This is my realistic tentacle model for I2V. It performs well in both cases where you already have some tentacles in the screen or not. Compared with Anime/2D training data, realistic tentacle movements have very few resources online and kinda just suck. But anyway this is still better than I hope it would turn out, and I'm sure you will have a lot of fun playing with it.
My example are super lazy because I enjoy lora training from coming up and cleaning dataset to setting up the training parameters more. To use this model, here are some sample prompts (simply copy pasta from my examples):
One person named 8f%gqg in the middle of the screen, a black tentacle enters from the top of the screen and fills her mouth. White liquid cum shoots out from the tentacle. Stable camera.
One person named 8f%gqg in the middle of the screen relaxing. Suddenly rusty flesh tentacles with suction cups appear from all sides of the screen and she is shocked. A tentacle from bottom thrust back and forth in her vagina violently, and another right below it enters her anus, and snaking throbbing. The tentacles shoot white cum liquid.The reason I use person named 8f%gqg is because this seems to preserve the original person's face better(maybe not, didn't test thoroughly). If you use Asian girl, White girl you sometimes end up changing their faces midflight.
Some other points:
You can use my sexmachine lora to enhance the thrust movements, or breast squeeze for enhanced cumshot.
This model is pretty "dominant" in a sense that sometimes your WAN will ignore other lora's functionalities.
Lastly, I'm sorry I'm so far unable to produce any T2V models, I spent my whole yesterday on this but the result is simply abysmal. I mean it works but not up to my standard. I want every lora I publish to be good so not going to release them.
Description
Initial version
FAQ
Comments (28)
Thanks for trying T2V
author, may I ask why you need all this crap that makes you want to puke?
This lora isn't for people like you.
@gaoxuxu exactly
try this with a bong...
My wife likes it
@gaoxuxu Yeah, definitely not for me - I'm not a mental patient yet.
@ilya808rolf994 Yeah sometimes she does that
@ilya808rolf994 you are new to ai art. give yourself some time and check out this comment next year, in 2026
@ilya808rolf994 Glad you took time out of your busy ‘not liking this’ schedule to tell us all about it.
A true pioneer of wasted energy.
laaaame bla bla bla! SHURA I LOVE YOUR LORAS
epic stuff!! great stretching of kink! need a bong now! lol
Disgusting, need more. Here is some buzz.
I can't get this to work with Wan2GP, I get the following error-
Exception: No compatible weight was found in Lora file 'loras_i2v/tentacles-shura.safetensors'. Please check that it is compatible with the Diffusers format.
Any idea what I'm doing wrong?
I don't know what Wan2GP is, can you share your workflow?
@shura490 https://github.com/deepbeepmeep/Wan2GP
@ictos Are you using command line, any particular reason not using ComfyUI? Can you use Kijai's ComfyUI workflow or vanilla workflow by official WAN2.1 huggingface account instead
What's the purpose of the "a person named" portion of the prompts?
Was told this may improve face stability so trying it out, I think I can't really tell the difference when model fidelity is high. But for low quality lora this seems to have effect, or maybe it's just placebo
to keep it from mangling your female subject and turning her into something\someone else
@floweranon888 Does it work better as that specific string of characters, or can it be any unique "name" you come up with?
please make this for hunyuan so i can use it with framepack
Hi, could you show me the dataset for this model, I would like to train a similar model but with tentacles from the eyeball.
Sry I can't don't want to get into trouble but I can tell you I find all the clips on Rule34 and PH
This won't work with Wan Vace 14B or am I doing something wrong?
What's it trained on? 5B or 14B?
please re-train this for wan2.2