Adorable girls - Vagina close-up from below

Adorable girls - Vagina close-up from below - Low

NSFW

Adorable girls: Vagina close-up from below 🫦

I first tried to make a face sitting Lora but now i like to use it to make a low-angle shot, then screenshot and re-use as an input image with another Lora

## Usage

I highly recommend you to use these lora with it: The reference videos of the model are generated using them

Camera tilt down undershot At @0.7 strength

Ultimate Pussy and Anus helper At @0.5 strength (In the archive)

Trigger word: facesitng

Recommended prompt:

facesitng, the video start with a woman, the screen instantly goes dark and then lights up again to show the same woman sitting on the viewer in the same place, camera angle from below. She is bottom less, her vagina is fully visible, she is above the camera, vagina close-up. Her face is visible.

Negative prompt: I personnaly use this one :

色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走, extra head, two heads, duplicate head, floating head, extra face, deformed body, mutated, extra limbs

## Training details

- Model: WAN 2.2 14B I2V

- Steps: 600

- Resolution: 480x832

- FPS: 16

- Duration: 5s

- Videos: 20

## Recommended settings

- Sampler: Euler

- Scheduler: Simple

- Steps: 3 high, 5 low

- Strength: 1

## Notes

- I used Wan2.2_Remix_NSFW_i2v_14b_high_lighting_v2.0.safetensors (And low) who is actually my favorite checkpoint to generate nsfw videos

- Found the images used in the showcase on CivitAI

Hf ;)

***** This LoRA is shared without warranties and with the condition that it is used in a lawful and responsible way. I do not support or take responsibility for illegal, harmful, or harassing uses. By downloading or using it, you accept that you are solely responsible for how it is used. *****

Description

FAQ

Comments (15)

tsuna_mikasaMar 28, 2026

CivitAI

In my environment, it didn't work like the reference video. ① A woman is shown, ② the screen goes black, ③ the screen brightens, ④ the camera is positioned at her vagina. According to the description, the camera move from its initial position to the position of her vagina?

1User

Author

Mar 28, 2026

Idk if u use the recommanded Lora combo but i added the workflow used to generate the reference videos as training data

tsuna_mikasaMar 28, 2026· 1 reaction

@1User Thank you!

Your workflow worked for me.

Lora_AddictMar 28, 2026

@1User maybe i'm stupid by i downloaded all examples and it always only contains the load video node for me

1User

Author

Mar 28, 2026· 2 reactions

@Lora_Addict In the download section, just below "LoRA (292.6 MB)" there is a "Training Data" containing the .json wf

Lora_AddictMar 28, 2026· 1 reaction

@1User ohhhh... thank you <3

AishaAIMar 29, 2026

CivitAI

I think you've been looking at my LoRAs, haven't you? ✋🏻🤠🤚🏻

Well, I'd like to make an observation about your dataset.

You stated that you trained with 600 steps, 20 videos of 5 seconds at 16fps (totaling 80 frames per video). The problem is that, due to the number of steps, the training couldn't even read all the videos, since 20 * 80 frames = 1600 steps to read the entire dataset once (1 epoch).

With 600 steps, only 7.5 videos from your dataset were read and trained.

To have good learning in High Noise, you need about 3 epochs (3 * the total number of frames in your dataset), and for good learning in Low Noise you need about 5 epochs.

By using the "Camera tilt down undershot" and the "Ultimate Pussy and Anus helper" your LoRA becomes completely unnecessary (I tested it, and it worked perfectly without your LoRA).

1User

Author

Mar 29, 2026

I'm using Ostris on a Runpod A100 SXM, each step take around 50-60 seconds, when the hardware cost is 1,5$ it rises very quickly

Do you have any recommendations ?

I saw your video, it's slow and not showing the vagina in the right angle, i think the Lora has his use case utility

AishaAIMar 29, 2026· 2 reactions

@1User I normally use an L40S or an A100-40GB. I usually get between 40-30 seconds per step on the RunPod, but I get between 30-20 seconds per step on the Modal. I don't know why, but the RunPod seems slower.

Now, about what I said about the video I made, if I had written in the prompt that she opens her legs, it would have exactly the same effect as face sitting.

Wan2.2 isn't a cake recipe that needs to be precisely followed to get results... You just need to know how to use the basic knowledge that Wan2.2 has and everything works normally.

For example, Wan2.2 doesn't know what "face sitting" is, but it obviously knows what "spread legs" is, just as it doesn't know what "cum dripping on her face" is but it knows what "viscous white liquid dripping on her face" is.

I see many people doing almost useless training just to redefine a word instead of taking advantage of basic knowledge, not to mention the unnecessary training done in High Noise/Low Noise.

I tried to reproduce the results of your videos using only your LoRA + Pussy Helper and it simply doesn't work, because the number of steps wasn't enough for it to learn anything beyond darkening the image.

When you want to know if your LoRA is working decently, look at the LOSS graph. When the variations between 10 steps are between 0.6 and 0.9 in RAW, it means it has learned to repeat the patterns it saw in the training videos.

Remember that whether you have 1 step or 1 million steps, your LoRA file will be the same size in megabytes, and that doesn't mean it's working correctly. So, to teach a few things, use RANK 16 and CONV 8; this will keep the file around 150MB. But if you need to teach very fine details and many micro-movements in Low Noise, use RANK 32 and CONV 16.

1User

Author

Mar 29, 2026

@AishaAI Thanks you for your advices dude

alex47213Mar 31, 2026

@AishaAI at least his face consistency is much bigger than yours (and stop talking about "forget speed loras"). :D

AishaAIMar 31, 2026

@alex47213 now you're blocked 🤫

StrangeLiminaApr 10, 2026

@AishaAI Is there somewhere that people have gathered a sort of Wan2.2 wiki? Like a database of the various concepts the model is familiar with by default? I know of many, but I'm sure it's only a tiny fraction of the total.

StrangeLiminaApr 10, 2026

Can you link to this "Ultimate Pussy and Anus helper" I can't seem to find it.

I found it on another site, that listing had a link back to CivitAI which was a dead link. The model was removed??

yefy_aiApr 2, 2026

CivitAI

Btw, why not link to Wan 2.2 Pussy and Anus LoRA by HearmemanAI here on Civit unless the results are way different? I once used the Ultimate Pussy and Anus helper (which got deleted), but HearmemanAI's model gave me better results (I however can't remember in which way 🤔).

LORA

Wan Video 2.2 I2V-A14B

by 1User

Download (Beta) View on CivitAI