This is a tongue kiss lora, based on wan2.1 t2v+clip visual training, and the effects are good on t2v-i2v, but I haven't done any large-scale testing yet. 13 3-second materials were used for training, 23 hours of training time, and 48g of existing storage. I recommend a weight of 0.65-0.8 for this lora. If it exceeds the threshold, there will be too much immersive pictures. As for the lens and light, you need to adjust them yourself. It is recommended to use a fixed character lora so that the character stability will be more consistent.
Description
This is a tongue kiss lora, based on wan2.1 t2v+clip visual training, and the effects are good on t2v-i2v, but I haven't done any large-scale testing yet. 13 3-second materials were used for training, 23 hours of training time, and 48g of existing storage. I recommend a weight of 0.65-0.8 for this lora. If it exceeds the threshold, there will be too much immersive pictures. As for the lens and light, you need to adjust them yourself. It is recommended to use a fixed character lora so that the character stability will be more consistent.
It should be noted that only side angles are currently supported, because this is more clearly displayed.
FAQ
Comments (14)
Wow. Example prompt?
tongue kiss
OH YEAH THE PEOPLE NEEDED THIS
yeah
Works incredible on image to video. Well done.
Any chance for an inp version?
mabe,thx
What is the pixel resolution of this material？
1920x1080
Hey this is great, but we can't use the lora on-site since the mods changed all the Wan video tags. Any chance you could either re-tag it or re-upload it for people to use on-site? It was great for image-to-video!
5B ti2v would be vwry welcome!
Awesome Lora! It works perfect even with wan 2.2 🙂
Could you provide some information about your training process please:
- how many images/videos did you use?
- how long should a video be?
- did you use a combination of images/videos?
- did you use video masking to remove the heads?
- did you use civitai for training?
I started some first tests. Creating a character with images only works fine, but I did not get good results using a combination of images and short videos using civitai trainer.
Sorry - did not read your explanations before 🙄 Ok - I got it, you used 13 clips each 3 seconds long👍 I would be glad if you could answer my remaining questions. And what did you mean with "clip visual training"?
does this work with the wan2.2 14b with the 4 step lora? If it does, I cannot get it to work much at all.
It does work. I'm running it in an WAN22 i2v workflow. The results look fantastic. My gripe is that the scene is rolling back to the first frame mid animation. That is really awkward and I'm guessing it's a WAN 2.1 leftover.
Edit: 96 frames is the upper limit. Everything after that will cause the animation to return to the first frame.
Details
Files
tongue kiss.safetensors
Mirrors
tongue kiss.safetensors
tongue kiss.safetensors
tongue_kiss.safetensors
wan_tongue kiss.safetensors
tongue kiss.safetensors
tonguekiss.safetensors
tongue_20kiss.safetensors
tongue kiss.safetensors
tongue kiss.safetensors
tongue kiss.safetensors
tongue_kiss.safetensors
tongue kiss.safetensors
tongue kiss.safetensors
tongue-kiss.safetensors
tongue_kiss.safetensors
tongue kiss.safetensors
tongue kiss.safetensors
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.