release it did have a tag but didn't have a filter added
So I put it in 2.0
thanks for the no audio variant !
It is pretty bad at keeping the face. looks like another person as it goes on. shame.
yeah this wasn't trained for long runs :D
after how many frames does it break for you ?
@blo01 It breaks pretty much instantly on anime styles for me, especially the lips mutate massively. For example this is what it looks like as soon as her mouth moves: https://i.imgur.com/iUPsMVr.png
Got the same issue on the SmoothMixAnimations LoRA, probably affecting style too much. Lowering LoRA strengths helps slightly but it also affects motion quality. I don't know enough about LoRA training etc. but I'd guess that there's a few blocks affecting style too much.
@LastAssignment are you using model with mixed in loras ? or clean one without adding anything to it except this lora ?
@blo01 Clean BF16 dev model. I only really get it working by lowering this LoRA to atleast 0.5 (used 0.8 in the example), motion suffers slightly and style still bleeds through but it's atleast usable this way, motion is MUCH better than without.
It's definitely not easy to get flat anime styles working with mouth movement with the current NSFW LoRAs.
Details
Files
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
Mirrors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
ride.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
1.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors
nsfw_riding_backshot_frontshot_ltx23_v1.0.safetensors