Constricted pupils / wide-eyed - CivArchive (CivitAI Archive)

Constricted pupils / wide-eyed - v0.2a

NSFW

Support my work on Patreon and Ko-Fi and get access to tutorials and exclusive models.
Rate and leave a like if you enjoyed it, and follow for new models!

LoRA trained to generate constricted pupils / wide-eyed characters, to enable better expressions.

Trigger: constricted pupils

Other useful tags: wide-eyed, open mouth, :o, parted lips, etc.

You can also describe the expression: angry, surprised, scared, crying, horrified, screaming, etc.

Description

FAQ

Comments (15)

AnnawnJun 22, 2023

CivitAI

*sips*

mr_JackJun 22, 2023· 4 reactions

CivitAI

woW! This is what I want, T.T

mangaka92Jun 22, 2023· 3 reactions

CivitAI

Love this LoRA. Especially all of the subtle variations of the expression.

Would love to see something similar capturing the many variations on the "expressionless" look with some different options for the more "kuudere" look, the expressionless face with the upside-down triangle mouth, a "bored" look, a "bored and slightly annoyed" look, a "blank/dead eyes" look, etc.

psoft

Author

Jun 22, 2023· 1 reaction

thanks for your feedback.

I'm looking to see how feasible more subtle expressions could be trained and datasets available out there. the tricky part is capturing the small details while keeping the lora from affecting style and composition too much.

mangaka92Jun 22, 2023· 1 reaction

@psoft Thanks for looking into it! Agreed, it would probably be difficult to train on such subtle expressions.

necroryonaJun 22, 2023

CivitAI

amazing job, could we get something like this but for general models?

https://civitai.com/models/12909/hypnosis-or-blank-eyes-or-anzu-or-bonnie-or-hinaichi

x_mkivJun 25, 2023· 8 reactions

CivitAI

any chance you could do heart eyes?

shunwakae167Jul 9, 2023

CivitAI

かわいいよ。こっちをむいて

huun3535Jul 13, 2023

CivitAI

Great honor to ask: how can I adjust to achieve the movement of my eyeballs up, down, left, and right while keeping my head still?

psoft

Author

Jul 13, 2023

That depends a lot on the base model, or if the lora was specifically trained for that, which isn’t the case. Depending on the base model you’re using, if you use looking up/down/to the side/away it may work. But bear in mind that SD has some difficulties with that.

dulltrimAug 7, 2023

This guy made some good LORAs for changing eye directions:

https://huggingface.co/2vXpSwA7/iroiro-lora/tree/main/test

shadow0Jul 22, 2023· 1 reaction

CivitAI

It consistently generates text in the image for me... good otherwise.

psoft

Author

Jul 22, 2023

@shadow0 just with this lora? what base model?

you can use "speech bubble" "text" "english text" "chinese text" in the negative prompt, as it usually helps.

but that also means I may need to check if text is too prevalent in the dataset.