I2V
I2V has finally landed, performing much better than using the T2V lora with the I2V model. Trained on the exact same dataset, excluding images as they cannot be used to train i2v. The low noise model has been trained at the native resolution of the dataset for maximum fidelity (1024x1024).
The concepts and prompts are the same as the T2V version.
T2V
Trained on a quite large collection and a lot of different concepts, the main one being women lying on their back showing their asshole. Regularization data were also mixed in to prevent bias.
The model seems to perform well in a wide variety of scenes and has the potential to perform well in much more tasks than the one it was intended for.
Training/Dataset parameters
259 * 512x512 videos of 3 seconds.
189 * 1024x1024 images.
Rank 128.
18 epochs on the high noise.
22 epochs on the low noise.
Higher repeats on concepts with less data.
Learning rates started at 0.0001 for both, and the last ~20% of training ran at about 0.00002.
Videos were sliced and cropped using a custom tool I wrote.
Captioning:
Qwen2.5-VL was used for the initial captioning.
The system prompt was adapted for each different concept to provide the model with a context and instructions on what to look for.
Manually reviewed with minor corrections.
Trained using diffusion-pipe on a single 5090.
What it what trained on:
Women in piledriver position
Spreading her buttocks with own hands
Viewer spreading her buttocks
Close-up of vaginas from under
Penis anal insertion
Penis anal pull-out
Penis anal thrusting
Side views
POV views
A variety of pubic hairiness, but I am somehow still unable to control it
Ejaculation, but hard to control
Faces were mostly cut out of frame to minimize character bleeding
What it cannot do:
Do NOT expect good results on standing poses. There is none in the dataset.
See the prompts associated with the samples.
Prompt geniuses please share your work.
Description
FAQ
Comments (17)
259 clip isn't that complete overkill though? (not trying to be an asshole here just wondering...) still impressive result, even if I feel you could've done this at rank 32 as well with so many clip and bonus images
Good lora, but why I always get anus looks like a pink hole?
great job! any chance an update that includes the back view?
I love that this lora exists, but does it only contain loose gaping buttholes? Seems like it wouldn't since you said you trained it on a lot of different concepts, but the results seem to only have the one version/state.
Excellent work on this high-quality lora!
And thank you for your explanations about the training.
Could you specify the parameters used in your TOML? I have trained character loras using only images with great results, but I would like to try training HIGH with videos.
BTW, manually tagging videos is very tedious, lol. I'm used to doing it automatically with images and minimal review/correction. Any solution for that, maybe Qwen Abliterated? I've tried it with ComfyUI, but the two nodes I've tried give me an error. I don't know if there's something like TagGUI but for videos.
Finally, those of us who create Loras know all the work that goes into it, and you must have had a LOT of work with this Lora, so I just want to thank you for sharing it ;-)
Great job!
Strangely I get a blurred image every time, anyone can help?
can this do men too?
"Trained on a quite large collection and a lot of different concepts, the main one being women lying on their back showing their asshole."
I don't train LoRAs so I can't offer advice on fixing this, but I'll say from my testing it really insists on that "main one" and ends up being a rigid, one-trick pony. It's a good trick, but over two gigabytes for one trick kinda hurts.
It really insists on invoking that pose to some extent or another (meaning body glitches ensue). I tried weights anywhere from 0.55-0.8, alone, other pose/motion, other genitalia concept enhancers, etcetera. And even in instances where it didn't partway invoke the pose and glitch the figure, it would always insist on "anus on bottom, vagina on top" in 100% of my trials. So, girl on all fours, for instance. Hmm, that's not where that goes...
Anyway, the effort is appreciated otherwise. Things to think about if/when pursuing V2, which I hope you're inclined to do. Not enough asshole lovers trolling the latent space.
The Sheer creativity in the gallery is insane
I've been using this for a while now and have only just realised I haven't said thanks.... so thanks for this amazing piece of work, I don't think I will ever get bored of playing around with it.
Strangely, many times some objects can turn into penises. What is the reason for this
I'm quite curious about why the anus is a bit higher than I expected. For instance, when I was outputting some videos of squatting and shooting the anus from above, the anus should have appeared in a lower position, but it ended up in the front, almost adjacent to the genital area. Whatever, is this the best anus lora of wan2.2
Are you ready, guys? :)
I have so many regrets.
Could you please tell me what setup you train on? How much RAM do you have? And what program do you use?
I get results of much lower quality than in your examples, although with other LoRa everything is. The vagina and anus are not as detailed as in your examples. Maybe because of the lightx2 LoRa? The Strength of yours LoRa 1
Thanks for the great work. Which version of the model are you using? Vanilla or some NSFW remix? I just tried the I2V lora with K6 GGUF and the anus & genitals are almost non existent, nothing close to the details you are getting in the examples.
Details
Files
WAN2.2 - Asshole - LOW - e22.safetensors
Mirrors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
Asshole_LOW.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2-asshole-low-e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN 2.2 - Asshole - Wan 2.2 T2VI2V 14B LOW.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
test20-low.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2_-_Asshole_-_LOW_-_e22.safetensors
WAN2.2%20-%20Asshole%20-%20LOW%20-%20e22.safetensors
Asshole_LOW.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors
WAN2.2 - Asshole - LOW - e22.safetensors