It looks really good. Did you use a lot of different faces during training ? And silly question but ... why separate High and Low noise in differents posts ?

Post the settings in your examples. Just tried it at strength 1, and saw no effect.

Its sad that there is no saliva

- heavy E-girl-style makeup
- Westernized facial changes
These two issues that appear in other Loras don’t show up here, so the results look natural. However, it occasionally makes the eyes look upward to the right, as if giving a playful side-eye. I’m not yet confident about controlling this through prompts, but I’ll keep experimenting.

I think it's nice that the character face isn't deformed and exaggerated, so thank you!!

Edit: Solved thx to soulblazer :D

Do you have a Workflow for this? i use both high and low on 1.0 strength and even with the Tags that you mentioned the quality of the Ahegao is not nearly as good as in your Animations (she does a Ahegao face but its nearly the same as without a model/with another model, also she doesnt cross her eyes, she just closes them.)

牛逼100/10

Works really well! Will you make more expressions?

Many thanks!

(futanari) https://files.catbox.moe/9ct8cg.mp4

Great job this is the only Ahegao video lora thats worked well for me, works great with the lightning 4step loras as well

This is the BEST ahegao Lora, by far, it can be combined with any lora, on the basic workflow 2.2 from comfyUI lol, no zoom, REAL crossed eyes, even on 10 sec video, good job. 10/10 👏👏👏

I don't know why, but randomization seeds works better than incrementation.

Because after 2 or 3 seeds, the Lora stops working.

Either you find a new random seed and fix it, or you use randomization if you're doing multiple generation, for 100% chance of doing the ahegao.

I used :

"(she crosses her eyes:2)."

"She does the ahegao face. She sticks her tongue out and crosses her eyes, (crossed eyes:2), rolling eyes."

"(Make sure she crosses her eyes during the scene:2)."

(She keeps her eyes wide open with the ahegao expression towards the camera during the entire action and she doesn't blink her eyes:2)

More you insist on the action, more it help, but not too much, exemple of my prompt :

"(A young woman doing the Ahegao face expression:2), while doing "action of others LoRAs"

"A Adult/old, muscular/thick/thin, tall/small, man/women/penis enters from the righ/left/bottom/corner/below

(she crosses her eyes a lot, almost every time:2).

She does the ahegao face. She sticks her tongue out and crosses her eyes, (crossed eyes:2), rolling eyes.

"Describe the action of other LoRAs"

(She keeps her eyes wide open with the ahegao expression towards the camera during the entire action and she doesn't blink her eyes:2)

(Make sure she crosses her eyes during the scene:2)."

in the negativ prompt you can also try :

"blinking eyes, closed eyes"

The lora that I combined :

Bouncing Boobs - Wan 14B (1st, only on high 1.00)

"WAN 2.2 Anime Cumshot Aesthetics – Precision Load I2V (Beta version)" (2nd, high + low strengh depend on the picture and angle so try 1.00 for first try).

OR

"F4C3SPL4SH - I2V Wan 2.2 Video LoRa - K3NK"
(high + low strengh depend on the picture and angle too so try 1.00 for first try).

"Ahegao_IRL" (3rd, sometime Strengh high 0.35  to 0 .60 without too much lora, and 1.00 to 1.20 with multiple lora, 1.45 Strengh low)

The high setting seems to control the size of the eyes, the intensity, and the duration of the lora effect.

The low setting, on the other hand, seems to control the cross-eyed effect and the tongue details.

If the face of the girl is close from the camera, and without too many lora , try :
0.35 to 0.60 high / 1.00 to 1.35 low

If she is if it's a little far away :

0.60 to 1.20 high / 1.35 to 1.50 Low

(But beware of her wide-open eye)

Lower the strength of the other low LoRAs,  if they inject too much emotion, to 0.75 or 0.80, if your prompt is heavy.

Or

Increase the low strength of "irl_ahegao" if you don't see a crossed-eyes effect. Generally, 3 or 4 LoRA max per video, you can try more, but the configuration will be tricky.

Increase the high strength for the duration and intensity of the expression (but not too much if you don't like wide-open eyes and you want to keep the original face).

cfg 1.8 to 2.0 (high) 1.4 to 1.6 (low) , all sampler i used worked, (Sampler : euler > euler, is natural.
scheduler : simple > simple OR beta > beta is good too).

Basic workflow + Loralightx2v :
4 step high (0 / 2) 4 step low (2 / 4) : Faster, decent quality.
8 step high (0 / 4) 8 step low (4 / 10000) : Longer, but, better quality and with better following prompt

One last thing, almost every lora i tried, works with this one, i recommand only one face expression per workflow.
I don't know if the placement of the Loras matters, but I'm still putting this Lora in the last position to prevent the other Loras from crushing it, this still needs to be verified.

Wish she kept the face the entire time. They always drop the expression no matter the length of the video.