Follow me to make sure you see new poses and Nobodys when I post them. Things move fast on this site, it's easy to miss.
Meet Kristen Nobody, this person doesn't exist. You may think she looks like someone you know, but she isn't. Just the imprecise nature of diffusion. Stylized images were generated using the legendary A to Zovya RPG Artist's Tools model.
If you need a consistent character but don't want a known celebrity, Kristen is your gal. Kristen has an exotic look and prefers to have really long blonde hair with flowers unless you change your prompts.
Follow me for more of the Nobody family coming soon for 1.5 and 2.1.
Click the zNobody tag for more, or the Nobody tag to see other's contributions.
Do you have requests? I've been putting in many more hours lately with this. That's my problem, not yours. But if you'd like to tip me, buy me a beer. Beer encourages me to ignore work and make AI models instead. Tip and make a request. I'll give it a shot if I can. Here at Ko-Fi
Description
FAQ
Comments (28)
Could you please share workflow of how you made coherent same face to train inversion? How many images you used? I need to make a lora or inversion on a character that don't exist, made up by me for my project, but I don't have any images of it yet, only a description in my mind.
You just combine actresses that SD knows at a constant ratio. It will usually produce a consistent face. If you look at the other Nobody models, you will see resemblances to existing actresses,
it's actually both simple and complicated. I play around with the strength of existing tokens in the base model using the Embedding inspector extention for auto1111. get a face I like and save it as an embedding. not the same as training, but delivers the same result. easy since I'm not specifically trying to make anyone's face.
Ah, I see. I need more unique character with special features on which I could train my on Lora with this character. :-)
@drakmour Can also do it the really hard way like I did lol. Look around the existing Embeddings and LORAs, find a couple that are in the ballpark of the general face. For example this one plus the Elsa LORA. Then try mixing em at different strengths like (kristennobodysd15:0.4), <lora:elsaDisneyWrise_v20:0.3>, (Irish woman:0.4), just whatever. Try different things till the outputs are in the ballpark of what you want. Then start generating em! Tweak prompts and setting along the way. Might take several thousand image generations of adjusting to get what ya want. Then keep throwing them away till you have like 15-20 images that are really really close. Then if you need, use Photoshop or free Gimp if you need to add like a mole to the left cheek or something to all 15-20. Then make a new LORA and train it, this video is already outdated but this is the basic idea: https://www.youtube.com/watch?v=70H03cv57-o
Could be really cool to have unique ugly nobodies, and also people who look particularly interesting and weird without being either ugly or pretty.
I have made Loras of (No-one), may be they are somehow distant relatives of Nobody. :D
Great work BTW.
Everyone's related in latent space. Thanks!
Are you planning to also add some male Nobodies? :)
Yes! I have a list. Getting to one soon. I get one done per day.
I really love your Nobody series, but they all seem to skew towards the young adult side... would be really interested to see a Nobody (or a dozen) in their 30s or 40s...
That's a good idea. I'll add it to my list of upcoming nobodies. In the meantime, the embeddings are pretty rigid so that the character is consistent across different styles and models. But you can turn down the strength and modify them a little. Like make (KristenNobodySD15:0.5) and then add "middle-aged woman" instead of "beautiful woman". That would age them up slightly.
Also, almost forgot to mention that these custom models are all biased towards young and beautiful. Try the original SD models for a more realistic, non-glamorized result.
@zovya To be fair, there are a lot of women in their 30s and 40s who are beautiful too.
@haikenedge754 I appreciate that as an older woman. but the Diffusion Model needs prompts based on the stuff people put into it. So you gotta prompt like they trained.
@zovya Fair enough; I just thought maybe, if the textual inversion was trained on a beautiful older woman, it'd automatically lean produce said older woman just by prompting the textual inversion. For example that's not exactly 1:1, when using URPM, if I include "Helen Mirren", it automatically includes her as an older woman in the final product.
@haikenedge754 Yep, it would indeed. It was just a "in the meantime fix". I'll add some wiser women soon.
mix of Alexis and generic Girl? looks pretty.
One of the best embeddings on the whole site
awww, thanks. I like Kristen too
Why is this a .bin file instead of a .pt?
in the early days, in the long long ago, embedding inspector made .bin files for embeddings. now it makes .pt
this is the way
I apolagize to anyone I may have inadvertently offended in my lack of understanding that the .bin files in question are used the same as the .pt files... I'm kinda new at this, and simply had NEVER seen the .bin files in that way.
I really like having A person that can be consistent in Images across multiple Styles.
The sad thing is that you are restricted to the Age of the Character/ Person.
It would be really nice to have a "Face" that is Consistent across Multiple Ages etc.
The Embedding and Loras which I have tried so far weren't able to give that kind of flexibility.
SHE IS UNBEARABLY HOT OMG
thanks
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.