Alison Brie embedding created by Airathias
Embedding was created in 1000 steps, and a large number of vectors per token. Try to keep your prompt short, but I've seen it work fine with longer prompts.
I added prompts to the example images, should work fine. These prompts were basically blatantly copied from other examples throughout the website. If I copied yours, mea culpa. You can use this without crediting me.
The embedding was trained on stock SD 1.5, so it should work for many if not most models based off of 1.5.
Images listed are made using a handful of models to test, including:
Stock SD 1.5
Deliberate (https://civarchive.com/models/4823/deliberate)
DreamShaper (https://civarchive.com/models/4384/dreamshaper)
UberRealistic Dreamy PFG Porn (https://civarchive.com/models/3751/uberrealistic-dreamy-pfg-porn)
Description
FAQ
Comments (5)
Hey my man, I contacted you on Reddit recently and you helped me out quite a lot. I've been stuck on something though, and thought maybe you have some ideas or Input.
How do you train an embedding for someone that wears glasses sometimes? My brother has glasses on in about half the recent pictures I have, and each half is not enough to train a model (30-40 each)
Combining them in the same model had mixed to bad results, do you have an idea on how to approach this? Thanks for your time and effort my dude :D
While I haven't had to really deal with glasses as a problem, my dataset always contains pictures with some foreign object on the face be it glasses, a mask, a hat or a phone to their face, whatever. Just pictures with something that doesn't belong.
Whenever a picture has such an object, I always make sure I very explicitly mention it in the caption for that picture. So any time someone has glasses or a mask or something, I go "woman wearing a mask ..., with a facemask" just to point out that it's something of note that isn't usually there.
I'm not sure if that's what's been working for me or just my selective process for growing a dataset, but that's my thought process. I'd suggest being very descriptive of foreign objects in your captions, and to not mention them where they're not. So don't be like "a man without glasses". Unless you want your model to really focus on it.
@Airathias That's what I understood too, according to Aitrepreneur's tuto :
If you mention that the glasses are present, it will train the face "wearing glasses", as a foreign object.
Not mentioning them (or mentioning "with no glasses") will act like they are part of what the face should look like all the time.
https://www.youtube.com/watch?v=2ityl_dNRNw for reference.
We try not to sexualize Annie.
The preview images may look a little weird but I think this is the best embedding of her out there, it beats Loras handily with Adetailer on
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.