Indroduction:
Simple and clear - half-closed eyes concept LoRA, for those, who can't get in on basic model.
Model used for previews: DivineElengaceMix v6
Recommendations:
0.5 - 1.0 weight
half-closed eyesas an activation tagSpecify eyes color
Add
looking at viewertag
Credits:
Trained on NAI Anime Full. Dataset of 150 pictures. 10 Epochs. Clip Skip - 2.
Really BIG thanks to:
Some good folks from Discord server
You, for reading, downloading and reviewing this LoRA
Description
Initial Release
FAQ
Comments (6)
Omg! Amazing! No more derpy eyes or weird winking.
Hi, thanks for commenting. I hope it would work fine for all your needs.
Can also do erotic and derogatory looks.Can be done, but depends on the surrounding environment and conditions.
If you invite him to bed, he will sleep.
Will sleep even when wearing pajamas.
Depending on SEED, he will sleep on side or back.
Examine well.
Never got completely closed eyes with this LoRA. ♂Thats not my problem, its yours♂
Also, i think this -> <lora:add_brightness:0.5><lora:add_detail:0.5> <- LoRAs can ruin other LoRAs or model output in general.
1. There is no bed invitation pose on your images
1.1. None on bed-related tag doing completely closed eyes for me
2. There is no pajamas on your images (you've mistyped this word ->pjama<-, so it can't affect eyes
3. None of prompts you've shown on review ruins my generations
4. I don't have those 2 LoRAs and i'm not sure how it works, some quality/detail/lighting LoRAs just messes up noise on generation process
But if you're having troubles with ->half-closed<- part of eyes, of course you can just add "closed eyes" to negatives. Or just specify eye color or "looking at viewer" in positives. It should help.
Please, check this -> https://imgur.com/a/TyvbD1M
@TroubleDarkness I rechecked it. closed eyes occurs.
The rate of occurrence is 0~60,70% depending on the model.(https://imgur.com/a/Buel8aJ)
(4)<lora:add_brightness:0.5><lora:add_detail:0.5> is to modify the image with LORA in CIVITAI. Specifically, it increases brightness and detail. I do not think that the picture will be disrupted by these. I think there is a possibility of overlap and conflict if they are similar overall corrections, but in this case, I think it's just an eye correction. I checked the prompt matrix to be sure, but it is closed eyes.
(1) What I am saying here is that declaring a prompt triggers it, so it doesn't matter if the action is done on the picture or not (it should matter, but waiting for it to happen would generate a lot of comparison images, which is not realistic. If it is [Prompt matrix], wait until it comes out.)
Thanks for pointing out the omission in (2). I was inadvertent.
Adding [looking at viewer] opens the eye, but it was one eye. This should be avoidable by selecting SEED and Model, but I'm reporting that this is also the case.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.