Better poses for taking selfie
This version is intended for SDXL, a version for Pony is here: TakeSelfie Pony.
Usage Hint:
Use strength 1
Keyword:
TkSelf
Build your prompt like this:
TkSelf, a girl in..., taking selfie
Sample:
TkSelf, (slim girl) in a mall. The girl wears ..., taking selfie
Even this LoRA is only trained with SFW images, it can be used with a NSFW checkpoint to produce NSFW content.
Description
Beta, I will refine source pictures an release further version
FAQ
Comments (4)
Been playing around with this. It's easily overridden by other (stronger) LORAS, so it tends to lose the whole selfie aspect quickly. What it seems to create unfortunately are really, really elongated fingers even when it doesn't render a phone.
Therefor, I declared it as beta. I think I have too much photos in training data, where on the training photo the hand holding the phone is out of focus. I'll try to get more images, where also the hand holding the phone is in focus.
@niper53 I'm sure you'll find the sweet spot. I was curious to see what using the same prompt without this lora did, and basically it would render similar images but with a much smaller phone in hand. Which was why I'm curious why it tends to omit the phone so often (I didn't post any of those, but I might later as it made one of my generations look like she was giving the devil horns gesture,). For example, in one of my posts on the red carpet I might get a phone in hand once in 10 images even with the weight at 1. I know it's a work in progress, I'm just hoping my feedback is helpful.
@dabrown3434 The model I usually use produces on the prompt '... girl taking a selfie' an image that looks like the girl would look onto her smartphone. Low position and no stretched arm.. What I wanted to reach is a higher position of the phone, a stretched arm and a perceptive enlargement of the phone. That's the reason, why in the trainings images are (unfortunately) a lot of images, where the phone (and the hand holding the phone) is out of focus.
And thanx for your positive 'I will reach the sweat spot', but to be honest, I am rather a beginner, not a pro. I will try my best, but when the result does not meet my expectations, I often have no glue, what I can do. My try on a LoRA 'SitWithOneLegUp' ended w/o result, and I have no glue, what I can do better. Maybe I end at the memory of my graphics card, I can produce a maximum of 64 network dimensions, otherwise I run into out-of-graphics-memory, or I am just to inexperienced to tag the training-images better.