This is the first LoRA I ever trained other than some AI influencer LoRA for Flux a while ago so go easy on me please :)
I tested mostly T2V using the 14B model and it works pretty well.
Tested very briefly with the I2V 480P model and also got decent results.
Trigger words: pussy, vulva, vagina
Some example prompts I used for T2V:
Candid video, A petite 19 year old blonde female tennis player wearing a white cap, white sports bra with a tennis racket logo on it and a white mini skirt, she is sitting on a wooden bench next to a tennis court, there's a metal fence behind her, she is shown from low angle, through her lifted mini skirt her pussy is clearly shown, her labia majora is spread and her vagina is completely shavedAn amateur video showing a gorgeous Argentinian 19 year old woman with light brown hair, deep blue eyes and round glasses sitting on a fuzzy rug she is wearing a light blue and white football jersey, her pussy is the center of the video, her labia is spread and her vagina is shaven, her skin tone has a smooth even texture, her inner labia minora is visible, she is thick and dense pubic hair, the woman is smiling playfully showing her perfect white teethTraining info:
Trained on an H100 using my diffusion pipe RunPod template
Dataset consisting of 58 images generated with Big Lust / Big Love
Dataset resolutions: 896x1152, 1024x1024, 1344x768, 768x1344
Training parameters:
Epochs: 70
Batch size: 4
Learning rate: 3e-5
Rank: 12
Batch size: 4
Captioning:
All images were captioned using joy caption with the word pussy as a prefix
Join my Discord server for updates on new LoRAs, tips & tricks, Workflows, RunPod templates and the holy right to be close to an amazing charismatic person like myself:
https://discord.gg/fyha5Pzm
Description
FAQ
Comments (37)
Wow, great detail on this one!
Thank you and thanks for the buzz!
I love that it respects all kinds of vaginas lol. Great job!
Thank you for the kind words!
I tried to keep the dataset diverse
Finally, a lora that teaches wan about this part of the female anatomy.
Thank you!
Glad you liked it
I hope there's gonna be the V2 with wider variety of shapes. Innies for example.
Have you tried it? It worked for me by prompting innie.
There's 2-3 innies in the dataset.
I might expand it in the future
@Hearmeman +1 for innies ;) At least good to read there were a few in the dataset as all samples show very similar shaped pussy. Cheers. Thanks for the training!
@dryuha15117 @SD_AI_2025
I added 30 more pictures to my dataset most of them butterfly and innies and captioned them appropriately.
Currently training, let's see... I'll post it if it's any good.
@Hearmeman How did it go?
Bro, you should’ve released this LoRA earlier, so I wouldn’t have had to spend $3 training my own version last night lol.
release it too!
@sunburnt_toasty I wanted to release it too, but it is honestly far from satisfactory lol.
I've found using 2 loras that do the same(ish) thing at low weight (e.g. 0.5 each) can be better than either alone. I guess together they've seen more examples? Worth a try.
@logenninefingers888 which ones exactly?
@sunburnt_toasty my doggy POV and the other (better!) lora for example
Lora was trained too close
Maybe, Most of the pictures were relatively close so it understands more detail.
Tried and tested. The impression is more negative: the “rose” is too red and “rubbery”. It does not look very natural.
That's probably because it was trained on AI generated images. But it's definitely way better than nothing.
@kangaru861 Of course - when there is only “nothing”, the situation is sad. But in any case we will have to wait for better samples.
This works surprisingly well, though it would be good if at least some of the training images were real photos, as it looks kinda weird / rubbery sometimes.
The biggest room for improvement, other than controllable style of vagina, would be different angles - it can't seem to do front-on view/standing, only spread legs.
I love your work and your lora
I'd like to use them but currently aren't available on civitai with wan 2.1,a monk to have a powerful PC locally
Great, you're awesome.
Seems to change the face for some reason
Works but vagina is still bad, need Innie version.
Hello, brother, it seems that you forgot to upload the cover preview!
he didnt add the prompt.. i mean wtf xD
Thank you bro
im fairly new to all this. i use the wan 2.2 14B default template 4step lightning lora. how do i set it up ? what sampler and scheduler do i use ? default give me blur and ghosting even with neg prompts set up. thanks for your work !
You got better yet? I'm in since 1 week and improved alot recently, maybe i can give some advice.
@teeay25225 any help appreciated. yea i have improved a bit, but with my ADHD and comfyui - whoa ! lol. i just get confused which models to put where eg a dasiwa HI and a remixnsfw in low can give good results, then having lightx2v loras on/off is apain to figure out as some models have "baked in" so if i have them on sometimes faces are melting, and vice versa it's a minefield !
@488evo625 Yeah and because its so complex its so hard to help someone. I only can tell that the best thing for a beginner is to use the boundbite10lightspeed checkpoint with a baked in lora and to not mess around with your own speed loras. Focus on how the normal loras work because they can change the face/identity with the wrong strenght. Then all the different samplers.. Here too: Dont mess around too much as a beginner but what i know is that Euler and Unipc are good for identity and overall. Dasiwa Complex 2 is the workflow i can advice, very beginner friendly and i use it almost every time. But there are also workflows where you able to make 10-20 seconds long videos thats crazy and damn nice and works very well. Mostly there are speed loras in these workflows, deactivate them and use the bondbite10lightspeed checkpoint with the baked in 4step speed lora. But anyways: With time comes knowledge and you will get better. Greetings from germany
@teeay25225 I already use dasiwa, midnightflirt, smoothmix, and remixnsfw, i think any more would blow my mind ! Thanks for your words though and Hello from Scotland !
Many Loras have been posted later, but this one is the only that keep consistent quality, not messing up other body parts and especially faces, and not degrading overall quality.
The day you trained this Lora was really lucky one for everyone in this room xD
Details
Files
Wan_Pussy_LoRA_Hearmeman.safetensors
Mirrors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
pussy_LoRA.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
wan_p.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman (2).safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Wan_Pussy_LoRA_Hearmeman.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.