Take it easy š±
Okay, I'll go harder š¤¤
šIt is recommended that you preferably generate in 16:9
Wanna support me?
Don't send Buzz, send me USDc or USDT through Polygon Network:
0x3f1e1323fa90b4c9f5a386bfa5b1e72f5a6c25e9$0.10 or $10, any amount is welcome and helps me make more LoRAs for the community
Workflow
I don't use ComfyUI, so I don't have a workflow to provide š¤·š»āāļø
The training is done on Wan 2.2 A14B Diffusers (BF16)
Training
400 steps on high noise
1 edited video (80 frames)
HLR
Lots of hope
$10 (š)
Usage (Low noise)
This LoRA was trained only on High Noise, which means you'll have to use some Low Noise that knows what a penis is (I used the Low Noise from DR34ML4Y v2).
Got an idea for LoRA? Then let's do it! (or at least try š¤ )
In my repository, I created a model called "LoRA Requests", so if you want to request a LoRA, send videos and photos to help build the dataset for a public LoRA that will be posted here, that's the place to go.
Open LoRA Requests on HuggingFace š¤
V2 Update:
The first version wasn't working very well, so I decided to redo it, and to my surprise it wasn't much different š„
I understand that the problem is in the training video, which ended up generating a bad blur in High Noise, creating trends and such. To solve it, you can increase the weight of your woman's description prompt.
If you describe the woman and still miss some characteristics, put them in parentheses and add a weight between 1.2 and 1.4, as in the example below:
(the woman is asian, has white short hair, black eyes and slender body:1.2)This will force this characteristic, and in my tests here it was the solution, because Wan 2.2 needs to see a similarity between the "two" people to pick up the reference, such as the same face, hair and clothes.
I2V:
I know the type of prompt you want, but I can't put it here. To see all the prompts, visit this LoRA on HuggingFace š¤
Prompt:
The man's penis is going in and out of her vagina. Her body trembles as he moves his hips.Note that the trigger word wasn't there
High Noise LoRA Scale: 1.2
Low Noise LoRA Scale: 1.0
Shift: 4
T2V:
I haven't tested it on T2V, and it's quite likely it won't work. If you want to test it, just keep the same structure as I2V and add the details.
About HLR + ZCD
This is a fast-learning technique, which makes LoRA less flexible. This can drastically reduce creativity, but it yields stable results while using few resources. Negative effects:
It will probably only work at the same camera angle
High chance of not responding to different prompts
High chance of forcing the original characters of the training video Training only the High Noise reduces this chance, but it can force things that are VERY similar, like the same hair length, same color, etc.
Download
Download High Noise LoRA (On HuggingFace)
Download Low Noise LoRA (DR34ML4Y_I2V_14B_LOW_V2)
Help me creating more
If you want to help me continue making LoRAs, or if you want me to make a LoRA for you, buy 5000 PlayCoins at Aisha-AI and transfer them to my account (account number 2).
This helps Aisha-AI to stay alive and produce new LoRAs for you all š
Description
First version
FAQ
Comments (31)
I am there to learn to fly. :-D
I believe I can fly
Iām not sure why all of your loras have a high failure rate and need to seed fishing to get decent results on my setup, while other authorsā loras work fine without seed fishing.
comment is useless without sharing your prompt and settings.
I only post the LoRA when I get good results in at least 10 different generations, with completely random seeds. When it's not working properly, I put a warning in the post. I don't know if it was a coincidence that mine worked or yours didn't, but it would be really more helpful if you told me where it's generating and what settings you're using.
@HarryPsalmsĀ it's actually not when the suggestions given by the author are so out of whack. And that could also be why their examples have that very "cooked" low quality look. Even one of their posted examples isn't actually doing the prompt and the male is basically just doing a spasm behind the girl.
edit: that said, several of their lora work "good enough" and better than decently, when the RNG seed gods shine, using a few WF's that have been okay since Wan22 launched, if I keep the lora strength in check. Check out their loras on the civitarchive. The cowgirl and missionary and morning wood ones are loads of fun.
@BRai4nowĀ And what did you expect? A close-up in 4K? If you thought that was bad, you can try to do something better by creating your own LoRA, spending your own money and resources š¤
Or you can give me your money and I'll try to improve the quality š
@BRai4nowĀ "Cooked" because you are using lighting LoRA with a very high strength but these LoRAs dont required lighting... Btw I am sure you are using a checkpoint that already has lighting integrated, so, dont use a lighting LoRA if the checkpoint already has it, About the sampling/scheduler use Euler-simple.
I had the same problem as you until I realized that these LoRAs are very well trained to be used just with the very basic settings/standard settings and yup, they do work!
This artist has many Lora characters, and I've tried it myself. The main problem is the face-changing; I can't maintain consistency. Is there a suitable workflow? If I can replicate the artist's effect, I'm willing to pay.
@TUOYIMLU88Ā If you want to maintain overall consistency, simply write down the features. And if you want to keep the same face, try assigning an expression. Wan 2.2 uses an LLM as a text encoder, so it actually interprets text and follows commands. So saying something like "same woman" will usually keep exactly the same woman, but when you get a video that didn't have a reference cut of the same person before and after, it thinks it should cut to another scene with another person, usually more similar to the one in the training video.
You can even test variables, like calling it woman_1 and citing it later. I made a LoRA with 2 women and 1 man, and describing the features of woman_1 and woman_2 actually made it possible to put both in the same scene after a cut.
So if you want to maintain body or facial features, describe them.
Start with 0.8 on High Noise and rise if you need
@TieFighterPilotĀ you know nothing at all what I'm using. I'm commenting on the OP's examples. Odds are his inference image is also cooked. And I'll have you know I use no lightning on high and standard 1.0 strength on low. Using it at all on high degrades quality and high values do stupid things on low. This was established very early on in Wan 2.2's life.
@BRai4nowĀ Thats even the worst part!! Using lora on low will create artifacts, will entirelly damage the end result š¤¦š»āāļø and even 1.0 cfg it is too high on both H/L, I have designed a workflow that allows me to identify these type of problems, I can reverse steps on Low pass to identify "cook" settings.
@TieFighterPilotĀ what the fuck you guys are talking about?! I just redo this LoRA, see my last post on LoRA's gallery to see the consistency of face in the prompt
@AishaAIĀ Just explaining that these LoRAs with around 0.8 works well and no lightingx2v LoRA because it "cook" the results
@TieFighterPilotĀ I train with HLR (High Learning Rate), which causes the LoRA to learn movement patterns very quickly, sometimes even identically. This is called overfitting, and most of the time it's not a good thing. This means that when using 1.0, the LoRA will force a replication of the training, as if you were taking a LoRA with a normal learning rate and setting the scale to 1.5 or 2.0. That's why even with 0.2/0.5/0.8 it continues to work.
Remember that I'm only training the High Noise; it's only responsible for the initial "blur" of body movement and position. The fine details are handled by the Low Noise. I'm creating an overfit in the High Noise, while the Low Noise is normal, and that's why the result ends up being satisfactory, allowing the result to be similar but not identical to the training video. The Low Noise LoRA never got to see my training video, and that's why it prevents it from being identical to the video.
He need to fuck more and use thighmaster less
I agree, and the training video was really intense, but LoRA couldn't learn the intensity. But that's also relative, because for example, if you increase the FPS, it will seem faster. I try to work with the WAN 2.2 standard, which is 16 fps with a total of 81 frames. Videos with a lot of movement end up looking very bad with such a low framerate (like 16). If you keep the frames high to have motion quality, it will look slow, and if you keep the framerate according to the duration (keeping the movement abrupt), LoRA ends up interpreting it as "teleporting" frames, and WAN 2.2 tries to fix this to maintain the smoothness of the movements. If you increase the LoRA scale, you will get more abrupt results, but you will also get some pretty crazy bugs.
I'll redo this one, I'll update it soon
We need this position with legs closed no one has done it yet!
With her legs closed, you wouldn't be able to see anything š¤Ŗ
@AishaAIĀ positioned from behind looking at the penetration you could. It's one of my fav positions
@AishaAIĀ reverse cowgirl is another one with legs raised in the air thighs together with a view of penetration. No one has done these yet.
@Justing1234Ā If you have a video, it's easier to understand
@AishaAIĀ Sent you a DM
Let's do another missi onary pose.
Just send me the video, and I'll do it as soon as I can
Why don't you train the low-noise LORA at the same time? This makes it easier to attempt to reproduce the result.
There's no need to reinvent the wheel; if a low-noise already exists that can create the textures, all we need is a high-noise to define position and movement
@AishaAIĀ Could you share how you organize your training materials? I find that the time and effort spent on editing and preparing materials far outweigh the actual training process. Only training the high noise feels like a waste of both the material and the time invested.
@AishaAIĀ I have a private model with extremely impressive training results. However, the process of collecting and editing the materials took me an uninterrupted 12 hours. The high-noise and low-noise of this model were fixed at specific strenth and mixed together in almost all workflows. But this was extremely exhausting, to the extent that I usually gave up halfway through organizing the materials due to time constraints. If you have a better understanding, please feel free to share it.
@zakakkiĀ I know a few tricks, but I don't intend to reveal them, at least not here. I spend more time searching for videos than actually doing the training. I usually only use 1 or 2 videos, 200 to 400 steps, and the LoRA is ready. Many people complain, but they're just using the Workflow incorrectly, which usually doesn't bother me since I create these LoRAs for Aisha-AI, and if it works there, I don't care if some random person isn't getting good results.