Natalie Lee, better known as NataLee, Instagram model and influencer.
Update: I wasn't 100% happy with version 2.0 and I've retrained the TI from scratch with my current settings. The current version (3.1) is step 130 of the new embedding. I'll probably add step 135 later at some point as version 3.0, since I think it's also pretty accurate and I had some trouble deciding between the two.
Curious about my work process? I have summarized it here.
Appreciate my work? My TIs are free, but you can always buy me a coffee. :)
Building a good prompt with my TIs
You're obviously free to experiment, but bear in mind that my TIs are trained with a more or less fixed phrasing, that normally starts with:
"photo of EMBEDDING_NAME, a woman"
So I recommend always starting your prompt like that and then building the rest of the prompt from there. For instance, "photo of (natal33:0.99), a woman as a movie star, modelshoot style, (extremely detailed CG unity 8k wallpaper), photo of the most beautiful artwork in the world, professional majestic oil painting by Ed Blinkey, Atey Ghailan, Studio Ghibli, by Jeremy Mann, Greg Manchess, Antonio Moro, trending on ArtStation, trending on CGSociety, Intricate, High Detail, Sharp focus, dramatic, photorealistic painting art by midjourney and greg rutkowski, (white turtleneck top:1.2), ((movie premiere)), (long skirt), ((standing near a movie theater)), ((paparazzi in the background)), (looking at viewer:1.2), (detailed pupils:1.3), ((closeup portrait:1.1))"
Description
I thought it was about time I gave my very first embedding a paint-job. When I created it, I thought it was a poor attempt because I didn't really know how TIs worked and I was building pretty poor prompts. In retrospect, I now think it was at least a very decent TI, and I'm not really sure this new version improves upon it. I'll let you judge.
The new version is an 1800-step TI. I removed 5 pics from the original dataset of 23 to leave it at 18, and applied my usual settings.
FAQ
Comments (5)
wooooow!!! She's gotten better *-*
This is awesome work man. I did go back and look at the first one... and I was actually going to say it looked better at first. But I was wrong.
Sometimes I think when we look at image generations we do, especially when we're invested in it as a project and want it to turn out well.... it's hard to turn our brains off in certain ways and remain objective.
Something that happens sometimes is that if the actual image just straight up as an image is good looking, then we sometimes get lost in that and aren't able to see things as objectively, to see them neutrally and judge with that kind of eye. And we of course by being invested in the project and doing the work want it to turn out well, so that affects our judgement with how "wowwed" we are by the result as well.
When I was looking at the 1.0.... I originally was saying it was better because in my mind I was expecting it to be a lot worse, because I know you're dedicated to your craft and have learned so much. So when I looked at the originals, they were beautiful, the actual style of them is gorgeous, they're great looking images.
But I've never actually seen the person, so I opened up the Instagram you linked and tried to take her in as a person to better judge which of the TIs was better at capturing the essence of who she is. I think your original, it just has some cool elements in the photos that make them beautiful to look at. But I think with the new images, it really does seem like she as a person comes across represented more accurately.
One thing that might be difficult to gauge is how much of that has to do with the actual prompts themselves, and how much of it has to do with the change in technique.
I've been at it for a bit now with doing some personal TI projects and trying to experiment and learn, but I'm going to specifically take on a project next that is for the purpose of uploading here on CivitAI, so I can try and see how it's gone and possibly share some info as well.
Thank you for this TI, it is amazing.
I'm tring to train a TI but I can't get good results.
Can you share your dataset ? I would like to train this TI again to see if I can get same results as you or if there is something wrong with my training.
This together with the NataLee LoRA ought to do magic. Stunningly beautiful woman!
Lovely work, do Debora Gomez
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.