A request of @rigobert309680.
Keilah Kang is an American model and social media influencer, with more than 2.5M followers on Instagram.
This is a 140-step TI trained on a dataset of 15 images with these settings.
Curious about my work process? I have summarized it here.
Do you have a specific idea for a TI in mind? Visit my website and let me know.
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 beautiful (kk4ng:0.99), a woman with perfect hair, wearing Golden Yellow (cassock:1.1), (cosmic background:1.1), (closeup), modelshoot style, (extremely detailed CG unity 8k wallpaper), professional majestic photography, (Leica M6 Camera), 24mm, exposure blend, hdr, faded, extremely intricate, High (Detail:1.1), Sharp focus, dramatic, soft cinematic light, (looking at viewer), (detailed pupils), 4k textures, elegant, ((((cinematic look)))), soothing tones, insane details, hyperdetailed, low contrast".
Please also note that I'm using the "add detail" LoRA for my example pics. I recommend setting it around 0.5 for best results.
Description
140-step TI trained on a dataset of 15 images with these settings.
FAQ
Comments (4)
Looks really good!
Is there a reason you restrict to just 15 images? I'm getting good results with much larger datasets.
I used to create TIs with larger datasets months ago. It's simply much harder, in most of the cases, to keep a consistent dataset when it's composed of say 30 images instead of 15. And as soon as you start mixing different eras, you may easily get a TI that looks like an "average" of those eras but does not resemble the subject almost at all, in any specific era. Hope that makes sense.
So I started testing with less and less pics, because I had already heard from some creators that worked with smaller datasets, and soon found out that results seemed both better and most consistent. I've trained models with 6 pics, 10, 12... I've settled on 15 because I think it's a good balance between consistency and variety. But I've had good models trained with even less pics.
@JernauGurgeh Thanks! Does using such small datasets result in overfitting?
@hourglassfcupslimwaist I've had some bad experiences with datasets of 6 images, but with datasets of 10 or more images, as long as you select good pics, you shouldn't have any issue.